1 GEOSPATIAL ESG THE EMERGING APPLICATION OF GEOSPATIAL DATA FOR GAINING ‘ENVIRONMENTAL’ INSIGHTS ON THE ASSET, CORPORATE AND SOVEREIGN LEVEL WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 3 Authors: David Patterson, Head of Conservation Intelligence - WWF-UK (Lead Author) CONTENTS Susanne Schmitt, Nature and Spatial Finance Lead - WWF-UK Pablo Izquierdo, Senior GIS Lead - WWF-Norway Paolo Tibaldeschi, Senior Advisor, Environment & Development - WWF-Norway Summary 5 Helen Bellfield, Policy Director (Trase Lead) - Global Canopy Dieter Wang, Sustainable Finance Specialist, The World Bank Bryan Gurhy, Senior Financial Sector Specialist, The World Bank Introduction 8 Alexandre d’Aspremont, Chief scientist - Kayrros Paola Tello, Climate Solutions Manager - Kayrros How does Geospatial ESG Align to Existing Environmental or Biodiversity 11 Claire Bonfils--Bierer, Product Manager - Kayrros Screening Approaches? Steven Brumby, CEO/CTO & Co-Founder - Impact Observatory John Barabino, Co-Founder & Head of Business - Impact Observatory The Fundamentals of a Geospatial ESG Approach 14 Nick Volkmer, Vice President, ESG - Enverus Jingwen Zheng, Lead Data Scientist - Enverus Key Limitations Faced when Applying the Open ‘Environmental’ 16 Cath Tayleur, Head of Nature Positive Supply Chains - NatureMetrics Frank A. D’Agnese, President and CTO, Earth Knowledge, Inc. Observational Datasets for Geospatial ESG Application Julia Armstrong D’Agnese, CEO, Earth Knowledge, Inc. 1. Temporal Consistency 18 2. Accuracy 18 Contributors: Alison Midgley, Senior Sustainable Finance Specialist - WWF-UK 3. Spatial Resolution 19 Donald Wubbles - Professor of Atmospheric Science - University of Illinois & Senior Advisor to Earth Knowledge 4. Data Interdependencies 20 Fee Reinhart, Specialist, Sustainable Finance - WWF Switzerland Hugo Bluet, Senior Advisor, - WWF International 5. Relevancy 21 James Lockhart Smith, VP, Markets - Verisk Maplecroft Joanne Lee, Technical Specialist, Finance Practice Core Team - WWF International 6. Biodiversity 21 Peter Eerdmans, Head of Fixed Income and Co-Head of Emerging Market Sovereign & FX - Ninety One Richard Peers, Responsible Risk Ltd - Founder Sarah Rogerson, Corporate Performance Programme Manager - Global Canopy Reflections on the Current Open ‘Environmental’ Data Landscape 23 Bobby Shackelton, Head of Geo, Bloomberg LP Michael Lathuillière, Senior Research Fellow - Stockholm Environment Institute Box 1 – Earth Knowledge 24 Ben Levett, Lead ESG Researcher - Neural Alpha Case Study One Acknowledgement: Asset-Level Assessment – Mines in Brazil – WWF 26 Special thanks to all the various individuals who provided expert guidance and input into this paper. Box 2 – Satellite imagery – visual assessment 28 Copyright: Digging Deeper into the Data 33 Published January 2022 by WWF-UK. Any reproduction in full or in part of this publication must mention the title and credit WWF-UK as the copyright owner. Text © WWF-UK, 2022. All rights reserved. Box 3 – Kayrros – Examples of insights via Commercial 35 Please note: A geospatial view on droughts and employment Brazil (Pages 48-55) © by International Bank for Satellite Remote Sensing Products Reconstruction and Development/The World Bank. License CC BY 3.0 IGO. See https://openknowledge.worldbank.org/pages/terms-of-use-en for the required Box 4 – Enverus – Satellite Insights into the Oil and Gas Sector 41 attribution format and disclaimers. Contact pubrights@worldbank.org with any queries on rights and licenses. Disclaimer: Case Study Two Company-Level Assessment – Soft Commodities in Brazil – Trase 42 This work is a product of the staff of WWF with external contributions. The opinions expressed in this publication are those of the authors. They do not profess to reflect the opinions or views of WWF. The contents employed in this publication and the presentation of material therein do not Case Study Three imply the expression of any opinion whatsoever on the part of WWF or any of the authors concerning the ESG Sovereign-Level Assessment – A Geospatial View on Droughts and 48 performance of any assets, companies, or nations or any delineation. WWF does not guarantee the accuracy of the data included in this work. Employment in Brazil; the World Bank No photographs in this publication may be reproduced without prior authorisation. Key Performance Indicators – What Could be Generated Now? 56 Copy Edit: Chris Cartwright Future Developments 59 Box 5 – Impact Observatory – Observational Data Improvements – Upcycling with AI 60 ABOUT WWF WWF is an independent conservation organization, with over 30 million followers and a global network active in Box 6 – NatureMetrics – Improvements with eDNA 62 nearly 100 countries. Our mission is to stop the degradation of the planet’s natural environment and to build a future in which people live in harmony with nature, by conserving the world’s biological diversity, ensuring that the use of renewable natural resources is sustainable and promoting the reduction of pollution and wasteful consumption. Final Reflections 65 Find out more at wwf.org.uk References 67 Appendix 69 COVER IMAGE © ANDRE DIB / WWF-BRAZIL WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 5 SUMMARY An ongoing challenge with Environmental, Social, and Governance (ESG) efforts is access to robust data. In response, commercial data providers are continually developing solutions to improve insight. Here we discuss one of these potential improvements: the use of geospatial data within ESG focusing on the environmental (E) aspect. Geospatial data can, and is, being used for social (S), and governance (G) purposes, but these are beyond the scope of this paper. This paper explores and tests with real-world examples the potential of geospatial data approaches as means to provide additional insights into the environmental impacts of specific assets, companies, states or nations for sovereign debt investment. Starting with the current data landscape, the document runs through the open ‘environmental’ geospatial data portfolio, outlining its strengths and weaknesses. From this vantage point, the report outlines three case studies in Brazil across differing scales, highlighting various key metrics. The first looks at an asset level example, mining operations; secondly a corporate level example looking at soya production (where asset data is unavailable); and finally a national scale example for sovereign debt insights. Throughout the paper, commercial actors provide technical illustrations as to what more would be possible with additional resources. The document demonstrates that it is possible, even with limited resources and only open data, to generate robust geospatial ESG insights that often can be scaled globally – aiding financial institutions to better differentiate environmental impact at different scales and across different applications. However, as with any method there are limitations. Subsequently, throughout we have tried to illustrate some of the complications which arise with potential outputs, emphasizing the need for actors to carefully consider results in context. The paper concludes by discussing the various future technical developments, highlighting real-world developments, such as eDNA and machine learning, and their implications for the future of geospatial ESG. Finally, we look at a breakdown of the critical components of geospatial ESG tools, showing where they fall on a spectrum, with most underutilizing the technical toolkit available. As a result of this potential technical growth, combined with greater demand from the financial sector, we expect to see a rapid development of more refined geospatial ESG products and insights in the near future. © GREG ARMFIELD / WWF-UK WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 7 KEY POINTS: • It is increasingly evident that tailored sector and site specific geospatial ESG methods and metrics are required to maximize insight. However, such sector specificity creates potential difficulties when attempting to integrate different sector specific metrics at the portfolio level. • Geospatial ESG is emerging into the mainstream, and as yet, there are no universal Adding to these aggregation challenges are the differing needs of users, such as portfolio frameworks or metrics for defining the environmental impact (and dependencies) of various analytics for equity investment versus tailored site-specific investigations for project finance. asset classes. • The majority of open and commercial geospatial ESG platforms do not yet fully utilize all • Robust insights are already possible for some sectors, limited primarily by the extent the data and technical methods available (see diagram below), creating the potential for and availability of asset data (which define location and ownership of assets) and supply rapid development. Yet, the immaturity of the data marketplace for asset, supply chain and chain data. observational data is likely at least in the short term to act as a constraint in some areas. - Asset level to corporate level screening has been achieved for sectors such as oil and gas, mining, fishing, shipping, cement, steel and the power sector. Indeed, commercial factors such as Asset Resolution, Verisk Maplecroft, Reprisk, Bloomberg and others already offer geospatial ESG-derived data products. Some, such as the Trase tool, even manage to generate insights from incomplete asset data, providing estimates of a ASSET DATA OBSERVATION DATA DATA PROCESSING company’s supply chain deforestation risk. Asset location Single layer Direct comparison • Geospatial ESG methods are scalable, across both number of assets and sectors; in this paper we generate insights for mining assets in Brazil which could be applied globally: One vector layer or raster layer Asset overlaid by one or multiple included in analysis. observational data layers. - Defining the high-level impact of mines; considering impact to habitats, conservation areas (considering the intactness and importance variance of each individual conservation area), freshwater exposure, etc. + Sector and site specific weightings + Sector specific attributes + Multiple layers - Ongoing monitoring, land degradation, emissions, tailing dam growth and volumetric expansion of the mines. E.G. power plant, real estate, Two or more vector layers or raster layers Impact adjusted to sector and site variables farm - cotton. included in analysis. - Defining an ‘environmental ratio efficiency’ of mines (or aggregating to the corporate level) against local or global peers, where the weighted extent of habitat destroyed (and + Observational inferences other key environmental costs) is considered against mining production year on year (and any other major positive values). Refining, backfilling observational data from other variables. + Site specific attributes + Dynamic data • The open data portfolio has limitations. However, data is improving year on year, with major intergovernmental initiatives pushing to significantly improve the ‘environmentally relevant’ data portfolio for initiatives such as the UNʼs Sustainable Development Goals (SDGs). E.G. hydro power plant reservoir size, Near real-time feed of data, weather data. + Interdependence power production Mw. - One important tension in the future of geospatial environmental insights is the role of The site specific impacts considering the interdependencies of natural assets, the international governmental agencies (IGOs) and the non-governmental organizations e.g. forest loss impacts on wider local (NGOs) in data provision. Often these agencies are uniquely placed to deliver water security. environmental datasets yet may choose to restrict access to their data for commercial application or may lack the means to be able to generate data products at the required + Additional external data + Sector specific frequency or detail. The private sector will continue to fill data gaps; however it is likely monitoring data + Near real time adjustment to remain with the IGOs and NGO to provide data in some specific areas. Consequently, E.G. web scraped data. a question mark remains over the role, responsibilities, funding and ethics of the IGOs i.e. methane detection, marine oil spill Results updated frequently and capable detection, night time flaring, for oil and gas of adjusting to near real time data feeds, and NGOs in gatekeeping critical environmental data. e.g. oil spill. assets. • Future data developments, such as using machine learning to update multiple observational data layers from one high resolution land cover layer, improvements in ground species monitoring and habitat disturbance detection, are likely to play an important role in + Supply chain asset data + Historic and future data AI providing improved insight. The asset data of all major or significant E.G. past temperature averages, extreme - Significantly our understanding of threats, impacts and the health of ecosystems at scale suppliers and their suppliers. weather events. + Machine rationalization are likely to dramatically improve with new ground sampling methods, such as eDNA, and landscape audio, with complex machine learning models amalgamating these new Analysis is adjusted to the best regional species of ground data insights with other geospatial data, i.e. climate, geographic and data and regional models based on dynamic machine rationalisation of the land cover data. options present. + Other data + Other data Traditional ESG data points, economic, E.G. social, economic, governance data + Machine learning social data points, ground data etc. points, ground data, etc. Throughout any of the various data sourcing, data processing or results, machine learning is applied to iteratively improve outputs. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 9 INTRODUCTION Figure 1 Top, adapted from Stephenson, and Carbone (2021), an example indicator hierarchy, linking ground in-situ biodiversity data from a commercial site to corporate performance. Bottom, adapted from World Bank and WWF (2020) – a hierarchy linking sub-asset For decades, the financial sector has incorporated geospatial data to better understand assessments to corporate performance to the portfolio to national scales.16 opportunities and risks, whether non-material or financially relevant. The insurance industry, for example, has long used complex geospatial models for catastrophe management, such as determining the risk of extreme weather to real estate, and are now arguably at the leading edge of modelling how extreme weather events are likely to change under different climate scenarios. In recent years there has been an uptick in interest around the application of geospatial data to support Environmental (E), Social (S), and Corporate Governance (G) (ESG) insights.1 GOAL: ‘Geospatial ESG’,2 is defined here as the use of geospatial data to generate ESG relevant DATA AGGREGATED GLOBALLY = SPECIES DIVERSITY OF FOREST- insights into a specific commercial asset, company, portfolio or geographic area. Geospatial CORPORATE PERFORMANCE DIVERSITY DEPENDENT BIRD SPECIES ESG begins with the accurate location and definition of ownership of a commercial asset (e.g. INCREASED factory, mine, field, retail estate), known as ‘asset data’. Then using different geospatial data approaches it is possible to assess the asset against ‘observational data’, to provide insights into initial and ongoing environmental impact and other social and governance variables. Geospatial data can be integrated with ground monitoring data (e.g. smart meters, eDNA) and traditional ESG data points for even greater insights. The advantage is clear: an additional DATA AGGREGATED BY COUNTRY = SPECIES SPECIES data source, capable of providing independent, global, high frequency insights into the NATIONAL PERFORMANCE DIVERSITY DIVERSITY environmental impact and risks3 of single assets or companies (by grouping the assets of a company and its supply chain), or within a given area such as a state or country. This rising interest, to determine environmental risk and financial materiality, coincides with improvements in satellite technology and machine learning, spurring the development of an energetic world of related start-ups offering niche or more general data services, such as DATA COLLECTED AROUND SPECIES SPECIES SPECIES SPECIES marine oil spill detection,4 wildfire prediction,5 methane emission detection,6 carbon emission EACH POWER PLANT = DIVERSITY DIVERSITY DIVERSITY DIVERSITY prediction from the heat profile of factories7 or exposure to deforestation within supply chains.8 SITE PERFORMANCE This arrives at a time when pressures on financial institutions (FIs) are increasing on three major fronts: 1) growing calls for voluntary and increasingly mandatory disclosure (e.g., TCFD9) and regulation (e.g. launch of EU Taxonomy and SFDR; UK Green Taxonomy); 2) the need to address ‘double materiality’ (e.g., in TNFD10 scope), which recognizes not only the financial materiality to companies arising from ESG risks and opportunities (dependencies) but also the materiality for society and the environment arising from the companies’ operations (impacts), which in turn can result in financial risks; and 3) the growing importance around the topic of the ‘environment’.11 Recognizing the increasing attention and opportunities, over the last few years, the larger business intelligence providers have increasingly begun integrating various ‘environmental’ TIER 0 - COUNTRY LEVEL geospatial data points into their ESG products. Alongside this mainstreaming, some financial Summed or aggregated scores for countries, based on Tier 3 and 4 data. institutions have begun to expand their technical capacities to make use of geospatial data in- house, often with an initial focus on climate change. TIER 1 - PORTFOLIO LEVEL Unfortunately, the complexity of natural systems and the diversity of commercial operations Summed or aggregated scores for countries, have made it difficult to develop clear metrics to define environmental impact and based on Tier 2 company scores. dependencies. Generating robust insights – across diverse commercial operations each with a differing impact; across vastly different natural habitats with differing sensitivities; and combining complex global supply chains each with differing impacts and dependencies – has TIER 2 - PARENT/COMPANY LEVEL Summed or aggregated scores for parents proven problematic. The situation is further hampered by a lack of data availability at a high companies, based on Tier 3 and 4 results. granularity at a sub-national level, where data simply may not exist for key variables and with no disclosure required in most cases. Combined, these issues have compounded to make measuring environmental impact at the company scale and above extremely challenging. TIER 3 - ASSET LEVEL Assessment of the asset - GIS overlaps, remote sensing, plus Tier 4. Prior attempts to resolve this issue have made important gains, resulting in a large array of various climate, nature and biodiversity standards, methods and tools available today.12 13 Despite these efforts, no approach or standard has yet been widely adopted. Most approaches TIER 4 - SUB-ASSET LEVEL DATA now agree that the solution is to scale results from site operations to higher levels.14 This has Assessment within the asset - IOT, smart been presented in several ways, but essentially it is a hierarchy aggregating environmental meters, traditional ESG reporting etc. insights from the asset to the company to the parent company, or region insights from municipality to sovereign (Figure 1).15 WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 11 HOW DOES GEOSPATIAL ESG Measuring environmental variables on the ground in-situ, whilst effective, is laborious, expensive and unviable at scale. ESG analysts require results ready to go at company and ALIGN TO EXISTING portfolio level, ideally assessed frequently with consistent methods to provide up to date comparable insights. As a result, alternative solutions need to be found. Satellite remote sensing is arguably a good candidate; while it will never be able to answer all questions, it is increasingly able to provide environmental insights at a global scale that are consistent, ENVIRONMENTAL OR BIODIVERSITY independent and repeatable at a high temporal frequency – ideal for creating consistent ESG relevant insights across the globe for millions of commercial assets. Combined with the improvements in machine learning, this leads to the realization that we are entering a world SCREENING APPROACHES? where assets, corporates and nations themselves will no longer be the key factor in disclosing their environmental impact: geospatial ESG approaches, combined with other AI developments such as natural language process (NLP), are increasingly capable of unpicking the large data trails to provide robust insight independent of the actor. Biodiversity and many environmental variables are notoriously hard to quantify: there are no tons, degrees or centimetres of biodiversity. Species, or numbers of species, can be used as Traditionally, large actors like development banks have gained insights into environmental units of measurement, but each unit ‘species’ is not itself a consistent unit – as is, say, a ton impacts through resource and time-intensive social and environmental impact assessments of carbon – each species having differing impact sensitivities, rarity, range connectivity, etc. for specific projects – although increasingly these are integrated with remote sensing and And whilst efforts are under way to develop quantifiable and comparable biodiversity metrics, geospatial components, modelling or a combination of methods. A ground-proofed approach, such as Mean Species Abundance (MSA) or Species Threat Abatement and Recovery (STAR),17 whilst excellent, is not suitable for many types of investment, where one company may have they have limitations (See Key Limitations – Biodiversity). Additionally, environmental risks hundreds of assets and supply chains connected to tens of thousands of sites. In the absence and impacts are often non-linear, can occur over long time horizons, and materialize abruptly of having ground assessments to hand, actors are forced to turn to modelled or geospatial when they do occur, due to threshold effects or tipping points. Satellite remote sensing and approaches for insights. geospatially derived metrics are not exempt from these challenges. Yet progress has been made over the past decades, and in the coming years, more advanced technology (See Future A large range of modelling, generalist environmental and specific biodiversity footprint Developments), sensors and models will be capable of providing greater insight into near-real tools19 have emerged to aid corporates or financial institutions (FIs) in understanding the time trends of ecosystem health and other relevant insights.18 Combined with improvements in environmental implications of various types of operations or investments. Commonly these machine learning, it is now inevitable that geospatial insights will improve and offer increasingly tools combine publicly disclosed corporate information with open-source scientific datasets valuable data points to be integrated within traditional ESG methods. and then apply a custom model to define risks or impacts. Frequently they attempt to capture upstream and downstream effects, using some form of value chain analysis linked In this paper we explore the topic of geospatial ESG, looking at the challenges and what can to the location of the company’s production facilities and generalized biodiversity impacts be achieved today both with the current open ‘environmental’ geospatial data portfolio and by or pressures. Portfolio-level outputs, commonly sum company-level assessments. All these commercial actors. This will be illustrated with three real-world case studies in Brazil, showing tools are relatively new, and as such, there is very little standardisation or benchmarking to geospatial environmental insights, with a focus on defining impact across three different test results between tools. scales: project finance, corporate investment, and sovereign debt – specifically: Often these footprint tools, lacking direct site-level environmental impact measures, convert • Mining Projects in Brazil for Project Finance – WWF publicly disclosed revenue figures into production volumes as a starting point to scale impact. To achieve this, they use some means to classify the various activities of each company • Soft Commodity Companies in Brazil for Corporate Investment – Global Canopy (e.g., GICS, NACE, FactSet’s Hierarchy). These are then combined with other open-source or custom methods (e.g., EXIOBASE, ReCiPe/Life-Cycle Assessment) to translate production • Environmental Performance of Brazil for Sovereign Debt – World Bank and resource usage into a range of environmental pressure metrics, such as land-use change, CO2 and CH4 emissions, and freshwater pollution. These are then converted again into   biodiversity impact metrics, often via an open-source model such as the Global Biodiversity Model for Policy Support (GLOBIO). BIODIVERSITY AND MANY ENVIRONMENTAL VARIABLES ARE NOTORIOUSLY HARD TO QUANTIFY: THERE ARE NO TONS, DEGREES OR CENTIMETRES OF BIODIVERSITY WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 13 DIRECT MEASUREMENT - INDIRECT MEASUREMENT - These approaches use standard values for location, industry and activity, and as such GEOSPACIAL ESG MODELLING / FOOTPRINTING require tailoring for each company to provide robust results. The tools frequently include geospatially derived variables within their models. Examples of these footprint focused tools include: Currently infeasible due to lack of Aggregated from parent PORTFOLIO SCORE assets data for all sectors company scores • Corporate Biodiversity Footprint – Iceberg Data Lab • Biodiversity Impact Analytics – CDC Biodiversité / Carbon4 Finance Aggregated from company scores (only for sectors with well defined PARENT COMPANY SCORE Aggregated from company scores • Biodiversity Footprint for Financial Institutions – ASN Bank / Pre / CREM asset datasets) • Sustainable Investment Framework Navigator – KPMG / CISL Aggregated from asset scores COMPANY SCORE Aggregated from asset scores • Portfolio Impact Analysis Tool for Banks – UNEP FI Positive Impact Initiative Defined within the asset dataset, Commonly defined within a global On the other side of the equation, a range of geospatial ‘environmental insight’ focused large variation exists on the dataset, such as GICS, NACE, extent and accuracy of the COMMERCIAL ACTIVITIES FactSet’s Hierarchy. Often not tools have emerged. These tools tend to be designed around screening for project finance, attributes included capturing well the full range of often without pre-packaged asset data, requiring the user to upload and compile their activities undertaken. own assessments. Examples include Global Forest Watch Pro, Ecometrica, Maphubs, and Integrated Biodiversity Assessment Tool (IBAT). Others, such as Asset Resolution, Verisk Maplecroft and Reprisk, contain asset data and can in some cases provide insight at asset, corporate and sector levels. Commonly does not apply a highly Defined using asset location ASSET SCORE location specific assessment, rather against observational datasets The use of geospatially derived data within modelled approaches is common. Broadly regional specific values speaking, however, it is possible to categorize the two major approaches as those which focus on 1) indirect measures via modelling of financial data and sectoral classification and those which focus on 2) direct geospatially derived measurements. Neither approach is necessarily better than another; both have limitations. However, the goal is the same: to Commonly defined to ASSET LOCATION define at multiple scales as comprehensively and accurately as possible the environmental within 100m impact (and often dependencies) of a wide selection of companies. Figure 2 simplifies the steps these two approaches often take to define environmental impact (and dependencies) at the asset scale or higher. Geospatial approaches can aid Publicly disclosed revenue figures are converted into production in defining the environmental impacts of commercial activity – e.g. the extent of power volumes. Then translates consumption via infra-red heat proxy, the extent of land degradation, marine oil spills, and Commonly not predicted, rather ACTIVITY REQUIREMENTS production and resource usage into CH4 emissions. Importantly, compared to modelling approaches that often rely on annually impacts are directly measured (see (ENVIRONMENTAL PRESSURES) a range of environmental pressure below) metrics, such as land use change, disclosed data, geospatial metrics can potentially be generated at a very high granularity CO2 and CH4 emissions, and and temporal frequency, assuming robust asset data is continuously available. This allows freshwater piollution. potentially highly accurate independent daily, weekly and monthly metrics. Of course, improved environmentally relevant geospatial data is potentially a win for any approach as it Historic and ongoing near real provides more robust data which could be integrated into any method. time satellite derived, quantitative These are then converted again into measurements for specific biodiversity impact metric, often variables such as land use change, ENVIRONMENTAL IMPACTS via an open-sources model such as The next sections look at these geospatial approaches in more detail. infra-red heat profile as a proxy for Global Biodiversity Model for Policy CO2 and CH4 emissions, and forest Support (GLOBIO) loss, fragmentation etc Frequently attempts to capture upstream and downstream effects, Repeats the asset assessment as using some form of value chain SUPPLIERS SCORE analysis linked to the location of the above for suppliers assets company’s production facilities and Figure 2 (following page) biodiversity impacts or pressures. Diagram illustrating the two approaches for corporate biodiversity environmental screening, the more established economic driven modelling approach (RIGHT), and the emergence and potential of direct measurements via geospatially driven methods (LEFT). For simplicity we have separated the two approaches; however it seems inevitable that due to data gaps in the geospatial portfolio, hybrid approaches pulling from both sides of the equation are likely to be developed. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 15 © ANDRE DIB / WWF-BRAZIL THE FUNDAMENTALS OF A ‘ENVIRONMENTAL’ OBSERVATIONAL DATA GEOSPATIAL ESG APPROACH The world of geospatial data that might be relevant as etc.). Where data is not available at the asset level, it might observational data is vast. Platforms like Resource Watch be necessary to average results regionally or apply regional and UN Biodiversity Labs, with large public data portfolios, datasets or some other measurement (see Case Study Two). serve to illustrate the diversity and depth of some of the This approach requires a wide-ranging and up-to-date set of major environmentally relevant datasets available. The initial environmentally relevant observational datasets for the analyst challenge is that a broad range of topics – from weather to to draw upon. soil carbon to biodiversity to an area’s legal status to the net A geospatial ESG approach is straightforward. The location/s of an asset or a company’s primary productivity of ecosystems and climate change22 – With the geospatial ESG approach for environmental asset and their suppliers’ assets are geolocated. Known as ‘asset data’, once defined these might be relevant depending on the use case. As a result, impact developing, there remains confusion around which locations or areas can be compared or modelled with ‘observational data’ – datasets thousands of datasets, either local or global, may offer value. observational datasets should be applied. Which are the most that provide insight. Within the environmental space, these might provide insights into essential? And what exactly can this data tell us separately variables such as a factory’s heat profile as a proxy for power usage, methane emissions, At a most basic level, a user can compare assets against and in combination? What data is missing and how might it or direct impacts to the natural world such as by considering overlays with protected areas, a single observational data layer in a direct one-to-one be improved in the future? How should methods be tailored to deforestation, habitat fragmentation, endangered species, habitat connectivity, biodiversity, comparison. For example, global power plants (asset data) specific sector needs? To begin to answer these questions, etc.20 Throughout this document we assume access to robust asset data, with the can be compared against World Heritage Sites (observational it is first necessary to understand the data realities and the necessary ownership information to allow results to be linked to specific assets, data) to identify which power plants are within or near a issues currently shaping geospatially driven methodologies. companies and then portfolios. World Heritage Site. Whilst useful, often parties are interested in more complex questions, such as, ‘Which power plants In the next section (Key Limitations) we consider the constraints More complex geospatially driven approaches are possible that consider environmental are having the greatest environmental impact on World of the current data situation by looking at a portfolio of open dependencies (e.g. water risks) and wider risk modelling, assessing environmental values in Heritage Sites per MWh?’ To begin to answer this question23 ‘environmental’ geospatial data. Here, as an example drawing connection with another and not in isolation. This would consider, for example, the interrelated will require the application of many observational datasets upon the geospatial data portfolio of the UN Biodiversity Lab, a environmental impacts of a company’s assets, all its operations globally and its ongoing together, in combination with other non-geospatial data points UN website that is focused on curating and managing a robust impacts (and how these impacts affect other operations within the context of the immediate – looking at, say, the power plant’s attributes: its type, fuel environmental geospatial open library to support nations’ and global landscape); the near-real-time direct weather risks to the assets and indirect type, output, etc. Using these sector-specific variables, it is delivery of the SDG and CBD goals. As such it serves as a risks (e.g. extreme weather damaging supporting infrastructure); and the long-term climate then necessary to consider the asset’s initial environmental robust example of the current authoritative and relevant data implications – all relative to the asset’s positive outputs (e.g. production and performance vs. impact (e.g. clearance of site for construction and scaling available and useful for considering environmental variables its peers). As an introductory starting point within this paper, we simplify the discussion of that impact against habitat types; endangered species using globally consistent datasets. It by no means captures all to focus on geospatial approaches to measuring direct environmental impacts.21 presence; proximity to highly sensitive conservation sites) and the data available but provides a good indicative sample of the ongoing impacts (e.g. CH4 emissions, infra-red heat profiles, open global scale datasets available.24 WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 17 KEY LIMITATIONS FACED WHEN 2000 2006 2009 2002 2003 2004 2005 2008 2001 2007 2020 1990 1996 1993 1999 1980 1983 1984 1985 1986 1988 1989 1992 1994 1995 1998 2021 1982 1987 1991 1997 1981 2010 2012 2013 2016 2019 2015 2018 2014 2017 2011 Dataset Name Aboveground Biomass Density of Vegetation APPLYING THE OPEN ‘ENVIRONMENTAL’ Aboveground live woody carbon density change 2003-2014 Accessibility To Cities Annual Water Recurrence (Global Surface Water Explorer) Areas of Global Significance for Biodiversity Conservation, Carbon Storage, and Water Provision Bare Ground Change 1982 - 2016 Biodiversity Intactness Index (2016) OBSERVATIONAL DATASETS FOR Change in Cumulative Human Impact to Marine Ecosystems City Water Map Catchment TNC Contiguous Zone (24 nautical mile limit) Continuous Land-Sea Adminstrative Boundary Crop Suitability 2011-2040 Crop Suitability 2071-2100 GEOSPATIAL ESG APPLICATION Crop Suitability Change 1981-2040 Crop Suitability Change 1981-2100 Crop Suitability Change 2011-2100 Cumulative Ocean Impact - Sum of Pressure Data 2013 - KNB Digital elevation model (DEM) 90m res. Ecological Land Units (ELUs) - USGS-ESRI Ecoregion Degradation (National Level) Ecoregion Loss 1993 - 2009 (National Level) Ecoregions 2017 - By Biome Ecoregions 2017 - By Ecoregion Exclusive Economic Zone (200NM) (2018) Fire Locations (2013 - 2018) Forest Connectivity Geospatial datasets, either alone or in combination with other datasets, can be used to Forest Fragmentation 1000m Forest Fragmentation 500m provide ESG insights on ‘environmental’ variables and even biodiversity impacts and risks. It Forest Height (Africa) Forest Height (SE Asia) is not yet clear what the data requirements are for geospatial ESG, where standards are still Forest Height (South America) to arise. Different applications will have differing data needs – for example, sovereign debt in- Forest Structual Condition Index (FSCI)(2019) Forest Structural Integrity Index (FSII)(2019) sights will differ from project finance. However, several assumptions can be made. Firstly and GEOCARBON Global Forest Aboveground Biomass Global Distribution of Saltmarshes (2017) most importantly, the data must be capable of meaningfully capturing the environmentally Global Distribution of Seagrasses (2020) Global Distribution of Warm-Water Coral Reefs (2018) relevant variable. Secondly, they must provide insights at a meaningful resolution: results at Global Distribution of Wetlands - CIFOR (v2) Global Forest Change 2000–2019 40 km may suit landscape, state or national insights but may lack the granularity to report Global Forest Cover 2000 on the impacts of specific commercial assets. Thirdly, they must provide insights at a high Global Forest Cover: Gain 2000-2012 Global Forest Cover: Loss Year (Highlight) 2019 enough frequency to be meaningful to the analyst but also manage to capture events – for Global Grid of Probabilities of Urban Expansion to 2030 Global Habitats example sampling once a year is not suitable for capturing CH4 emissions. Finally, datasets Global Human Modification Index Global Intertidal Change (1984-2016) need to be consistent, to enable trends to be calculated, where for most sovereign applica- Global Land Cover (ESA/ESA CCI/UCLouvain) 2015 Global Mangrove Forest Cover for the 21st Century (2000) tions five years of data seems to be required. Global Soil Organic Carbon Map (v 1.5) Gridded Livestock of the World – Cattle Gridded Livestock of the World – Goats Considering these variables against the current open data portfolio raises some interesting Gridded Livestock of the World – Sheep Human Footprint 1993 (Terrestrial) issues which potentially act as a limitation on insights. Here we review six key issues Human Footprint 2009 (Terrestrial) Human Footprint Difference 1993-2009 (Terrestrial) commonly found within the open data portfolio: Human Impact on Forests Human Pressures Human Pressures in Protected Areas (2018) Increase in SOC on Croplands After 20 Yr - High Scenario • Temporal consistency Increase in SOC on Croplands After 20 Yr - Medium Scenario Intact Forest Landscape  IUCN Green List of Protected and Conserved Areas (Jan 2020) • Spatial resolution Last of the Wilds 1993 + 2009 Live Biomass Carbon Density Mangrove forest Soil Organic Carbon (2018) • Accuracy Marine Ecoregions of the World Marine Ecoregions of the World Protected Coverage Marine Proportional Range Rarity (Aquamaps) Marine Protected Areas (WDPA) • Data interdependencies Marine Range Rarity (Aquamaps) Marine Species Richness (Aquamaps) Marine Wilderness • Relevancy Maximum Water Extent (Global Surface Water Explorer) Nutrient pollution (fertilizers) 2013 - KNB Ocean Pollution (Shipping Lanes, Ports) Pressures 2013 (KNB) Pelagic provinces of the world • Challenges of ‘Biodiversity’ Physical Exposure to Tsunamis Population Density (CIESIN, 2018) Potential for Tropical Forest Carbon Sequestration - Catchment Potential for Tropical Forest Carbon Sequestration - Country Protected Area Connectivity (ProtConn) - Country Level Protected Area Connectivity (ProtConn) - Ecoregion Level Ramsar Sites (WDPA) Ramsar Wetland Sites Rarity-Weighted Richness Realised Clean Water Provision SDG 6.6.1 - Global Water Transitions 2000-2018 Short Vegetation Change 1982-2016 Species Richness Terrestrial Ecoregion Protection - 2018 Terrestrial Ecoregion Protection - 2018 Terrestrial Protected Areas (WDPA) Terrestrial Wilderness Terrestrial Wilderness Lost Territorial Sea (12 nautical mile limit) The City Water Map (CWM) The World Database of Key Biodiversity Areas (KBA) Threatened Species Richness Tree Canopy Change 1982 - 2016 UNDP GEF Mapping - UNDP GEF Funded PA Projects UNESCO Biosphere Reserves Figure 3 (following page) UNESCO World Heritage Sites (WDPA) Water Occurrence (Global Surface Water Explorer) A visual illustration of temporal coverage of 105 open and commonly used environmentally Water Seasonality 2014-2018 (Global Surface Water Explorer) relevant geospatial datasets. World Atlas of Mangroves WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 19 1. TEMPORAL CONSISTENCY Vector Datasets Issue: Open25 environmentally relevant datasets have poor temporal consistency. Cause – Within vector datasets – shapes defining areas such as protected areas – accuracy is not absolute, and the boundaries of given areas are not always correct. Often situations arise where there To illustrate this point, below we reviewed 105 data layers26 listed on the UN Biodiversity Lab, highlighting is no ‘agreed’ boundary, with different stakeholders presenting different delineations. This is a common in blue every year a data layer has a measurement. Only 40 data layers (38%) had values for more than issue with border disputes between countries. Beyond basic errors, large environmentally relevant vector one year. Only 20 (19%) had consistent records for over five years. datasets often contain technical faults, such as topology or geometry errors, making the datasets difficult and in some cases impossible to analyse without a significant correction. The lack of data points over time is compounded within some datasets, such as the Global Human Modification Index, where, due to different methodologies applied, different years are not directly Implications – Error or disagreement on boundaries for an area (e.g. a protected area, country, comparable. Finally, research and development and then publication delay have an impact, where indigenous area) as defined within vector datasets means that results reported may not be accurate or frequently several years may have passed between developing products and publishing results. For may be perceived as inaccurate by some stakeholders. Technical errors slow assessments and waste example, the Biodiversity Intactness Index was published in 2016, reporting results for 2005. resources and reduce the application of the dataset. Cause – Global datasets are expensive to develop and maintain. Frequently, data layers are developed Any systematic or random error is undesirable in terms of generating useful insights and undermines for academic publication. Once the publication is achieved there may not be resources or incentive to the potential strength of geospatial insights, although most boundary errors can be addressed by regularly update and produce year on year updates. Or technology or methods may improve, outdating reporting results conservatively with buffer areas included or with error margins. Perceived error is more the approach and data product. Those data layers which are produced consistently at high frequency are challenging to address and has slightly unusual ramifications, for example, government officials may not almost always those backed by major programmes, such as Global Forest Watch, or major databases readily accept results generated for sovereign debt geospatial assessments of their nation if they disagree such as the World Database on Protected Areas. with the boundaries applied in the assessment. Added to this issue, an asset’s area may change over time. This issue, or its potential, needs to be addressed when designing metrics. For example, at a simple Implications – Geospatial ESG insights are only of value if they are correct. In general terms, the older level, total forest loss within a palm oil producer farm is not directly comparable over time if the producer’s the observational dataset applied, the greater the potential that the data will be out of date and incorrect farm changes in area, but total forest loss, per km2, would be comparable. as a current measurement. The lack of consistency over time also limits the ability to consider trends over time, where ideally at least five years of consistent data is required. Low temporal frequency, with only a 3. SPATIAL RESOLUTION few datasets offering monthly updates, also makes it impossible to monitor emerging issues in near-real- time or track trends at a finer scale, or in some cases define the initial impact of projects. Issue: Open environmentally relevant raster datasets often have a low spatial resolution. The value of an observational dataset rests in part on how frequently it is updated, but also on its spatial 2. ACCURACY resolution. This refers to how large the pixel size is (Figure 4). Those with a fine resolution (under 30m) allow more detailed insights; above 30m, necessary detail, such as deforestation or land degradation, begins Issue: The accuracy of environmentally relevant spatial datasets is not absolute. to be averaged. Although as we will see this is not relevant for all applications – the resolution required depends on the task in question. Within the data portfolio, assessed resolution ranged from approximately To help understand how this data can be applied, it is necessary to understand a little about the data. 30m at the equator to 100km. Of the 105 layers assessed, 24 (22%) had a resolution of or below 100m. Firstly, there are two major types: vector, and raster data: • Shapes (vector files) often define man-made delineations, country boundaries, protected areas, indigenous areas, key biodiversity areas, marine protected areas, important marine mammal areas, estimated species ranges, etc. • Grids of pixels (raster files), often used to represent continuous phenomena or variables, are equal- area squares with a given specific value, frequently used to provide global maps of land cover, elevation, forest loss, forest gain, flood risk, ground carbon, extreme weather risk, human disturbance, biodiversity indices, species counts, habitat connectivity, etc. Raster Datasets Cause – Raster layers with environmental relevance are often based on complex image classification algorithms of satellite imagery, in which methodological choices have had to be made to define how to interpret images. Ground validation, required to improve the accuracy of data products, is often costly and as a result limited. In addition to the methodological challenges, some classifications provided might not be narrow enough for the sought application – e.g. ‘forest’, and not ‘pine forest’. Modelled layers are frequently developed from data sources which contain data gaps, gaps that are often not expressed in the results, potentially providing a false impression of their accuracy.27 Implications – Fortunately, most major open datasets are the results of peer-reviewed research, and Figure 4 A graphical representation of spatial resolution, illustrating how rapidly detail is aggregated at relatively high levels of a high standard exists. As such an assumption of a fair degree of accuracy is often possible. This is resolution. Taken from Tian et al., 2020 images show in false colour (Red, NIR, blue bands as R-G-B images of a subregion in combined with the consideration that a fair degree of imprecision is likely to be tolerated in geospatial Beijing at different spatial resolutions. 28 ESG, where the goal is often high-level screening to find outliers rather than to delineate between specific values – although as attention on the subject increases, and as data products are increasingly used as decision variables within the financial sector, we can expect to see greater scrutiny placed on the accuracy of products used. However, caution should always be applied, and accuracy should not be assumed. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 21 Cause – Almost all openly available global remote sensed derived data products have been developed based on freely available ESA and NASA satellite imagery. This imagery resolution has commonly been 5. RELEVANCY a limit on the resolution of global environmentally relevant datasets. Whilst high-resolution imagery is Issue: The current open environmentally relevant datasets lack topic coverage. commercially available up to 30cm, the cost of acquiring the imagery and then the computational power required to convert these images into useful insights means this is unviable for almost all academic and A robust overview of environmental impacts (and other potential use cases) for ESG application requires NGO applications, where even developing 30m resolution layers with freely available imagery requires observational datasets which provide relevant insights into many different subject areas, such as above-ground significant computing resources. As a result, many of the finer resolution data products rely on donated biomass, net primary productivity, vegetation height, fragmentation, soil moisture and species abundance, at resources from the tech sector, such as Google, Microsoft, or others. Modelled layers which use ground high spatial and temporal frequencies. data face a similar compute restriction, often limiting resolution to around 1km. Whilst a wide portfolio of geospatial data is available, it does not always explicitly capture the specific metrics Implications – Often commercial assets are relatively small (e.g., individual power plants, roads, mines, required. Confusing the situation is the fact that there are often tens of geospatial datasets available for a single farms, etc.). Frequently users are interested in topics such as deforestation, land-use change or pollution topic, all with slightly differing methodologies. The analyst faces the challenge of identifying which one to apply. events, at a fine scale to link to specific assets, which typically requires high-resolution observational This is worsened by the poor temporal consistency (as outlined above), where datasets are often not updated, datasets. However, there are many exceptions where high-resolution data is not required, such as some forcing actors to switch between datasets, undermining the consistency of results and their ability to track national scale indicators. It could be argued that many of the critical environmental data products relevant metrics over time. to geospatial ESG insights are limited by spatial resolution. However, this is unlikely to be the case for long since, as satellite technology develops, we can expect to see more and more high-resolution imagery and Cause – Beyond the practical costs and challenges in updating global data products, there is also the technical derived data products become obtainable at a viable price point. difficulty in quantifying variables for some topics, where some remote sensing measurements are simply easier to achieve than others. As such, there is often a bias towards the more technically feasible topics. 4. DATA INTERDEPENDENCIES Implications – As demand increases from financial investors to be able to better define their ‘biodiversity’ impact or risk exposure and various ‘E’ metrics, there is an emerging risk that until the various stakeholders Issue: Environmentally relevant observational datasets may draw from the same source data. work to overcome complex technical challenges, the current data gaps will be filled with subpar data products. Fortunately, there have long been calls for an improved environmental data portfolio for wider Cause – Due to the challenges and efforts in developing global geospatial data products, researchers conservation goals,30 and there are established efforts underway working to resolve this issue.31 Other technical often use one or multiple existing spatial datasets to build a new product, for example, ‘Terrestrial developments are also occurring, which are likely to combine and complement the remote sensing efforts, see Ecoregion Protection – 2018’29 is a combination of: Future Developments.   • IUCN and UNEP-WCMC (2018). The World Database on Protected Areas (WDPA), April 2018. 6. BIODIVERSITY Cambridge, UK: UNEP-WCMC. Available at: www.protectedplanet.net Issue: ‘Biodiversity’32 is extremely difficult to capture and define in near-real-time33 at a global scale. • Olsen, D.M. et al. (2001). Terrestrial ecoregions of the world: a new map of life on Earth. BioScience, 51(11): 933-938. It is often claimed by data providers that their geospatial data, model or tool is robust and provides holistic insight into biodiversity impacts or related areas. However, we would argue that at this time, no team, group or • The Nature Conservancy (2012). Marine Ecoregions and Pelagic Provinces of the World. GIS layers product has yet achieved a means of defining the impact of commercial operations on biodiversity, at a global developed by The Nature Conservancy with multiple partners, itself combined from scale at a high temporal frequency. - Spalding et al. (2007) Marine ecoregions of the world: A bio-regionalization of coastal and shelf If, for example, we consider the Integrated Biodiversity Assessment Tool (IBAT), which is often presented as a areas. Bioscience 57: 573-583.and Spalding et al. (2012) Pelagic provinces ecoregions of the world: A method of screening for biodiversity risk, we can explore some of the challenges. IBAT is a highly useful data biogeographic classification of the world’s surface pelagic waters. offering primarily made up of three global datasets: Protected Areas (WDPA), Key Biodiversity Areas (KBAs) and - Ocean & Coastal Management 60: 19-30. GIS DATA (the non-cut on the coastline version has been the IUCN Red List of Species. Protected Areas (PA) and KBAs may or may not have a high level of biodiversity used) downloaded on 20160720 from http://data.unep-wcmc.org/datasets/38. within them; some may be highly degraded, some may be pristine. From the source dataset itself, the differentiation of intactness of sites is not possible. Since there is no near-real-time input on the physical status of the intactness of assets, results in some cases may not be well-grounded. For example, although the WDPA Implications – This means that errors or issues in prior datasets can be compounded. In addition, the is updated monthly, if a PA has recently been converted to, say, a palm oil plantation, this would not necessarily ‘new’ product may be formed of much older datasets and may not be as up-to-date as first considered. be reflected rapidly within the IBAT data, requiring further triangulation with external datasets outside of IBAT Finally, as actors move towards more complex geospatial assessments and merge multiple datasets by the user and likely additional licensing rights. These datasets and the others available of course provide in combination, there is the potential for duplication of values within the model, with the same dataset valuable insights, but even the most robust data offerings have challenges. This means that in almost all cases, effectively influencing results multiple times. for geospatial ESG application caution and additional analysis is required when applying them.   Cause – In the race to compete and provide products for the growing demand, actors may inadvertently overstate the relevancy or accuracy of their products. Or conversely, various FIs keen to rapidly upskill in this space may not have the time or resources, or the incentive, to scrutinize the solutions they are offered. Implications – Assuming that a dataset meaningfully captures an environmental metric more than it actually does creates the potential for actors to falsely believe their exposure is less than it is, or that specific companies or assets have higher exposure. This ultimately has the potential to aid greenwashing, derailing trust in the ESG process of trying to realign capital to support nature recovery, or to slow the effectiveness of the realignment. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 23 REFLECTIONS ON THE CURRENT OPEN ‘ENVIRONMENTAL’ DATA LANDSCAPE The open34 data portfolio assessed in this document is by no means comprehensive. It does, however, highlight the general themes and the common issues currently faced by those providing environmental geospatial ESG solutions drawing upon these reserves. Actors within the commercial space, as we will explore, have developed workarounds to some of these challenges. Yet often the underlying data used by commercial operators comes from publicly available data sources provided by NGOs, IGOs, academia or multilaterals. Indeed, due to the irreplaceability of these global environmental datasets, where only one or two exist, future commercial geospatial ESG developments will most likely in many cases be restricted by these datasets unless radical solutions are found. From a geospatial ESG perspective, perhaps the most pressing issue is the that surrounding the temporal and spatial resolution of datasets. The graph below (Figure 5) illustrates this: out of the 70 raster datasets assessed, most (46) have a low spatial resolution and a poor temporal resolution; only a few (6) can be considered to have both high spatial and temporal resolution. If geospatial ESG is to deliver meaningful results, it will require a more temporally consistent and wider environmental data portfolio to work from, particularly around ‘biodiversity’. Perhaps the most established push for a unified plan for monitoring global biodiversity comes from the Group on Earth Observations Biodiversity Observation Network (GEO-BON), via its common framework of essential biodiversity variables (EBVs). A recent paper by Skidmore et al (2021) proposed a set of 30 key remote sensing biodiversity products35 for global biodiversity monitoring to fill data gaps for wider conservation purposes, such as tracking performance to global targets, United Nations Sustainable Development Goals (SDGs) and Aichi targets. The paper,36 and others before it, have repeatedly highlighted the need for harmonized, open, accurate, repeatable and reproducible, analysis-ready remote sensing biodiversity products (and with that the need for more ground-truthed biodiversity data) for national monitoring,37 policymakers and scientists – a need we echo here, but for a newer use case: the geospatial ‘Open’ ESG application for the financial Data Temporal Resolution sector. No. of years with data -5 0 5 10 15 20 25 30 35 40 45 1 10 100 Resoltuion (m) 1000 10000 100000 Figure 5 Graph showing the spatial and temporal resolution of 70 raster layers assessed from the UN © NATUREPL.COM / NICK HAWKINS / WWF Biodiversity Lab data portfolio. Circle size indicates number of datasets. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 25 LINKING IN-SITU OBSERVATIONS WITH REMOTE SENSING EFFORTS Earth Knowledge’s framework aligns to the same five drivers of global change developed by the IPBES in their global WILL BE ESSENTIAL TO CREATING assessment of biodiversity and ecosystem services,40 41 the WWF in their Living Planet Database and Report,42 the WEF in their Nature Risk Rising Report,43 and which were originally defined by the IUCN in their Standard Lexicon of IMPROVED DATA PRODUCTS. Biodiversity Conservation.44 This fundamental alignment of the Earth Knowledge’s Indicators Framework, or indeed any platform, to these authoritative bodies is foundational. It provides consistency in process and language so that more direct translation can be made between science conclusions and global change and sustainability outcomes evaluated by financial institutions. Vitally the data generated from Earth Knowledge Indicators are structured to identify and forecast both discrete Within the commercial space, actors such as Ecometrica , Earth Knowledge and others 38 39 environmental processes and the interrelated resulting conditions of global change on biodiversity and other aspects of have built upon the open data space to create improved data portfolios, generating their own natural capital. Each Indicator is a composite measure of different conditions aggregated at multiple spatial resolutions data products to fill temporal gaps or improved resolution through techniques like backfilling. at different time periods for specific locations across a landscape or a seascape. However, many of the key environmental topics, such as biodiversity, cannot be developed purely from remote sensing but require in-situ ground data much of which is held by the NGOs or intergovernmental institutions. Linking in-situ observations with remote sensing efforts will THE SPECIFICS be essential to creating improved data products. And until this is resolved, open or private Earth Knowledge generates its ‘digital twins’ by constructing and running numerous Earth system and Earth subsystem sector developments are likely to continue to face restrictions or will require entirely novel process models to characterize past, present and potential alternative future environmental processes and conditions. technical approaches to improve the global environmental data portfolio. Where required, lower resolution (more global) data or model outputs are appropriately downscaled using spatial, statistical or dynamical downscaling methods that are suitable for the data and the model from which the data originated. The question that comes next is, considering these limitations, what can be achieved now? To give insight into the extent of what is possible, we provide an example from the commercial space on the next page, showing how complex models are capable of overcoming Process models used to describe biophysical processes at multiple spatial and temporal scales must meet several key many of the data limitations to provide greater insights than the sum of their geospatial parts. criteria in order to be used. These include that the biophysical process models must: From there we explore three case studies across multiple scales in Brazil, showcasing step by step the various insights which can be gained and the various data challenges involved, 1. Be developed or available in the public domain starting at the asset level, looking at mining in Brazil. 2. Have undergone significant peer review in many different journals and/or organizations 3. Be used in many different landscape environments and settings 4. Be used in many different geographic locations BOX ONE 5. Be applicable at multiple spatial and temporal scales and 6. Be used by a broad user community EARTH KNOWLEDGE For each Earth system and Earth subsystem model, Earth Knowledge identifies and selects authoritative data sources Authors: Frank A. D’Agnese, President and CTO, Earth Knowledge, Inc. and Julia Armstrong based on: D’Agnese, CEO, Earth Knowledge, Inc. 1. Global uniformity and extent (global, regional, locally specific data set) The Earth Knowledge Planetary Intelligence Platform continually assesses Earth 2. Date of collection (or future projection) systems. Our platform leverages a ‘digital twin’ of the Earth integrating authoritative data and models of the Earth’s interconnected systems, from the subsurface to the 3. Spatial resolution upper atmosphere. This digital Earth represents the varying conditions of landscapes 4. Methodology of data acquisition and development and seascapes at multiple spatial and temporal resolutions, from approximately 125 years into the past to 150 years into the future. 5. Official verification of the data developers and their organization and 6. Assessment of their source organization’s quality assurance and quality control procedures The Earth Knowledge Platform translates scientific data, geospatial data and Earth- systems models into 300+ indicators related to the direct drivers of global change Candidate source datasets are profiled, qualified and sampled to establish their suitability for acquisition and processing. and the commonly described three pillars of sustainability (Natural Capital, Social Capital, and Economic Capital). These direct drivers of global change, which lead to biodiversity loss and habitat degradation, include climate change, pollution, Once the models are calibrated and evaluated to determine how well the models represent natural conditions over the invasive species and disease, over-exploitation of natural resources, and land and 125-year historical period, the models are then re-run under varying conditions that represent different potential future sea conversion. The indicators help assess global change and sustainability actions states that may occur as a result of climate change and other forms of global change. and provide a quantitative way to measure impacts and related potential risks and opportunities at any location on the globe. Future projections of a 150-year period are calculated beginning in 1950 and simulated through to 2100. The overlap period from 1950 through 2020 is used so that there is sufficient repetition in the historical model and the forward- looking projections to determine the potential for any model bias that could exist that may be introduced from field observation data. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 27 CASE STUDY ONE ASSET LEVEL ASSESSMENT – MINES IN BRAZIL – WWF To illustrate the value of a geospatial approach at the asset level, WWF’s Conservation Intelligence team45 directly compared all commercial mines (763) within Brazil against several environmentally relevant vector and raster observational datasets. Three case mines are used throughout to illustrate the various complications which arise even in the most basic assessments. The example mines are as follows: 1) Aurizona (gold) operated by Equinox Gold Corp; 2) Capanema (iron ore) operated by Vale S.A.; 3) Northern System (iron ore) operated by Vale S.A (Figure 6).46 The results can be presented in a variety of ways depending on the user’s application. Directly, each asset versus each observational dataset, or each asset impact modelled in some way against multiple observational datasets in combination. Or, asset scores can be aggregated to each parent company to provide insights at a parent level rather than at an individual asset level (explored in the next case study). To aid understanding of the approach, we focus here on directly reporting results per variable at the asset level. Results can of course be integrated alongside other traditional ESG metrics. Figure 6 All mines within Brazil and the locations of the three mines used to illustrate basic issues in geospatial ESG assessments: 1) Aurizona (gold) operated by Equinox Gold Corp 2) Capanema (iron ore) operated by Vale S.A. 3) Northern System (iron ore) operated by Vale S.A. 1 3 2 ACTIVE MINES © EDWARD PARKER / WWF INACTIVE MINES WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 29 BOX TWO SATELLITE IMAGERY – VISUAL ASSESSMENT METHOD Mining projects as defined by S&P Global Metals and Mining dataset (data sourced as at During a technical geospatial assessment, it’s worth noting the value of access to time-lapse satellite March 2021) were given a 1km² area47 and compared using ArcGIS 10.8.1 against several imagery and satellite imagery on demand. This provides an ESG analyst a rapid means of placing the vector layers: project in a spatial and temporal context, without needing to source any additional details. Where is it? What surrounds it? When did the project start (if post 1980s)? What was the status of the • Protected Areas, World Database of Protected Areas, 202148 environment before the project was initiated? How has the project expanded? Time-lapsed imagery • Key Biodiversity Areas (KBAs), 2021 can rapidly help provide an analyst with context to these questions, by effectively playing a short ten second video of how the site has changed since the mid-1980s. For example, below is freely available • World Heritage Sites, 2021 NASA and ESA imagery showing one of the three mining sites, Aurizona, from the 1980s to 2020/21. It shows that the site was mangrove forest in 1986, developed in 2013, and expanded up to the present, We also considered Brazil’s mines against several openly available raster layers: with the tailing dam increasing in size. • Ecoregions49 Of course, if you have the necessary resources, this data can be quantified, as outlined on Page 36. • Biodiversity Intactness Index50 • Ground Carbon51 • Forest Loss, 201952 • Forest Structural Condition Index (FSCI), 2020 – data for the Tropical and Subtropical Moist Broadleaf Forests Biome53 • Forest Structural Integrity Index (FSII), 2020 – data for the Tropical and Subtropical Moist Broadleaf Forests Biome54 Subdividing mining assets into categories, such as the ecoregion, elevation, etc., allows us to use that data point to later adjust other variables against. For example, the same mine in open grassland would have differing immediate and ongoing impacts if in a different habitat, say rainforest. Here, as one example, we use ‘Ecoregions’ to illustrate the approach, but more Figure 7 Figure 8 complex approaches can use any number of these differentiating variables. Aurizona site limited mining activity – Aurizona site limited mining activity – Landsat – 27th August 1986 Landsat – 8th October 2013 Other observational layers are used to provide direct measures. Any number of observational datasets could be applied, using both static and dynamic inputs (e.g. near-real-time fire data or live feed weather data). Here we consider just a small number to describe the concept. In more detailed assessments it is common to consider 50+ observational datasets with interdependencies. Each dataset, depending on its design, needs to be treated differently to achieve useful insights. Some can be considered as is, without processing; for example, the Biodiversity Intactness Index provides a simple value that can be extracted and averaged. Most, however, require analysis, such as forest fragmentation, which needs to consider the length of fragmented habitat linked to the linear infrastructure of the mine site to provide insight into the mine’s associated secondary impacts. Figure 9 Figure 10 Aurizona site mining activity – Aurizona site mining activity – Sentenel 2 - 4th October 2020 Sentenel 2 - 4th October 2020 (Rendered Shortwave Infared) WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 31 Biodiversity Ground Protected Key Property Forest Loss - List of Owners Ecoregion Name Score Intactness Carbon FSCI Score FSII Score Areas Biodiversity Name 2019 Score Index Score Score Score Areas Score RESULTS Tropical & Subtropical Moist Pitinga Industrias Nucleares Do Brasil SA (Owner) 100% 6.66 0.15 0.35 0 0.56 0.79 2.04 1.28 Broadleaf Forests Tropical & Subtropical Moist Salobo Vale S.A. (Owner) 100% 6.6 0.14 0.35 1.54 0.09 0.02 0.68 1.28 Broadleaf Forests In total, 763 commercial mines were identified within Brazil, of which 263 (34%) are considered Boa Vista GoldMining Inc. (Optionor) 84.05%; Boa Vista Gold Inc. (Optionor) 15.95% Tropical & Subtropical Moist Broadleaf Forests 6.6 0.15 0.35 0.35 0.53 0.74 0.68 1.28 ‘active’. Overall, out of the 263 active mines, 31 overlapped with KBAs, 26 of which were Salobo West Vale S.A. (Owner) 100% Tropical & Subtropical Moist Broadleaf Forests 6.59 0.15 0.35 0 0.58 0.82 0.68 1.28 entirely within KBAs. For Protected Areas, 40 active mines overlapped with one or more Morro Dos Seis Cia Brasileira de Metalurgia e Mineracao (Owner); CPRM (Owner) Tropical & Subtropical Moist 6.59 0.14 0.35 0 0 0 2.04 1.28 Lagos Broadleaf Forests protected areas,55 22 of which were entirely within PAs. Only one currently inactive mine was Tropical & Subtropical Moist Serra Norte Vale S.A. (Owner) 100% 6.58 0.12 0.35 0.96 0.18 0.05 0.68 1.28 identified within a World Heritage Site within Brazil. Broadleaf Forests Tropical & Subtropical Moist Xingu Unnamed Owner (Owner) 100% 6.57 0.15 0.35 0.41 0.51 0.72 0 1.28 Broadleaf Forests Each mine site was scored against the example geospatial layers: defining ecoregions, the EMA BBX Minerals Limited (Optionee) 100%; Private Interest (Optionor) Tropical & Subtropical Moist Broadleaf Forests 6.57 0.15 0.35 0.03 0.64 0.9 0 1.28 mean score for Biodiversity Intactness Index, Ground Carbon, Forest Structural Condition Alemao Vale S.A. (Venturer) 67%; Federal Government of Brazil (Venturer) 33% Tropical & Subtropical Moist Broadleaf Forests 6.56 0.13 0.35 0.03 0.39 0.39 0.68 1.28 Index, Forest Structural Integrity Index, and total area of Forest Loss within a simple 1km2 Rio Cristalino Colossus Minerals Inc. (Owner) 100% Tropical & Subtropical Moist 6.56 0.14 0.35 0 0.59 0.83 0 1.28 Broadleaf Forests circle56 around each mine site. The three example mines had the following results; Tropical & Subtropical Moist Amazonas Potassio Ocidental Mineracao Ltda. (Owner) 100% 6.56 0.15 0.35 0 0.56 0.79 0 1.28 Broadleaf Forests Estanho de Tropical & Subtropical Moist Companhia Siderúrgica Nacional (Owner) 100% 6.55 0.12 0.35 0.21 0.16 0.23 0.68 1.28 Rondonia SA Broadleaf Forests Mine Name Aurizona Capanema Northern System Aurizona Equinox Gold Corp. (Owner) 100% Mangroves 6.55 0.14 0.17 0.5 0 0 1.36 1.28 Tropical & Subtropical Moist Bahia Tecstones Geologia Ltda (Owner) 55%; Private Interest (Owner) 45% 6.55 0.12 0.35 0.08 0.37 0.11 0.68 1.28 Broadleaf Forests Tropical & Subtropical Tropical & Subtropical Ecoregion Mangroves Grasslands, Savannas Moist Broadleaf Trauira BTG Pactual Mining S.A. (Owner) 88.08% Mangroves 6.55 0.16 0.11 0.24 0 0 1.78 1.28 & Shrublands Forests Tropical & Subtropical Moist Para-Amazonas Cowley Mining plc (Owner) 100% 6.54 0.15 0.35 0.22 0.36 0.51 0 1.28 Broadleaf Forests Tropical & Subtropical Moist Biodiversity Intactness Index N5 Vale S.A. (Owner) 100% Broadleaf Forests 6.54 0.12 0.35 0.36 0.08 0.02 0.68 1.28 0.94 0.66 0.73 (Mean Score) Tropical & Subtropical Moist Azul Vale S.A. (Owner) 100% 6.53 0.12 0.35 0.26 0.05 0.01 0.68 1.28 Broadleaf Forests Ground Carbon Salobo South Unnamed Owner (Owner) 100% Tropical & Subtropical Moist 6.53 0.15 0 0 0.61 0.86 0.68 1.28 9650 8700 0 Broadleaf Forests (Mean Score) Tropical & Subtropical Moist Vale do Ribeira Cia De Pesquisa De Recursos Minerais (Owner) 100% 6.53 0.14 0.09 0.02 0.54 0.57 0.68 1.28 Broadleaf Forests Forest Loss 2019 Tropical & Subtropical Moist 0.99 0.0026 0.0215 Serra Sul Vale S.A. (Owner) 100% Broadleaf Forests 6.52 0.13 0.35 0.07 0 0 0.68 1.28 (km²) Tropical & Subtropical Moist Iporanga CPRM (Owner) 100% 6.52 0.13 0.35 0.01 0.5 0.14 0 1.28 Broadleaf Forests Forest Structural Condition Index (FSCI) No Data No Data 1.26 Brazil Unnamed Owner (Owner) 100% Tropical & Subtropical Grasslands, 6.52 0.14 0.16 0 0 0 1.36 1.28 (Mean Score) Savannas & Shrublands Tropical & Subtropical Moist Patrocinio Belo Sun Mining Corporation (Owner) 100% 6.51 0.14 0.17 0.55 0 0 0.68 1.28 Broadleaf Forests Forest Structural Integrity Index (FSII) No Data No Data 0.12 Corrego do Sitio AngloGold Ashanti Limited (Owner) 100% Tropical & Subtropical Moist 6.51 0.14 0.16 0.25 0.3 0.06 0.63 1.28 (Mean Score) Broadleaf Forests Tropical & Subtropical Moist Igarape Bahia Vale S.A. (Venturer) 87%; Federal Government of Brazil (Venturer) 13% 6.51 0.12 0.18 0.02 0.23 0.31 0.68 1.28 Broadleaf Forests Protected Areas 6.2857 3.13 3.14 Tropical & Subtropical Moist (Area Overlap – km²) Volta Grande Belo Sun Mining Corporation (Owner) 100% Broadleaf Forests 6.5 0.13 0.35 0.29 0 0 0 1.28 Tropical & Subtropical Moist Santa Barbara Companhia Siderúrgica Nacional (Owner) 100% 6.49 0.13 0 0 0.37 0.52 0.68 1.28 Key Biodiversity Areas Broadleaf Forests 3.14 3.14 3.14 (Area Overlap km²) Ribeirao do Carmo Cia Minas da Passagem (Owner) 100% Tropical & Subtropical Moist Broadleaf Forests 6.49 0.13 0.16 0.23 0.39 0.23 0 1.28 Tropical & Subtropical Moist Paragominas Norsk Hydro ASA (Owner) 100% 6.48 0.12 0.17 0.49 0.09 0.05 0 1.28 Broadleaf Forests Figure 11 – Table showing the results for the three case study mines. Capanema Vale S.A. (Owner) 100% Tropical & Subtropical Grasslands, 6.48 0.1 0.16 0 0 0 0.68 1.28 Savannas & Shrublands Tropical & Subtropical Grasslands, Cata Preta Ouro Preta Mineracao Limitada (Owner) 100% 6.47 0.13 0.16 0.43 0 0 0 1.28 Once the above data are pulled together, it is possible to begin to build simple high-level Savannas & Shrublands Tropical & Subtropical Moist ‘environmental’ geospatial screening for commercial mines, using relative rankings to show Cata Preta Ouro Preta Mineracao Limitada (Owner) 100% Broadleaf Forests 6.47 0.13 0.16 0.43 0 0 0 1.28 outliers for each observational dataset. However, there are some additional considerations. Vetria Rumo S.A. (Venturer) 50.38%; Vetorial Siderurgica Ltda (Venturer) 33.83%; Triunfo Participações e Investimentos S.A. (Venturer) 15.79% Tropical & Subtropical Dry Broadleaf Forests 6.47 0.1 0.23 0.29 0 0 0 1.28 For example, many of the observational datasets above are ‘forest’ related (i.e. forest loss, Rabicho MMX Mineração e Metálicos S.A. (Owner) 100% Tropical & Subtropical Dry Broadleaf Forests 6.47 0.12 0.23 0.15 0 0 0 1.28 ground carbon, FSCI, FSII). Consequently, we risk biasing scores towards mines in areas with Corumba Vale S.A. (Owner) 100% Tropical & Subtropical Dry Broadleaf 6.47 0.13 0.23 0.12 0 0 0 1.28 Forests high forest cover vs. mines in areas without forest cover, e.g. savanna. Here benchmarked Tropical & Subtropical Moist N4W Vale S.A. (Owner) 100% 6.46 0.11 0.01 0.22 0.04 0.01 0.68 1.28 weightings on the ecoregion type could address these implications, or alternatively, users may Broadleaf Forests Tropical & Subtropical Moist be interested to identify mines with high risk to topical forest. Of course, any real application N4E Vale S.A. (Owner) 100% Broadleaf Forests 6.46 0.11 0.01 0.17 0.04 0.01 0.68 1.28 would need to carefully consider the application of observational datasets to best meet the Cajati Mosaic Company (Owner) 100% Tropical & Subtropical Moist Broadleaf Forests 6.46 0.14 0.17 0.11 0 0 0 1.28 needs of its intended application. Fabrica Nova Vale S.A. (Owner) 100% Tropical & Subtropical Grasslands, Savannas & Shrublands 6.45 0.12 0.16 0.17 0 0 0 1.28 Southeastern Tropical & Subtropical Grasslands, Vale S.A. (Owner) 100% 6.45 0.13 0.16 0.1 0 0 0 1.28 System Savannas & Shrublands The results for 50 mines are shown in Figure 12, illustrating how this method offers a high-level Qualimarcas Comercio E Exportacao de Cereai (Venturer); Socios Quotistas de Tropical & Subtropical Grasslands, Canastra 6.45 0.12 0.01 0 0 0 0.68 1.28 means to rapidly and consistently screen the active mines identified in Brazil, or indeed all Mineracao do Sul Ltda (Venturer) Savannas & Shrublands Northern Tropical & Subtropical Moist mines globally. Out of the three case study mines, all three are highly ranked, with Aurizona System Vale S.A. (Owner) 100% Broadleaf Forests 6.45 0.11 0 0.01 0.05 0.01 0.68 1.28 highest. This provides a useful high-level overview, but as outlined in the next section, it is vital Timbopeba Vale S.A. (Owner) 100% Tropical & Subtropical Grasslands, Savannas & Shrublands 6.45 0.12 0.16 0.01 0 0 0 1.28 to dig into the data further to understand the results. Cacapava do Sul Nexa Resources S.A. (Optionor) 75%; IAMGOLD Corporation (Optionee) 25% Tropical & Subtropical Grasslands, Savannas & Shrublands 6.43 0.1 0.11 0.01 0 0 0 1.28 Tropical & Subtropical Moist Passagem Cia Minas da Passagem (Owner) 100% 6.43 0.14 0.01 0.01 0.18 0.03 0 1.28 Broadleaf Forests Tropical & Subtropical Grasslands, Conta Historia Vale S.A. (Owner) 100% 6.43 0.13 0 0 0 0 0.2 1.28 Savannas & Shrublands Figure 12 (Following page) Candiota Cia Riograndense de Mineracao (Owner) 100% Tropical & Subtropical Grasslands, Savannas & Shrublands 6.42 0.12 0.01 0.09 0 0 0 1.28 Table showing a selection of the scores generated for the observational layers Mariana Vale S.A. (Owner) 100% Tropical & Subtropical Grasslands, 6.42 0.13 0 0.05 0 0 0 1.28 run against all active mines in Brazil, reporting 50 example mines.58 Savannas & Shrublands Urucum Vale S.A. (Owner) 100% Flooded Grasslands & Savannas 6.42 0.12 0 0.06 0 0 0 1.28 WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 33 DIGGING DEEPER INTO THE DATA The high-level assessment shows that the three case study mines are fairly highly ranked, with a higher score suggesting a higher likelihood of an ‘environmental issue’, and with Aurizona performing slightly worse than the other two on its initial high level environmental impact scoring. To improve insights, we need to dig into the data. ISSUES TO CONSIDER TEMPORAL AND WIDER CONTEXT When considering Protected Areas, or other area designations, it’s useful to consider if commercial activity predates the designation in question, or has an exception, or is legally allowed to occur within that area. For example, some states allow various forms of extractive activity within certain protected areas, and some designations, such as KBAs, have no legal standing. One way to consider this, if available, is to use the attributes of the datasets, such as dates of designation of the protected area and dates of the mining claim. If this is not possible, additional research may be required to fill data gaps. When we look at the three example mines (Figure 13), the mines predate the PAs. Here, a complication occurs: mines go through long development phases, and the majority of their environmental impact may have occurred under different ownership. Unpicking when the impact occurred and who was the responsible owner at the time can be complex. It may be initially sufficient for most high-level ESG purposes simply to identify when the current owners took control and if that predates any key area designations (Figure 13). Est. Project start Mine Name PA/s Dates IUCN Category KBA Dates (Est. Current owner) Aurizona 1978 (2019) 1991 and 1993 V, Ramsar Site 2009 Capanema 1983 (2014) 1994 V 2009 Northern System 1986 (1986) 1998 VI 2009 Figure 13 - Table showing the dates of designation of protected areas the mines overlap with against estimated project start dates of the mines. Temporal context is also vital for establishing the initial impact of an asset. If an observational data layer is applied after the development of the asset, the scores will be biased by the prior impact of the asset itself (Figure 11, 12). For example, a forest loss metric 10 years after the development of the mine is likely to be low or zero for the mine site, as the more recent dataset will not detect any change in forest cover for the site as it has long been deforested. Subsequently, the impact of a mine needs to be considered across time; this is made more challenging by a lack of consistent observational historic data to draw upon. In some cases, this is impossible with mines or assets predating the archive satellite imagery records (mid- 1980s). In these cases, and others, it may be possible by considering surrounding vegetation to predict the prior state of the area and estimate the site’s initial impact. However, if resources are available, the growth of a mine can often be calculated with remote sensing to show expansion over time (See Page 36). THE NEED TO DIFFERENTIATE WITHIN OBSERVATIONAL DATASETS It is important to differentiate variables within observational datasets to better understand initial and ongoing impacts. For example, no two protected areas, indigenous areas, key biodiversity areas, etc., are equal. Some are high status, some are pristine, others may be heavily degraded. An asset overlap with a conservation area should be considered by the site’s specific values, rather than as a binary value. Within the three case study mines, the protected areas they overlap have differing IUCN management categories and designations: IUCN Cat. V59 or VI60, and one of them is a Ramsar site, a wetland of international importance (Figure 13). ESG analysts might wish to use these or other relevant designation as a useful metric to © LUIS BARRETO / WWF-UK highlight operations with potentially higher risk and impact. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 35 However, beyond simple site attributes, it is vital to consider each site’s wider values. Is the protected area already heavily degraded? Does it have a high endangered species presence? ONGOING IMPACTS Does it contain multiple other commercial operations? Does it have pristine forest? High levels Here we have shown the outlines of a geospatial ESG approach, focusing on defining the initial of deforestation? Do conservation NGOs have a presence within the site? Does the site have a impacts of mines with the open data portfolio. Ongoing impacts (e.g. daily methane pollution) high international internet saliency? Is the site important for tourism? We provide no methodology can in some cases be monitored week by week via remote sensing products. Such monitoring for how to consider this, but it is possible to weigh the values of individual polygons, such as is by its very nature sector-specific – for example, marine oil spill detection and CH4 emissions PAs or KBAs, against hundreds of other variables,61 to provide more in-depth insights as to the are more likely to be relevant to the oil and gas sector and certain types of mining than to, say, likelihood that operations within a site present a significant and immediate reputational risk and to cotton production. Within mining, some remote sensing products provide a means for ongoing help aid in scaling the probable environmental impact at a high level. The three case study mines, monitoring, where, for example, forest loss and land cover change datasets help to show if the for example, each overlap with different PAs, each with a different number of IUCN endangered mine is expanding. Whilst other datasets, such as infra-red heat profile, and depth and extent of species present (Table 14). an open pit, might be useful to predict CH2 emissions and productivity. Commonly these products are provided by the commercial space, although some are within the open domain. In the future, additional ground data sources, such as in-situ smart meters or landscape audio, are likely to play IUCN Red Listed species with PA Mine Name (All / Least Concern / Near Threatened + Vulnerable WWF CI PA Screening Score a clear role in defining insight into an asset’s ongoing environmental impacts and stresses. + Endangered + Critically Endangered) Aurizona 890 / 794 / 96 0.34 Capanema 778 / 714 / 64 0.29 SUPPLY CHAIN IMPACTS Northern System 719 / 671 / 48 0.27 The emerging world of geospatial ESG is currently limited by which sectors have robust asset datasets. Currently, most are primary industries: mining, oil and gas, power plants, fishing, Figure 14 – Table showing additional information about the endangered species likely to be present in the PAs overlapped by shipping, cement, etc., whose environmental impact is primarily linked to actual operations. the three mines to improve differentiation of possible impacts.62 Secondary and higher industries, whose impact is mostly in the supply chains, will more pressingly need robust supply chain assessments. Naturally, this is only possible where supply Aurizona, which is on a Ramsar site on the coast, faces the highest scores, as could be expected; chain data is available,63 which currently is rarely the case. To resolve this, we commonly see but the two other sites also face regionally and internationally high scores, as would be expected regional ‘impact’ or ‘risk’ averages developed, where a company might not know which exact in tropical regions, which will naturally have a high global level of biodiversity. area the product was sourced from, but they know the state or region. Thus, we can provide an averaged regional risk or any number of regional values to provide some level of insight into that Of course, mining projects bordering or outside PAs, KBAs or other key designations can still supply chain. This is, of course, limited, but until we achieve greater transparency around supply cause significant damage to the natural world. Regardless of the situation, any commercial chains, geospatial ESG insight and applicability will remain constrained. In the next case study, operation outside key designations does not necessarily legitimize or negate the biodiversity we explore the value of this regional averaged approach (See Case Study 2). impact or reputational and material risks of the activity. Every site needs to be considered for its impact. Key area designation assessments as above provide a useful data point, but they should Within this mining example, we could, for example, define every power plant in Brazil, its type, and always be considered in connection with other geospatial and traditional ESG data points to build output, and then create renewables vs. non-renewables ratios per municipality as a proxy for the out a wider high-level understanding of the site’s impact. likely renewable power usage of mines within Brazil. This value and others can then be modelled against mining production, commodity type, etc. to give insights into the mine’s production SECONDARY ENVIRONMENTAL IMPACTS versus its high-level environmental impact efficiency. Secondary impacts are often tricky to capture and require detailed and tailored methods. For example, many mines in Brazil have opened previously intact forest to wider secondary CONCLUDING REMARKS exploitation by building roads to build new mines. This issue can be captured in a geospatial layer The approach outlined here is relatively simplistic. It is possible to develop the approach in far like forest fragmentation, but this needs to be correctly applied over historic years, for which there greater detail, considering the exact footprint of the asset against a more sector specific model, may not be open data. to better define the initial environmental impact and then the ongoing impact. The idea here is not to promote an exact method but to outline the basics of the approach in order to illustrate the SECTOR SPECIFIC APPROACHES concept of geospatial ESG. Discussing how various factors need to be considered, it becomes apparent that off-the-shelf data products often need to be refined. Yet even with this approach The various complications highlighted above show the strong need to develop sector-specific and with currently available data, insights are possible which arguably are useful to consider geospatial methods if actors wish to gain maximum insight and more precisely attribute alongside traditional ESG data portfolios. environmental impact. Within this example, with slightly more advanced manipulation of the data, it is possible to integrate the attributes of the mining asset data. For example, data defining the In the next section, we outline examples of additional sector specific insights that are possible via mine’s pit type (open or closed), commodity type (i.e. gold, iron ore), production status, work commercial remote sensing providers and show how more refined insights are possible. history and tailings volume can be used to differentiate mines, separating different types of mines   to assign different impact weightings more precisely. Beyond this, it is possible to build highly complex models utilizing all the various data and associated attributes, to consider assets over time, secondary impacts and near-real-time changes. For example, mining sites with tailing dams located in areas historically exposed to extreme rain events are potentially more suspectable to risk of tailing dam failure. This risk can be better triangulated using other variables, such as elevation of the dam, dam size, work history, surrounding urban population, water height, habitat types, water dynamics, wildfires and even dynamic weather data to determine at a high level those assets, and those companies, most exposed. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 37 BOX THREE KAYRROS – EXAMPLES OF INSIGHTS VIA COMMERCIAL STEREOSCOPIC 3D MODEL SATELLITE REMOTE SENSING PRODUCTS We built a high-resolution 3D model of a sub-area by combining image processing and machine learning technologies.64 Authors: Claire Bonfils-Bierer, Product Manager - Kayrros and Alexandre d’Aspremont, Chief scientist - Kayrros To illustrate additional remote sensing insights that are possible and that can be integrated to refine geospatial ESG approaches, we look at one of the case study mines, Northern System in Brazil. Specifically, we explore production, mined areas and removed material based on a combination of 3D Reconstruction, SAR Change Detection Index and Land Cover Multispectral Analysis. The goal is to detect changes in production rates as well as expansion of the mine activities and impacts on the surrounding ecosystem. CASE STUDY LOCATION The results below have been produced for the iron ore mining complex, Northern System in Brazil. Depending on the analysis, the whole mine complex of Northern System highlighted in white on the picture below or a reduced area in the south east highlighted in blue on the picture has been covered. A series of analyses from mid-resolution to high-resolution have been conducted to track how the industrial activity of the mine evolved and has impacted the environment and the land. Figure 16 Digital reconstruction of a Mine Northern System sub-area on July 16, 2020, derived from the SkySat constellation. (Sources: Kayrros Analysis powered by Planet.) After tasking a satellite to take a pair of stereo images— multiple images pointed to the same spot on Earth from different angles— we applied proprietary algorithms to transform this set of images into a 3D model, generating a digital topographical snapshot. This technology enables us to produce digital reconstructions of any remote location worldwide, and can be used as a critical tool to assess the volume of material removed from mines or the forest heights around it, for example. Complementary to the 3D rendering made possible by high-resolution imagery, medium resolution multi-spectral images— combined with deep learning algorithms— classify the land in near-realtime (with access to historical data). By using machine learning algorithms to detect different types of land cover, we can then differentiate forest areas from other types of areas, including mined areas (as shown below). This allows us to track forest cover levels in addition to the land occupied by the mine itself across time, as shown in Figure 17. Processing historical Landsat multi-spectral images enabled us to track the land surrounding the mine dating back to 1984. This allowed us to derive a comprehensive overview of the mine’s expansion over a long period of time, complementing the near-real-time coverage that we have today. Figure 15 – Image showing the study areas considered. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 39 Figure 19 Evolution of the Northern System Mine’s surface area from 1986 to 2020. (Source: Kayrros Analysis, Landsat-4 to Landsat-8 images, courtesy of the U.S. Geological Survey.) We generated the Kayrros Mine Activity Index to illustrate the impact of mining activity on the ecosystem. Figure 17 – Top: Optical Sentinel-2 images covering the Northern System Mine’s sub-area on three different dates. Bottom: Land cover classification mask associated with the three different dates; forest areas are shown To tackle this, we built a quantitative index to track production in full systems, derived in brown, water areas in white and other (including mining) areas are shown in yellow. The forest surfaces fromapplying change-detection algorithms onto Sentinel-1 Synthetic Aperture Radar evolve, as shown in the third row. (SAR) images. Then, we built coherence maps based on interferometric principles from the monitored mine; in simpler terms, images of the mine in which each pixel’s value varies from 0 to 1, which represents to what degree the structure of the field changed between two consecutive dates. The change index over the whole area is then derived from aggregating over pixels and normalizing. Figure 20 Top: Overview of the Northern System Mine and mapping of its sub-areas. Bottom: SAR coherence map produced using a Sentinel-1 image taken on September 5, 2020. The blue polygons show a sample of the pits. Dark pixels indicate important changes since the last acquisition. (Sources: Kayrros analysis; contains modified Copernicus data (2018–2020).) Figure 18 – Sample of Landsat-4 and Landsat-8 images on the Northern System Mine taken since 1986. (Source: Kayrros Analysis, Landsat-4 to Landsat-8 images, courtesy of the U.S. Geological Survey.) WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 41 BOX FOUR ENVERUS – SATELLITE INSIGHTS INTO THE OIL AND GAS SECTOR Authors: Nick Volkmer, Vice President, ESG – Enverus and Jingwen Zheng, Lead Data Scientist – Enverus Satellites are bringing unprecedented oversight into global oil and gas operations. Investors today are able to monitor flaring levels, methane leakage rates and development practices from a suite of satellites that is set to expand in the coming years. The visibility is welcomed by institutions looking to align capital with responsible development practices and by producers that want to showcase superior operations, particularly those in North America and Europe. Take Colorado, for example, where the Denver–Julesburg (DJ) Basin sits near the Denver metropolitan area. As the city has grown, neighbourhoods often overlap with oil and gas operations. Satellite imagery and telemetry data allow us to understand how operators are changing development practices in response. Since 2018, DJ operators have increased average land efficiencies, or the amount of hydrocarbons recovered per surface acre disturbed, by about 40%, largely by drilling more wells with longer horizontal lengths (or laterals) from a single surface location (Figure 22). Satellites paired with Land Efficiency (Mboe/Acre) Ft per reported data sets enable precise yet broad monitoring of these field-level activities. Similar workflows are being developed to analyse flaring and methane rates across North2018 2019 and will American operations 2020 be 2021 201 available in the ESG Analytics module in Enverus’ Prism platform. 536 670 729 749 12,81 Figure 21 Quarterly Kayrros Change Index vs. Reported Production (MT). The analysis shows the correlation between the index 800 25,000 generated and actual production. (Mboe Recovered/Surface Acre Disturbed) 700 Feet of Lateral Drilled per Acre (ft/acre) (Sources: Kayrros analysis, contains modified Copernicus data (2018–2020).) 20,000 600 Land Efficiency 500 15,000 400 10,000 300 CONCLUSION 200 5,000 100 Data fusion is key to deriving a comprehensive view of mines, both historically and in 0 0 near-real-time. Remote sensing sources and sensors— such as multi-spectral, stereo and 2018 2019 2020 2021 radar— offer a wide range of signals at different spatial and temporal scales. This provides direct insights on both industrial activity and its environmental footprint. Land Efficien cy Feet of Lateral D rilled per Acre Figure 22 (Left) Satellite imagery shows an algorithmically detected well pad location (red box) and subsurface well lateral locations (pink lines) for a pad in the DJ basin. (Right) The primary y-axis shows the average DJ basin land efficiency calculated as the sum of the estimated ultimate recovery (EUR) of wells on a pad over a 30-year production profile over surface acres disturbed, and the secondary y-axis shows feet of lateral drilled per surface acre disturbed. CASE STUDY TWO WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 43 COMPANY LEVEL ASSESSMENT – SOFT COMMODITIES IN BRAZIL – TRASE Author: Helen Bellfield, Policy Director (Trase Lead) – Global Canopy In many cases it isn’t possible to identify where specific products are produced. In these situations, without a defined asset location, what insights can a geospatial ESG approach bring to the table? One solution is to consider data at a regional level, as conducted by the Trase tool, which provides insights into the deforestation risk within soft commodity supply chains. The production and trade of soft commodities, including soy, beef and palm oil, is associated with the conversion and degradation of tropical forests and native vegetation. A number of banks, investors and companies have made voluntary commitments to remove deforestation from their portfolios and supply chains, and in November 2021 the EU Commission published a proposed regulation to prohibit the placing of products and commodities associated with deforestation and forest degradation on the EU market. Assessing the deforestation risks associated with a specific company’s supply chains and sourcing requires mapping products back to production regions. This presents a significant challenge due to issues that include long supply chains with indirect suppliers; aggregation and bulking of commodities such as palm oil, soy and maize; and the size of the supply base. This case study illustrates an approach pioneered by Trase to use publicly available data to map soft commodity supply chains at scale to connect per shipment trade data to subnational sourcing regions. In the case of Brazil’s soy exports, Trase maps 100 million tonnes of soy that was exported in 2018 by 300 trading companies and estimates that these exports were associated with 50,000 ha of soy deforestation risk. METHODS 1. Mapping soy supply chains Trase links Brazilian soy exports to sub-national municipalities of production via processing (e.g. crushing facilities, refineries) and storage facilities (e.g. silos), as well as deforestation impacts in these municipalities. First, we link individual export shipments back to municipality locations of taxation considering both the trader (with tax information) and the Brazilian state of production corresponding to the farms, silos, crushing facilities or wholesale retailing (i.e. trader assets) linked to the export shipment. Second, we link these assets with the municipalities of production where the soy was most likely produced (not to individual farms except in cases where we can make a direct link), through a minimum cost flow analysis using linear programming. This approach is optimised using the combination of trader assets, domestic consumption and export demand for soybeans, and transportation costs to identify the most likely municipality of production supplying these silos and crushing facilities. Exports are then aggregated annually and by trading company to provide an annual sourcing map by trader. Please see more information here. 2. Assessing soy deforestation In each soy producing municipality, Trase assesses recent soy deforestation. This is calculated by comparing the area of production associated with a specific harvest and export of soy to recent deforestation that has directly contributed to the production of that harvest. We estimate this based on the time it can take between the initial deforestation of an area of land and the processes of acquiring, preparing and selling the land before soy is typically planted. This is estimated to be five years for soy in Brazil. In addition to this ‘allocation period’, we also consider a one-year ‘lag period’ representing © MARIZILDA CRUPPE / WWF-UK the minimum time needed between a deforestation event and the harvest of soy. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 45 To derive direct deforestation associated with soybean production within the five-year 5. Translating data into ESG metrics allocation period, we: Red, amber and green flags are used to highlight areas of deforestation risk exposure for each trading company, • Put together an annual deforestation increment map (30m resolution) combining Amazon, These include the following environmental metrics: Cerrado, Atlantic Forest and Pantanal deforestation. In cases where data is not available annually (for example, for earlier years of the Cerrado time series, and for the Atlantic • The company is one of the top 10 exporters of Brazilian soy Forest), we obtain the annual mean deforestation by dividing per-pixel deforestation by the timeframe between two deforestation datasets. Note this covers primary and not secondary • Soy deforestation risk (ha) deforestation. • Relative soy deforestation risk (ha/1000 tonnes) • We process annual maps of soy extent (30m resolution) to remove fragments less than 20 • The company sources from high risk regions (e.g. the Matopiba region, a current soy deforestation frontier) hectares based on IBGE data on soy farms. • The company sources from top quartile of municipalities with the highest soy deforestation • We then compare total soy coverage in year y (e.g. 2019) to historical deforestation increment maps over the preceding five-year period (2014–2018 inclusive) as illustrated in Figure 23. DATA SOURCES • Finally, we aggregate this to the municipality level. Supply chain mapping • Per shipment data 2019 total soy coverage Deforestation from 2014 to 2018 Soy deforestation in 2019 (290ha) (140ha) (50ha) • Asset ownership and related activities – e.g. soy crushing: CNPJ, ABIOVE The soy deforestation is divided by • Soy production data: IBGE the deforestation allocation period (5years) and embedded in the exports as risk measure according See Brazilian soy methods documents for more information. to the total volume that an actor source from the jurisdiction, for example: Soy deforestation risk Trader A: 400 tonnes Trader B: 600 tonnes • Soy crop extent (30m resolution): Global Land Analysis & Discovery (GLAD) – University of Maryland Trader A: (50 * 40%)/5 or 4 ha Trader A: (50 * 60%)/5 or 6 ha • Deforestation (30m resolution): - INPE Prodes Amazon (1998–2019 annual) - INPE Prodes Cerrado (2000–2012 every two years; 201–-2019 annual) Figure 23 – To calculate soy deforestation in 2019, we overlay soy production in 2019 with deforestation increment maps from 2014–2018 inclusive to identify soy deforestation. - SOS – Mata Atlantica (2000–2005 (every six years); 2006–2008 (every three years); 2008–2010 (every two years); 2011–2016 (annual) - SOS – Pantanal (2003–2008 (every six years); 2009–2016 (every two years) 2017 (annual) 3. Assessing trader’s soy deforestation risk We connect soy deforestation to soy exports to create a measure of soy deforestation risk Note that recently more comprehensive data on deforestation that cover the entire country has become available from associated with each trader’s supply chain. To estimate soy deforestation risk associated with MapBiomas that will provide a single source of data negating the need to patch together multiple sources. exports, the estimated share of soy that is purchased by each trader or for each producing municipality (step 1) and assign soy deforestation in each municipality (step 2) proportionally See Commodity deforestation and commodity deforestation risk for more information to each trader. For example, if a trader buys 20% of a municipality’s soy in a given year, it gets 20% of the municipality’s soy deforestation. It is important to emphasise that this measure estimates the risk that a commodity trader is exposed to deforestation in its supply chain, Company legal hierarchy based on the jurisdictions it is sourcing from. • Financial service providers – e.g. Factset, Refinitiv Permid • Open corporates 4. Aggregating soy deforestation risk to parent companies • GLEIF In many cases, parent companies include different subsidiaries that are exporting and • National Companies House registries including CNPJ importing soy from Brazil. For example, Cargill Brazil exports soy, and Cargill France imports soy. This means that subsidiaries within the same company may trade with each other. Therefore, in aggregating soy deforestation risk at the parent company level, we need to avoid double counting the risk where one subsidiary exports volumes imported by another subsidiary. We calculate the total risk of parent companies as the total soy deforestation risk associated with all the company’s subsidiaries’ exports plus the total soy deforestation risk from all the company’s subsidiaries’ imports, excluding imports from the company’s own subsidiaries (as these have been accounted for under exports). WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 47 RESULTS Sourcing from the 6 municipalities that A hundred million tonnes of soy were exported from Brazil in 2018 by 300 trading companies. Sourcing soy from represent the top 25% Trase estimates that these exports were associated with 50,000 ha soy deforestation risk. The Matopiba (high of soy deforestation trade is highly concentrated, with the four ‘ABCD’ traders handling 50% of exports. These Soy exports in 2018 Soy deforestation risk risk region) in 2018 risk in 2018 Company soy traders’ exports are associated with 43% of soy deforestation risk. Company (million tonnes) in 2018 (ha) (million tonnes) (million tonnes) policy (Forest 500) ADM 11.4 6,474 1.01 0.66 60% However, depending on their sourcing patterns, these traders have different exposure to soy deforestation risks. Louis Dreyfus accounts for 10% of total exports but only 1% of total soy Bunge 15.7 11,197 1.31 0.58 57% deforestation risk, because it mainly sources from the south of Brazil where forests were cleared many years ago. In contrast, Bunge trades 16% of exports but accounts for 22% of Cargill 12.8 5,432 0.81 0.06 41% total soy deforestation risk due to its sourcing regions including the Matopiba region, where soy deforestation is currently happening (Figure 24). Louis Dreyfus 9.8 493 0 0 46% The concentration of risk in a handful of production regions is highlighted by the fact that 50% of soy deforestation risk associated with Brazil’s soy exports are from 1% of soy producing Figure 25 Environmental risk metrics for the ABCD traders of Brazilian soy. municipalities (18 out of 2318 municipalities). LIMITATIONS Trade Volume Traceability. Soy supply chain traceability gaps remain an important barrier in linking products handled by supply chain companies to soy deforestation impacts. However, Trase demonstrates that it is possible, through utilising existing public datasets, to create a supply chain map that links exports to producing regions and therefore to deforestation impacts in these regions. While this is an important step forward, Trase cannot directly attribute responsibility for deforestation to specific companies, as data on precise sourcing patterns back to individual farms are not publicly available. Among other data sources, Trase uses information publicly disclosed by companies in its supply chain mapping. As company sourcing data becomes more transparent, Trase can adjust its estimates of a company’s deforestation risk and reflect demonstratable deforestation-free sourcing being achieved by more progressive companies. Company’s Deforestation Risk Indirect land use change. Measures of direct soy deforestation only tell part of the story. While the majority of soy expansion in the Brazilian Amazon over the past decade has taken place onto land already cleared for pasture, the overall area of pasture remains more or less unchanged. In other words, as pasture is converted into agricultural land for soy and other crops, forest and savannah are cleared for new pasture. This suggests that soy expansion is indirectly driving deforestation. Gaps and time-lags in the availability of data. Trase data for Brazilian soy is only currently available for 2018 due to gaps in data availability for more recent years. More broadly, government data is often published with a time lag. While Trase time series data show that sourcing patterns do change over time, they also indicate that such supply chains remain ‘sticky’ – many of the larger trading companies are vertically integrated and have significant investments in soy silos, crushing facilities and port terminals as well as relationships with Trade Volume farmers, including via the provision of finance and inputs. There is an opportunity to also use this historic data to predict future deforestation risk. Company’s Deforestation Risk CONCLUSION Assessing environmental risks associated with soft commodity supply chains requires mapping products back to farms and concessions or at least to sub-national regions of production. While traceability and transparency remain significant barriers to mapping soft Figure 24 commodity supply chains, this case study demonstrates an approach for 1) mapping soy Sourcing map of soy exports in terms of volumes and associated soy deforestation supply chains and 2) connecting supply chains to soy deforestation using publicly available risks of the ABCD soy traders who dominate the trade. data that already can provide useful insights, such as the high concentration of risks in these supply chains, that can guide investor engagement with clients. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 49 CASE STUDY THREE SOVEREIGN LEVEL ASSESSMENT – A GEOSPATIAL VIEW ON DROUGHTS AND EMPLOYMENT IN BRAZIL – THE WORLD BANK Authors: Dieter Wang, Sustainable Finance Specialist, The World Bank and Bryan Gurhy, Senior Financial Sector Specialist, The World Bank INTRODUCTION The rapidly growing availability of geospatial data paves the way for better Environmental, Social and Governance (ESG) scores and ultimately, better informed ESG investing.65 This holds true not only for corporate entities, but also for sovereign nations, which this chapter focuses on. Asset managers, pension funds and other institutional investors are integrating ESG factors into their investment portfolios, which are a major source of capital flows in global financial markets.66 For example, in 2018, the investment bank J.P. Morgan and the asset manager BlackRock launched the JESG index, which incorporates ESG considerations into existing flagship benchmark indices that track government bonds in emerging markets (EM).67 As a report by the International Monetary Fund finds, investments that track benchmark indices have grown rapidly in EM bond markets, standing at around US$ 300 billion in 2019.68 Sovereign ESG scores, which lay the foundation for the operationalization of ESG investing in sovereign fixed-income markets, are not without controversy as two recent World Bank reports document.69 For example, ESG score providers generally agree on what constitutes a good sovereign performance for Governance and Social issues. However, this is driven by the ingrained income bias, which refers to the fact that 90 percent of sovereign ESG scores can be explained by a country’s national income.70 In comparison, there is considerably less agreement on what constitutes a good score on the Environment pillar. This is due to disagreements on what “good” performance is on a conceptual level, but also due to data gaps, out-of-date statistics, and heterogeneous reporting standards, which often force providers to fill in and estimate missing values. Moreover, even if records are available on the national level, corresponding subnational data rarely exists. Comparability across countries depends heavily on capabilities of national statistical offices.71 Geospatial data presents a promising solution with global, consistent, and highly frequent coverage that is objective in nature. The two World Bank reports also argue that better data measurement alone is not sufficient. Even though geospatial data helps better assess the environmental materiality of an indicator, such as better measurements of deforestation, desertification, or coral bleaching, it does not directly translate into economic materiality, e.g., economic output, employment figures, which in turn influences financial materiality, e.g., risk management or investment incentives.72 It is therefore crucial to process and convert geospatial data into economically meaningful numbers. This does not only refer to the units or the aggregation level of the data, as we will discuss shortly, but also to the very interpretation of the statistics. This chapter showcases how to establish an empirical link between environmental and economic for the case of precipitation anomalies in Brazil’s regional economies. To estimate the strength of this link, we use the local projection methodology that has been widely used to understand how economies respond to events, such as economic policy changes, market disruptions or natural disasters. While it would be interesting to examine the link with financial materiality, we leave this to future research. In the following, we first describe how geospatial data helps us better quantify the environmental materiality of © ANDRE DIB / WWF-BRAZIL droughts before we move on to estimate its link with the economy. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 51 Brasil Maior Plan and SPI 1-month 63,PRQWK 63,PRQWK SPI 3-month63,PRQWK 63,PRQWK SPI 6-month 63,PRQWK SPI 12-month 63,PRQWK 63,PRQWK 63,PRQWK Global Financial Crisis associated measures Brazilian Economic Crisis El Niño very strong  strong PRQWKVHDVRQDOO\DGMXVWHG )RUPDOHPSOR\PHQWJURZWK  moderate weak  0 weak  moderate strong Figure 27 63,PRQWK 63,PRQWK -2.0: Extremely dry 1.0: Moderately wet very strong 63,PRQWK 63,PRQWK Standardized Precipitation Indicator (SPI) over 1-, 3-, 6- and  -1.5: Very dry 1.5: Very wet 12-month horizons. La Niña -1.0: moderately dry 2.0: Extremely wet        Values depict the situation in December 2020. 0.0: Normal precipitation Each square represents 1 decimal degree (around 110km) 5HJLRQDOGLIIHUHQFHVLQIRUPDOHPSOR\PHQWJURZWKFRPSDUHGWRQDWLRQDODYHUDJH &HQWUDOZHVW 1RUWK 1RUWKHDVW 6RXWK 6RXWKHDVW 1DWLRQDODYHUDJH Figure 26 CASE STUDY SOCIAL CONSEQUENCES OF DROUGHTS Growth in Brazil’s regional formal employment and the El Niño–Southern Oscillation The World Bank classifies Brazil as highly vulnerable to Aside from the economic consequences of droughts, hydrological and meteorological disasters. Between 1900 for instance through the agricultural sector by harming Quarterly formal employment growth in each of the five Brazilian regions (left axis, colored lines) and the national average (left axis, black dashed line) are plotted against the backdrop of the Oceanic Niño Index (right axis, red and blue bars). and 2016, Brazil experienced economic losses of more than plant and livestock productivity, they also affect “[…] Visual inspection may lead to the conclusion that the El Niño was chiefly responsible for the 2015/16 downturn. However, $US 6.1 billion due to flash flood and riverine flood damages. public water supply, energy production, waterborne it is more likely that its role was aggravating the effects of the end of the commodity supercycle and the corruption In comparison, heavy and prolonged droughts affected the transportation, tourism, human health, biodiversity and scandal (“Lava Jato”). This warrants a more rigorous investigation. livelihoods of almost 80 million people and caused $US 111.2 natural ecosystems”, as described by a recent special billion in total damages. These events are expected to increase report of the United Nations.76 This is supported by findings both in frequency and severity in the future.75 in the literature, such as Rocha and Soares (2015), who MEASURING PRECIPITATION ANOMALIES At the same time, weather variation during and outside of wrote that early life health is determined by water scarcity and that droughts are “robustly correlated with higher The Standardized Precipitation Indicator (SPI) was introduced by McKee, Doesken, and Kleist El Niño and La Niña periods have profound effects on the infant mortality, lower birth weight, and shorter gestation (1993) to detect anomalies in precipitation patterns, such as unusually wet or dry conditions. Brazilian agricultural sector (Cirino et al. (2015)). Figure 26 periods.” Moreover, the same UN report also emphasizes The geospatial indicator used in this study is calculated by the Copernicus European Drought plots the Oceanic Niño Index against the quarterly formal the social consequences, since “droughts may affect men Observatory on a monthly frequency and with a spatial resolution of 1 decimal degree (around employment growth in each of the five regions of Brazil. The and women differently, and their impacts often amplify 110km, see Figure 27). SPIs are calculated over a specific accumulation period (e.g., 1, 3, black dashed line depicts growth on the national level, whose existing structural inequalities across social groups, ages 6 or 12 months) as deviations from the expected historical mean. Concretely, a high SPI-1 trajectory was shaped by major economic events. The coloured or other demographic categories.” Indeed, both floods and value in January indicates that it deviates strongly from historical rainfall values in January in lines isolate the region-specific developments in formal droughts hurt small rural farmers and poor urban residents, previous years. SPI-1 to SPI-3 (1 to 3 months), are short-term measures that detect reduced employment by calculating the difference between regional and who have limited means to respond to such disasters. soil moisture which could worsen crop health. SPI-3 to SPI-6 encompass entire growing national employment growth. A visual inspection of the figure Branco and Feres (2018) examined one of the possible or harvesting seasons where seasonal droughts can occur. SPI-12 represents an extended may suggest some relationship between the El Niño–Southern responses to weather shocks and found that droughts accumulation period and lower values could indicate reduced stream flow and water Oscillation (ENSO) and regional growth figures. One might be have an immediate, negative effect on rural household reservoirs. It is important to consider and compare different accumulation horizons, since a tempted to conclude that the El Niño was chiefly responsible income and thereby incentivize households to take up a shorter-term drought picked up by the SPI-3 indicator may in fact be part of a longer drought for the 2015/16 downturn. However, it is more likely that its secondary job, an effect they found to be stronger in poorer that is reflected by SPI-12.73 role was aggravating the effects of the end of the commodity municipalities in the Brazilian Northeast. supercycle and the corruption scandal (“Lava Jato”). Droughts are a complex phenomenon that no single indicator can fully explain. Accounting for local conditions, such as forest cover, irrigation systems or human settlements, and other hydrological and meteorological indicators is necessary to accurately characterize floods and droughts.74 In this study we focus solely on SPI and leave a more in-depth treatment for DATA In this case study we demonstrate how geospatial data Our dataset starts in mid-2004 and ends in 2021 with a future work. helps us gain a better understanding of this complex issue. monthly frequency for the 27 federal units. Environmental We focus on the regional economies of Brazil’s federal units data is obtained from the European Drought Observatory This rich geospatial data source by itself, however, cannot be directly used to answer rather than on the federal economy. This level of granularity and the formal employment indicator is retrieved from the economic questions, where researchers are used to deal with tabulated time series or cross- preserves the heterogeneity between states and allows Central Bank of Brazil (BCB) and the Brazilian Institute of sectional records. We therefore translate the SPI data from the geospatial format into a tabular us to make better use of the geospatial data. Alternatively, Geography and Statistics (IBGE). Land use and land cover format that aggregates observations onto the state level. This paves the way for statistical the data would have also allowed for the analysis to be data is obtained from Souza et al. (2020). models, which we employ to assess how unusually wet or dry weather conditions affect conducted on the municipal level. However, our main Brazil’s regional employment patterns. variable of interest, monthly formal employment growth, is only collected for the state level. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 53 RAINFALL ANOMALIES AND FORMAL EMPLOYMENT A CLOSER LOOK AT AGRICULTURE Formal employment growth does not respond to rainfall anomalies in a uniform way. Whether Employment figures in the agricultural sector are likely more dependent on rainfall patterns and how much regional economies react to periods of wet or dry weather depends on than are other sectors in the economy. For this reason, we narrow our analysis to the twelve the federal unit at hand, how long the anomaly persists and when it occurs. For example, federal units where farming is the most prominent land use and land cover (LULC), as shown does a short drought affect employment growth differently than an extended drought? in Figure 30.78 Given their higher share of farming LULC and the sensitivity of agricultural The response graphs, as described in Figure 28, answer this type of questions.77 After employment to rainfall patterns, we expect more pronounced effects of droughts. Figure 31 the accumulation periods of different durations ends, the solid lines show how the effect presents supporting evidence, by replicating the previous full-country analysis for these units. changes over different response horizons from the immediate, contemporaneous effect up The responses in Figure 31a and Figure 31b share similar characteristics. As before in Figure to a year after the event. The effects are estimated using local projection methods (LPM), 29, dry and wet weather have opposing effects. The long-term responses are strongest after which are a widely used methodology to estimate the effect of an event on an economy short-term anomalies. For example, in Figure 31b, after a short-term drought of one to three (Jordà, 2005). Events can for example be natural disasters (Dieppe, Celik, and Okuno, 2020; months, employment figures start to decline, and the effect becomes statistically significant Regelink et al., 2022) or new economic policies (Jordà, Schularick, and Taylor, 2020). A after four months. Interestingly, in Figure 31a, short-term wet seasons lead to increasing main benefit of LPMs is that they are lightweight and robust to misspecifications. The panel employment numbers that become significant after eight months. The growth may be due to version employed here also accounts for heterogeneities across federal units, such as harvesting seasons for soybeans, corn, rice and wheat, beginning around four to five months resilience, infrastructure, or economic structures. after planting and lasting another four to five months.79 For both types of anomalies, as the Figure 28 preceding accumulation periods exceeds three months, the response curves diminish. This is Illustration of response graphs. likely due to anticipatory and adjustment effects with labor moving to other sectors or regions. This study design estimates how the Accumulation period occurrence of abnormal weather during As mentioned earlier, droughts account for more than 92% of Brazil’s total losses due to the various accumulation horizons natural disasters. Figure 31c therefore examines the consequences of droughts during (SPI-1, -3, -6, -12) leads to responses -12 -6 -3 -1 t +1 +2 … in formal employment during the harvesting months specifically.80 Indeed, short-term droughts preceding harvesting seasons SPI-12 response period. Effects are estimated lead to immediate drops in employment. The effect size grows with longer response horizons SPI-6 Response period using local projection methods. and are highly significant.81 It is worth mentioning that by narrowing our focus on harvesting SPI-3 SPI-1 months we give more weight to cropland and less to pastureland agriculture. After short and wet periods, employment tends to shrink temporarily but recover in subsequent quarters (Figure 29). The opposite is true for short dry periods, with employment decreases up to twelve months later. Interestingly, the response graphs for both wet and dry weather flatten and diminish as accumulation periods become longer (six months or longer). This may be due to anticipatory and adjustments effects, as prolonged weather anomalies give time to the labor force to adapt, or agricultural and related industries adjust their employment needs. Predominant land use a) Very wet or extremely wet $EQRUPDOUDLQIDOOGXULQJSUHYLRXV  PRQWK PRQWKV PRQWKV PRQWKV Figure 29 Employment responds to unusual rainfall (all federal units)    The two panels (a) and (b) show the responses of formal employment to rainfall anomalies over various   accumulation horizons. After short  and wet periods, employment tends                             to shrink temporarily but recover in subsequent quarters. The opposite +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV is true for short dry periods, with (b) Very dry or extremely dry employment decreases up to 12 months later. The responses to wet and dry weather diminish with longer $EQRUPDOUDLQIDOOGXULQJSUHYLRXV PRQWK PRQWKV PRQWKV PRQWKV  accumulation periods, pointing towards possible anticipation and  adjustments effects.                                +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV The square markers locate the percentage change over increasing response horizons (horizontal axes). Dark markers indicate effect significance on a 5% level. The dashed lines Figure 30 – Predominant land use for each federal unit The map depicts the most common land use and land cover (LULC) classification for each (whiskers) demarcate the 68% (95%) confidence intervals. federal unit in 2020. Forests and farming are the two most common LULCs, compared to non-vegetated areas, water bodies and non-forest formations (Souza et al. (2020)). WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 55 Figure 31 – Employment (a) Very wet or extremely wet The effects identified are not particularly strong, with a cumulative employment decrease responses to unusual rainfall: predominant farm use LULC units. of -0.2% after 12 months. This is likely due to several reasons. First, despite Brazil being an agricultural power house and being among the largest producers for various crops, dairy $EQRUPDOUDLQIDOOGXULQJSUHYLRXV PRQWK PRQWKV PRQWKV PRQWKV The panels (a), (b) and (c) show the  products and meat, only 9% of the Brazilian work force is employed in agriculture. Second, responses of formal employment we only consider formal employment due to data limitations. We therefore almost certainly  to rainfall anomalies over various  accumulation horizons. In panels  underestimate the effect, as informal workers play an important role in Brazilian agriculture83 (a) and (b), unusual rainfall can  and are more vulnerable to adverse market conditions. Third, the estimated effect size relates take place during all months, while to a single drought only and could accumulate over multiple short-term droughts. Finally, since  in (c), only harvesting months  of soybeans, corn and rice are  rainfall anomalies mostly affect agricultural activities, we use land coverage and land use data relevant. The responses are in                             from satellite imagery to identify the twelve federal units where agriculture constitutes the line with findings in Figure 29 but largest coverage. Results show that indeed, if we only consider short-term droughts during +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV +RUL]RQ PRQWKV are more pronounced. Shorter droughts (1- to 3-months) have harvesting months in the twelve most farming reliant regions, formal employment drops by significant effects on subsequent (b) Very dry or extremely dry -0.4%, which is twice as large as when considering all regions. employment during harvesting seasons. As in Figure 29, the effect $EQRUPDOUDLQIDOOGXULQJSUHYLRXV of longer-term droughts may be OUTLOOK PRQWK PRQWKV PRQWKV PRQWKV alleviated due to anticipatory and  adjustment efforts.   A key contribution of this report is the flexible modeling approach which allows for quantifying  the materiality of biophysical or meteorological events for economic decision makers. As Gratcheva and Wang (2021) argue in their chapter of the recent flagship report “The  Changing Wealth of Nations 2021: Managing Assets for the Future”, a central challenge has     +RUL]RQ PRQWKV         +RUL]RQ PRQWKV        +RUL]RQ PRQWKV        +RUL]RQ PRQWKV   been translating an environmentally material quantity (e.g. droughts, forest area coverage, greenhouse gas emissions) into an economically material quantity (e.g. employment figures, inflation numbers, industrial output growth). Only then is it meaningful to discuss their (c) Very dry or extremely dry during harvest months (soybeans, corn and rice) financial materiality that influences the risk management and capital allocation decisions of financial market participants. The approach demonstrated here constitutes one possible $EQRUPDOUDLQIDOOGXULQJSUHYLRXV PRQWK PRQWKV PRQWKV PRQWKV way to establish the environmental-economic materiality link. The approach is not limited to  precipitation anomalies and employment figures but can be widely applied to study the rapidly  growing set of geospatial data sources.  This study could be expanded to integrate other relevant geospatial data such as weather  and environmental indicators or land cover transition data, to better identify drought    periods and discern which regions and municipalities are most sensitive rainfall anomalies.     +RUL]RQ PRQWKV         +RUL]RQ PRQWKV        +RUL]RQ PRQWKV        +RUL]RQ PRQWKV   Furthermore, additional economic indicators would be interesting to explore, such as regional inflation figures or real economic activity, to characterize the effect on the larger economy. A particularly interesting avenue of research would be to investigate spatial spillover effects between regions. For instance, one could estimate whether droughts trigger labor movement DISCUSSION towards neighboring states. Alternatively, one could empirically investigate the anecdotal evidence on how droughts in the 1970s lead to migration from the northeast to the southeast Sovereign ESG indicators are often available at the country level with an annual frequency. of Brazil, thereby contributing to the formation of the favelas. Additional data sources regarding Such a level of aggregation is not always ideal but justifiable, given the large data reporting informal labor would be particularly useful, given the large share of informal workers in disparities between countries and the goal of covering as many countries as possible. In Brazilian agriculture. comparison, geospatial data is not subject to these limitations and can therefore complement the existing sovereign ESG data landscape. This holds especially true for the environmental While the findings presented show materiality from an economic perspective, it is much more pillar, which is plagued by data limitations.82 Coverage of geospatial data is usually much larger difficult to ascertain financial materiality regarding Brazil’s sovereign bond market. Future work and more consistent but requires additional processing that can be considerable at times. could focus on linking economic materiality to financial materiality – and a corresponding effect What the researcher gains through the additional resolution – both temporally and spatially – on bond market pricing. As the concept of materiality is dynamic, the financial materiality are novel insights into a country that would be otherwise hidden on the aggregated level. impact could be short-lived or more prolonged and depend on the global economic patterns, such as the commodity cycle. In either case, understanding this link is important for policy This chapter illustrated how geospatial data could be used to model the effect of rainfall makers and financial market participants alike. anomalies, especially droughts, on Brazil’s federal economies. We found that dry periods are usually followed by lower employment figures in subsequent months. However, this only holds In summary, geospatial data is a valuable resource to mend the gaps in sovereign ESG true for shorter dry periods, up to six months. Longer periods have no strong effect as the indicators, especially on the environmental pillar. However, better data coverage alone is not local economy likely adjusts to the conditions. These findings hold on average for all Brazilian enough to improve sovereign ESG. Even though the results of the case study hinged upon the federal units and over all seasons. Given the nature of droughts and the labor market, these availability of sub-annual and subnational data, the higher resolution alone was not enough to effects would have been impossible to identify with annual country data only. understand how droughts affect Brazil’s regional economies. At the same time, the statistical model alone would not have generated any insights if we only had one data point for each year. It was the combination of geospatial data with an appropriate empirical model that enabled us to connect environmental with economic materiality. Establishing these materiality links will be crucial steps towards a better sovereign ESG framework. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 57 KEY PERFORMANCE In the case studies above, we have touched upon a range of observational datasets which could be used to create generic KPIs, and methods for differing scales. Outlined below is a INDICATORS – WHAT COULD list of potential metrics which are feasible to generate today. This is not to suggest this is new; such methods or metrics have previously been recommended in a range of standards and are already either directly or indirectly promoted by various commercial tools. However, whilst some indicators can be universal, as the field of geospatial ESG develops, we expect to see BE GENERATED NOW? more and more sector-specific metrics and models. POTENTIAL INDICATORS Here we outline a few examples of proposed high-level KPIs for defining ‘environmental’ geospatial ESG insights that could be widely applied today. Single metrics ideally should be Within this document, we have run through three case studies of what can currently be interlinked, weighted or modelled against other metrics for improved insight. Some must be achieved within the open data space. We have deliberately not attempted to present the tailored to the sector;86 others are more generic. results aligned to any standard, or existing initiative. However, of course, the results could be directly applied as metrics, or applied within existing footprint models, to track progress Overlap of asset with key location attributes, e.g.: aligned to various common frameworks, such as the SDGs, GRI,84 IFC Performance Standard 6, or the Natural Capital Protocol. • Biome, Ecoregion, Habitat, Land Cover classifications • Water Basin Much attention is now being given in various forums as to what the various high-level targets • Elevation are – what agreements nations, companies and financial institutions should adopt around the issue of nature and biodiversity. The question which sits alongside these endeavours • Assets urban/rural ratio – proximity to urban areas and linear infrastructure and movements is how does one measure performance against these frameworks and commitments? The ‘how’ arguably is more important than the ‘what’, as, without clear Overlaps of asset with key areas, i.e. PAs, KBAs, WHS, Intact Forest Landscapes, Ramsar means of measurement and tracking movement towards the proposed targets will remain Sites, Indigenous lands within the site, and buffers. unknowable. Data then is the underlying central component that enables application. Where data are not easily available, are inaccurate, or even if there is confusion surrounding what key • Weighting key areas by secondary variables, e.g.: methods or metrics are, we can expect to see disaffection. - Designation Within the emerging geospatial ESG space, there is now a race to develop the coherent - PA IUCN management category metrics, juggling the differing shortfalls (i.e. cost, accuracy, relevancy, legal rights, temporal - Species data, abundance, diversity, richness, and evenness of species consistency) of the data available against the need to develop clear metrics for various standards. The challenge will not be easy and will require improvements not only in the - Internet salience of site critical asset and observational data, but also in the machine learning and the sophistication - NGO / Conservation presence of sector-site-specific and user-case-specific products for financial institutions. To catalyse developments, there ultimately needs to be a collective push from a wide range of - Human population presence stakeholders to guide the rapid development of environmental indicators that are: - Temporal valves • Low cost, easy to produce • Weighting key areas by ‘intactness’ indicators, e.g.: • Accurate and reliable and scientifically robust - Assessment on the intactness of conservation area • Sensitive to change and allow the separation of impact to a specific asset - Assessment of land degradation • Comparable, across sites and scales - Site fragmentation • Applicable across a wide range of sectors, environments, and contexts. - The extent of linear infrastructure - The extent of commercial activity Several large-scale non-profit initiatives are now working to define useful environmental indicators, not necessarily considering application within the financial sector. These initiatives, - Assessment of human disturbance of conservation area such as Biodiversity Indicators Partnership (BIP) and IPBES core list of indicators, are now - Temporal valves grappling with the issue of identifying robust datasets which can directly or indirectly be applied to create key metrics for national performance to standards such as the CBD and SDGs – a topic mirrored in conversations surrounding the emerging TNFD. BIP, for example, is exploring a number of fully or partially geospatially derived indicators described below. Others such as efforts aligned to the GEO-BON initiative have developed lists of remote sensing indicators for tracking biodiversity 85 (Appendix 1). WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 59 Overlap of asset with forest, within the site and the buffer: Examples of physical KPIs Although not directly explored in this paper, where focus instead has been on ‘environmental’ • Forest Loss per km2 measures, geospatial ESG should be including climate change and physical risk inputs, many - Temporal values of which can be united with other environmental insights to improve overall insight and context • Primary Forest Loss per km2 Asset against extreme weather events, within site and buffer: - Temporal values • Historical Riverine, Coastal or Drought flood risk • Secondary Forest per km2 • Current dynamic weather data - Temporal values • Exposure of sector’s critical infrastructure, e.g. length asset of power lines to hurricanes • Forest fragmentation - Temporal values Asset against key climate change metrics within site and buffer: • Intact Forest Landscapes km2 • Temperature, precipitation, climate water deficit, wind patterns, reservoir levels, land use - Temporal values patterns for fire suppression, etc across differing scenarios • Mangrove forest km2(Trends) Of course, many of these potential metrics lack spatial or temporal resolution or fall foul of • Plantation / Managed Forest per km2 another limitation to meaningfully or sustainably be applied to drive robust geospatial ESG • Palm Oil per km2 insights. Significant effort in many cases would need to be applied to convert these datasets to a high frequency high spatial resolution consistent observational portfolio. Luckily, however, • Dry forest / Cloud Forest km2 with the emergence of new methods and technologies, it may be possible to leapfrog these issues and move to a new generation of environmental observational datasets. We explore Asset against key biodiversity proxies within the site and the buffer: these future developments in the next section. • Endangered species range per pixel or within area of interest (AOI) - Species data (i.e. total % of species range, abundance, richness, etc.) • Biome, Ecoregion, Habitat, Land Cover classifications FUTURE DEVELOPMENTS • Remote sensing products – i.e. Biological effects of irregular inundation, Above-ground biomass, Foliar N/P/K content, Net primary productivity, Gross primary productivity, Fraction of absorbed photosynthetically active radiation, Ecosystem fragmentation, Vegetation height, Specific leaf area, Carbon cycle (above-ground biomass) • Wider landscape ecoregions, habitats context - Connectivity Geospatial approaches can provide useful insights into the activities of companies and indeed help to differentiate assets, companies and portfolios on initial and ongoing environmental • Freshwater Biodiversity Exposure impact. With growing interest in the application, there are now actors working on building • Indices – i.e. Biodiversity Habitat Index, Species Habitat Index, Wetland Extent Trends Index, data exchanges and newer tools to support asset and supply chain data improvements Index of Coastal Eutrophication (ICEP) and Floating Plastic debris Density, Reef Fish Thermal for application within this space.87 Beyond the issues with asset and supply chain data, as Index, Species Protection Index, Wildlife Picture Index in tropical forest protected areas discussed throughout this paper, there is still much work to be done to produce an effective, consistent environmental observational data stream to enable and support robust geospatial Asset against key water risk metrics within site and buffer: ESG environmental insights, where it is vital to: • Water Stress, Water Depletion, Interannual Variability, Seasonal Variability, Groundwater 1. improve the temporal and spatial resolution of environmentally relevant geospatial datasets Table Decline, Untreated Connected Wastewater, Coastal Eutrophication Potential for application in geospatial ESG to aid the generation of up-to-date insights; 2. use improvements in technology and methods employed to break through critical Examples of sector-specific monitoring KPIs bottlenecks to ensure relevancy and coverage of critical topics; 3. improve commercial access to ‘environmental’ relevant data held by the IGOs, NGOs and • Methane Emissions frequency and density to specific areas, or assets academic institutions – either via open data standards or effective commercial licensing - Comparison to reported emission values and distribution solutions. • Marine Oil spill detection frequency and density to specific areas, or assets - Comparison to reported emission values Fortunately, various actors are already working to resolve these issues. In this section, we highlight some of the work being undertaken to provide solutions to the challenges faced. • Infra-red heat profile of site as a proxy for carbon dioxide emissions (estimate of the extent of cement factories power usage measured by heat generated) - Comparison to reported emission values of an asset • Ship AIS data - Shipping incidents – coral reef groundings, illegal fishing events, marine oil spills etc. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 61 BOX FIVE IMPACT OBSERVATORY – OBSERVATIONAL DATA IMPROVEMENTS – UPCYCLING WITH AI Authors: Steven Brumby, CEO/CTO & Co-Founder – Impact Observatory and John Barabino, Co-Founder & Head of Business – Impact Observatory There has been an explosion of research in the AI field of deep learning, resulting in open-source algorithms and automation software for rapid, large-scale image analysis. Adaptation of these techniques to satellite and aerial imagery has led to the creation of a number of space data analysis companies, including Orbital Insight, Descartes Labs, SpaceKnow and Impact Observatory, as well as major programmes to add machine learning capabilities within existing geospatial technology companies, e.g., Esri ArcGIS, Maxar GBDX, Planet PlanetScope, Airbus UP42, Google Earth Engine and Microsoft Planetary Computer. Open-source code repositories and coding communities enable sharing and reuse of geospatial algorithms and are democratizing access to knowledge previously available to only a few government agencies and technology companies. Applying on-demand geospatial data to foundational ESG challenges ESG applications supporting both private and public good organizations commonly require an understanding of land use and land cover (LULC) change processes across many years, and across seasons within a given year. Understanding patterns in LULC change can provide insights about resource exploitation, biodiversity habitat reduction, loss of ecosystem services and fluxes in natural storage of carbon. Impact Observatory has developed deep-learning algorithms that use 10m Sentinel-2 imagery to create dynamic, global LULC time series datasets to inform decision making and monitor impact.88 This global 10m LULC time series product provides over 100 times the spatial resolution of previous global open science products such as the Copernicus Figure 32 – Impact Observatory near-real-time, on-demand land use land cover (LULC) provides global 10m maps for planning, CGLS-LC100 100m resolution dataset or the NASA MODIS 500m dataset. Automation enables timely updating reporting and monitoring. These maps enable creation of dynamic science products that increase the spatial resolution and timeliness of important measures for climate, biodiversity and sustainable development. of the LULC map within the year and near-real-time monitoring, compared to traditional map products that are only updated after one to several years delay. Similar global 10m LULC products are planned by other teams, Data improvements on biodiversity e.g. the ESA WorldCover programme, and teams leveraging Google Earth Engine and Microsoft Planetary Computer. One of the most important observational variables to get right from an environmental perspective is accurately defining impact on biodiversity. Measuring biodiversity and then impact is extremely From land cover time series to dynamic science products challenging. Traditional field sampling, whilst excellent, is slow and resource intensive and as a result Building on this LULC monitoring capability, Impact Observatory partners with leading academic and impractical as a method to create accurate high frequency global insights. The alternative solution environmental NGO science teams with expertise in specific ecosystem services, such as carbon storage, commonly promoted is the reshuffling of the current global species datasets into new data products, biodiversity intactness and measuring the ‘footprint’ of human development. Impact Observatory automates and using various new statistical approaches or some other means. Such approaches are arguably unlikely scales these published science models to create automated global datasets as openly licensed science products. to break significant ground, as all these ‘new’ products rest upon the same limited data. Instead, if we are to radically improve insight on the extent and trends in species globally, entirely new data collection The human footprint dataset combines land use information with population pressure, built infrastructure approaches will be required. and accessibility data to map and assess human pressures across the globe at 1km resolution. These maps are critical for identifying remaining wilderness areas to aid in planning and management efforts. Impact Substantial global aggregates of species data already exist, such as GBIF,89 which provide vital data. Observatory worked with the human footprint academic team to reprocess the human footprint globally for However, these initiatives need to be supported by more regular and higher frequency field data if they 2017–2020 at 100m resolution. The results show high agreement with previous human footprint maps, with the are to provide strong geospatial ESG insights able to show subtle changes and trends in degradation ability to now run these on demand. to habitats over short time frames. Indeed, recognising the need for more data, under the Data4Nature initiative,90 corporates are encouraged to upload their species records created during impact assessments As with human footprint, Impact Observatory has also partnered with the UNEP World Conservation Monitoring into GBIF. Centre (WCMC) to operationalize the calculation of the biodiversity intactness index (BII) and above-ground biomass carbon change. The models used to create these science products use LULC time series, which can Fortunately, novel ground sampling methods are emerging that are capable of offering greater and now be automated and run on demand. Impact Observatory has partnered with WCMC to create global 100m greater insight into what species are actually present and potential insights into trends and degradation products that can be generated on demand and released annually as a public good. of ecosystems in much more real-time. For example, developments on bioacoustics – recording the ratio of natural vs. non-natural noise within a landscape and the structure and gaps within a soundscape – Impact Observatory’s industry partners fund the development and processing of this work, which provides combined with machine learning offers interesting potential of being able to better understand the high industry, finance and governments with near-real-time, on-demand, science-based insights, and allows Impact level health of a landscape with much lower sampling requirements. Initiatives such as Wildlife Insights91 Observatory to release public good, global LULC and derived science time series products on an annual basis. and the eBioAtlas Freshwater eDNA initiative92 offer a vision of the future where multiple forms of sampling data (e,g, eDNA, camera trap, audio) are combined together to help refine remote sensing products. Here we look at one of these technologies: eDNA. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 63 BOX SIX A substantial body of research literature now demonstrates that DNA-based methods can match or outperform NATUREMETRICS – conventional survey methods for many species and groups, and often bring advantages in terms of cost and survey effort, increased detection sensitivity and increased taxonomic resolution.97 For example, NatureMetrics has shown IMPROVEMENTS WITH DNA-BASED APPROACHES that in the field you can survey more species in just 10% of the field time when compared to traditional approaches. Compared to other high-tech biodiversity solutions, eDNA has the benefit of a relatively low-tech approach to field data Author: Dr Cath Tayleur, Head of Nature Positive Supply Chains – NatureMetrics collection, making it accessible to everyone including local communities and other stakeholders. DNA-based monitoring offers the ability to make surveys easier to conduct and replicate. Species identification is Challenges with traditional field sampling approaches automated in the lab, removing the subjectivity of field-based species detection. DNA samples can also be stored for future analysis; such repositories could be used in future to conduct independent audits of biodiversity claims. Traditional sampling suffers from its inability to scale, both over space and time, but also over species, as an astonishing 86% of land species and 91% of ocean species remain undiscovered.93 As we move beyond biodiversity being included in decisions simply for its intrinsic value, to its importance in underpinning economies and global health, we need ways to encapsulate the full complexity of life on earth. Biodiversity monitoring needs to capture those species, communities and their characteristics that are the real engine driving ecosystems. For example, without the multitude of bacteria and fungi that make up the soil microbiome, global agriculture would grind to a halt, yet our knowledge of life below earth is only starting to capture the complexity of this system.94 One of the key barriers has been the need to have trained experts in the field able to identify target species by sight, sign or sound. New approaches to biodiversity monitoring are required that democratize data collection, allowing a wider range of stakeholders to participate using standardized, simple and effective protocols. How DNA-based monitoring can overcome these challenges DNA-based approaches include environmental DNA (eDNA), the traces that species leave behind in the environment, and in other cases the sampling of the organisms themselves, such as with bulk samples of insects. A bit like crime scene forensics, the individual fingerprints detected through samples of soil, sediment, water and even air are then matched to reference libraries that contain the sequences of thousands of species (Figure 33), giving us a snapshot of whole biological communities. This process of DNA metabarcoding generates data at a scale that has never previously been feasible, allowing a more comprehensive overview of biodiversity, including the small and diverse organisms (e.g., insects, soil fauna, plankton, fungi, bacteria) that are often closely linked to the ecological functions of particular habitats95 and ecosystem services.96 Environmental DNA (eDNA) sampling Figure 34 – Using a filter to capture the DNA of a wide range of species that are present in and around water. Future directions for the DNA-based biodiversity data revolution The scalability of DNA-based monitoring means that we can gather more data on more species, and ultimately make more informed, evidence-based decisions around biodiversity. This has implications for a wide range of applications – from systematic conservation planning and evaluation of conservation outcomes to due diligence and environmental impact assessment. KPIs linked to species and/or ecosystem health could inform the sustainability linked products of the future. There is a huge opportunity to apply ‘big data’ approaches to DNA-based datasets in order to better understand how ecosystems function and respond to change. With large-scale monitoring, such as eBioAtlas, we can build algorithms to identify the signatures of healthy and resilient ecosystems and use these to inform KPIs. DNA- based approaches can be used to help validate the link between pressures and outcomes, ground-truthing the predictions of model-based assessments. We can set and measure progress towards meaningful restoration and ‘net positive’ targets, maintaining and improving the underlying functions of ecosystems. Finally, we can link field data to powerful earth observation data, allowing us to better see biodiversity from space, achieving even greater scale and allowing us to track changes in near-real-time. Figure 33 – The process by which eDNA enters the environment, is collected, processed, sequenced and turned into biodiversity data. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 65 FINAL REFLECTIONS The world of geospatial ESG is at an exciting point in time, with its potential only just beginning to be recognized and explored in the mainstream. This growth and novelty is reflected in many of the open and commercial geospatial ESG platforms currently emerging, where arguably many underutilize the full range of technical approaches available. In the diagram on the following page (Figure 35), we’ve simplistically classified the various key components of a geospatial ESG platform. If you look at any given tool, many are only getting started using the basic elements, such as a direct comparison to asset locations using multiple layers. Although of course there are exceptions – tools which use far more sophisticated sector and site weightings, observation data refinement and various forms of machine learning. Throughout this document we have shown the simplest approaches and introduced some of the more complex elements. It is exciting to consider that many actors are now moving beyond this to develop more robust solutions in this space. Combined with improvements in remote sensing and ground data collection, we expect in the near term improved geospatial ESG outputs with far greater accuracy and insight into the ‘environmental performance’ of assets, commercial actors and even national states. While there is much to expect from the emerging field, there are external limitations which may undermine progress such as access to asset and supply chain data. From the environmental data perspective, perhaps the key challenge is around the diversity of data sources. Where data providers will need to integrate, often continuously, a huge range of differing data sources held by a diverse range of NGOs, intergovernmental agencies, academia, the private sector and commercial data providers. Since no single actor will be capable of providing all these data sources, a shift will be required in how data is aggregated, suggesting that new approaches to data sharing will be required. It seems probable that this might be achieved via secure interconnected data marketplaces, perhaps using a tested and well-developed open standard to aid adoption; however, the design and terms of such an exchange is beyond the remit of this paper. ADRIANO GAMBARINI © WWF-BRAZIL / WWF-US / ADRIANO GAMBARINI WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 67 Figure 35 – Diagram illustrating various key components available to geospatial ESG platforms, many of which are expected to come into mainstream development shortly. REFERENCES AMIS (2012) AMIS Crop Calendar, Agricultural Market Information Gratcheva, E. et al. (2021) A New Dawn - Rethinking Sovereign ESG. System, Food and Agriculture Organization of the United Nations, EFI Insight-Finance. Washington, DC and New York, NY: World Bank Rome, Italy. www.amis-outlook.org. pp. 1–8. and J.P. Morgan. ASSET DATA OBSERVATION DATA DATA PROCESSING Arslanalp, S. et al. (2020) Benchmark-Driven Investments in Gratcheva, E., Emery, T. and Wang, D. (2021) Demystifying Sovereign Emerging Market Bond Markets: Taking Stock. IMF Working Paper ESG. EFI Insight-Finance. Washington, D.C.: World Bank Group. Asset location Single layer Direct comparison WP/20/192. Washington, D.C.: International Monetary Fund. Gratcheva, E. and Wang, D. (2021) Natural allies: Wealth and One vector layer or raster layer Asset overlaid by one or multiple Battistella, L., Mandrici, A., Delli G., Bertzky, B., Bastin, Dubois, G. sovereign ESG, in The Changing Wealth of Nations 2021: Managing included in analysis. observational data layers. (2018) Map of Protection Levels for the Terrestrial Ecoregions of the Assets for the Future. Washington, D.C.: World Bank Group. World as of April 2018. © European Union, 2018 Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, + Sector and site specific Branco, D. and Feres, J. (2018) Weather Shocks and Labor S. A., Tyukavina, A., Thau, D., Stehman, S. V., Goetz, S. J., Loveland, weightings Allocation: Evidence from Northeastern Brazil, 2018 Conference, + Sector specific attributes + Multiple layers T. R., Kommareddy, A., Egorov, A., Chini, L., Justice, C. O. and July 28-August 2, 2018, Vancouver, British Columbia, 2018 Townshend, J. R. G. (2013) High-resolution global maps of 21st- Impact adjusted to sector and site E.G. power plant, real estate, Two or more vector layers or raster layers Conference, July 28-August 2, 2018, Vancouver, British Columbia century forest cover change. Science 342 (15 November): 850–53. variables farm - cotton. included in analysis. International Association of Agricultural Economists. Data available from: https://glad.earthengine.app/view/global-forest- change. + Observational inferences Calice, P. and Miguel, F. (2021) Climate-Related and Environmental Risks for the Banking Sector in Latin America and the Caribbean: Hansen, A. (2020) SCI (2020 version) - Continent. figshare. Dataset. Refining, backfilling observational data A Preliminary Assessment, World Bank, Washington, DC. World https://doi.org/10.6084/m9.figshare.11608182.v2 The data is updated from other variables. Bank. p. 43. from the data provided in Hansen, A. et al. Global humid tropics + Site specific attributes + Dynamic data forest structural condition and forest structural integrity maps. Cirino, P. H., Féres, J. G., Braga, M. J., and Reis. E. (2015) Scientific Data 6, 232, doi:10.1038/s41597-019-0214-3 (2019). E.G. hydro power plant reservoir size, Near real-time feed of data, weather data. + Interdependence Assessing the Impacts of ENSO-related Weather Effects on power production Mw. the Brazilian Agriculture, Procedia Economics and Finance, 24, Hansen, A. (2020) FSII (2020 version) - Continent. figshare. Dataset. The site specific impacts considering 146–55. the interdependencies of natural assets, https://doi.org/10.6084/m9.figshare.11604588.v1 The data is updated e.g. forest loss impacts on wider local from the data provided in Hansen, A. et al. Global humid tropics water security. Cunha, A.P.M.A. et al. (2019) Extreme drought events over Brazil forest structural condition and forest structural integrity maps. from 2011 to 2019, Atmosphere, 10(11), p. 642. doi:10.3390/ Scientific Data 6, 232, doi:10.1038/s41597-019-0214-3 (2019). + Additional external data + Sector specific atmos10110642. monitoring data + Near real time adjustment Henley, A., Arabsheibani, G. R., and Carneiro. F. G. (2009) On E.G. web scraped data. Dang, H.-A.H. et al. (2021) Statistical Performance Indicators and Defining and Measuring the Informal Sector: Evidence from Brazil, i.e. methane detection, marine oil spill Results updated frequently and capable Index: A New Tool to Measure Country Statistical Capacity. The World Development, 37, 992–1003. detection, night time flaring, for oil and gas of adjusting to near real time data feeds, World Bank (Policy Research Working Papers). doi:10.1596/1813- assets. e.g. oil spill. 9450-9570. Hilton, S. and Lee, J.M.J. (2021) Assessing Portfolio Impacts - Tools to Measure Biodiversity and SDG Footprints of Financial Portfolios. Dieppe, A., Celik, S.K. and Okou, C. (2020) Implications of Major Gland, Switzerland: WWF. Adverse Events on Productivity. Policy Research Working Papers + Supply chain asset data + Historic and future data AI No. 9411. Washington, D.C.: World Bank Group, p. 44. IPBES (2019) Díaz, S., Settele, J., Brondízio, E. S., Ngo, H. T., Guèze, The asset data of all major or significant E.G. past temperature averages, extreme M., Agard, J., Arneth, A., Balvanera, P., Brauman, K. A., Butchart, S. suppliers and their suppliers. weather events. Facciolo, G., Franchis C. De. and Meinhardt-Llopis, E. (2017) H. M., Chan, K. M. A., Garibaldi, L. A., Ichii, K., Liu, J., Subramanian, + Machine rationalization ‘Automatic 3D reconstruction from multi-date satellite images,’ S. M., Midgley, G. F., Miloslavich, P., Molnár, Z., Obura, D., Pfaff, A., 2017 IEEE Conference on Computer Vision and Pattern Recognition Polasky, S., Purvis, A., Razzaque, J., Reyers, B., Roy Chowdhury, Analysis is adjusted to the best regional data and regional models based on Workshops (CVPRW) 2017, pp. 1542-1551, doi: 10.1109/ R., Shin, Y. J., Visseren-Hamakers, I. J., Willis, K. J. and Zayas, C. N. dynamic machine rationalisation of the CVPRW.2017.198. (eds.), Summary for Policymakers of the Global Assessment Report options present. on Biodiversity and Ecosystem Services of the Intergovernmental + Other data + Other data Fernández, P., et al. (2021) Comparing environmental DNA Science-Policy Platform on Biodiversity and Ecosystem Services. metabarcoding and underwater visual census to monitor tropical Bonn, Germany. 56 pages. Traditional ESG data points, economic, E.G. social, economic, governance data + Machine learning reef fishes. Environmental DNA 142-156 https://doi.org/10.1002/ social data points, ground data etc. points, ground data, etc. edn3.140 IPBES (2019) Brondzio, E. S., Settele, J., D.az, S., Ngo, H. T. (eds), Throughout any of the various data sourcing, data processing or results, Global Assessment Report of the Intergovernmental Science-Policy machine learning is applied to iteratively Fierer, N. (2017) Embracing the unknown: disentangling the Platform on Biodiversity and Ecosystem Services. IPBES secretariat, improve outputs. complexities of the soil microbiome. Nature Reviews Microbiology Bonn, Germany. 15, 579–590. https://doi.org/10.1038/nrmicro.2017.87 Jarić, I., Quétier, F., and Meinard, Y. (2020) Procrustean beds and Global Reporting Initiative (GRI) (2016). Consolidated Set of empty boxes: On the magic of creating environmental data. Biological GRI Sustainability Reporting Standards 2016. Amsterdam, The Conservation 2020, 237, https://doi.org/10.1016/j.biocon.2019.07.006. Netherlands: GRI. Jordà, Ò. (2005) Estimation and inference of impulse responses by local projections, American Economic Review, 95(1), p. 22. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 69 Jordà, Ò., Schularick, M., and Taylor. A. M. (2020): The effects of quasi-random monetary experiments, Journal of Monetary Skidmore, et al. (2015) Environmental science: Agree on biodiversity metrics to track from space, Nature, vol. 523, no. 7561, pp. 403-405. APPENDIX ENDNOTES Economics, 112, 22–40. https://doi.org/10.1038/523403a 1. Some commercial outfits, such as Verisk Maplecroft, have for over a decade Appendix 1 – Remote sensing product suggestions as priority focused on integrating geospatial data into ESG. Arguably now interest has McKee, T.B., Doesken, N.J. and Kleist, J. (1993) ‘The relationship Skidmore, A.K., Coops, N.C., Neinavaz, E. et al. (2021) Priority list biodiversity metrics (from Skidmore et al. 2021): become more mainstream, from financial institutions, within major policy, and of drought frequency and duration to time scale’, in Proceedings of biodiversity metrics to observe from space. Nature Ecology & within IGOs and NGOs. For example, new publications are emerging on the subject from financial institutions, such as Swiss Re (2021). of the Eighth Conference on Applied Climatology. Anaheim, Evolution 5, 896–906. • Biological effects of fire disturbance (direction, duration, 2. To avoid confusion, here we use the term ‘geospatial ESG’, defined as ‘the use California: American Meteorological Society, pp. 179–184. of geospatial data to generate ESG relevant insights into a specific commercial Souza, C.M. et al. (2020) Reconstructing three decades of land use abruptness, magnitude, extent and frequency) asset, company, portfolio or geographic area’, rather than other terms such as Mora, C., Tittensor, D. P., Adl, S., Simpson, A. G. B. and Worm, and land cover changes in Brazilian biomes with Landsat archive • Biological effects of irregular inundation ‘spatial economics’ or ‘spatial finance’, which are commonly used to refer to any application of geospatial data within finance, e.g. remote sensing derived crop B. (2011) How many species are there on earth and in the ocean? and earth engine, Remote Sensing, 12(17), p. 2735. doi:10.3390/ yields for commodity trading insights. rs12172735. • Leaf Area Index PLOS Biology 9(8): e1001127. https://doi.org/10.1371/journal. 3. Within this document, due to our expertise, focus is placed on the ‘E’ pillar in ESG pbio.1001127 • Land cover (vegetation type) 4. Such as Orbital EOS Stephenson, P.J. and Carbone, G. (2021) Guidelines for Planning and 5. Such as EarthKnowledge Newbold, T., Hudson, L., Arnell, A., Contu, S., et al. (2016) Monitoring Corporate Biodiversity Performance. Gland, Switzerland: • Ice cover habitat 6. Such as GHGSAT, MethaneSAT ‘Dataset: Global map of the biodiversity intactness index.’ In IUCN. • Above-ground biomass 7. Such as Kayrros Newbold, T., et al., has land use pushed territorial biodiversity 8. Such as Global Forest Watch Pro and Trase beyond the planetary boundary? A Global Assessment. Science Swiss Re Institute (2021) Remote Sensing Innovation: Progressing • Foliar N/P/K content 9. Taskforce on Climate related Financial Disclosure 353 (2016): 288-89. http://dx.doi.org/10.5519/0009936. Sustainability Goals and Expanding Insurability. Zurich. • Net primary productivity 10. Taskforce on Nature-related Financial Disclosures 11. Demonstrated by growing attention to the topic in wider forums and bodies such O’Connor, B. et al. (2015) Earth observation as a tool for tracking Teulings, C.N. and Zubanov, N. (2014) Is economic recovery a • Gross primary productivity as the World Economic Forum progress towards the Aichi Biodiversity Targets. Remote Sens. myth? Robust estimation of impulse responses, Journal of Applied • Fraction of absorbed photosynthetically active radiation 12. Stephenson and Carbone (2021) Ecol. Conserv. 1, 19–28 Econometrics, 29(3), pp. 497–514. doi:10.1002/jae.2333. 13. UNEP-WCMC (2017) • Ecosystem fragmentation 14. Stephenson and Carbone (2021) Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Theobald, D. M., Kennedy, C., Chen, B., Oakleaf, J., Baruch-Mordo, • Ecosystem structural variance 15. World Bank; WWF (2020) Powell, G. V. N., Underwood, E. C., D’Amico, J. A., Itoua, I., Strand, S., and Kiesecker, J. (2020) Earth transformed: Detailed mapping of 16. It’s important to note that sovereign level insights would not exclusively draw from asset level results, but would derive their own sub-national and national H. E., Morrison, J. C., Loucks, C. J., Allnutt, T. F., Ricketts, T. H., global human modification from 1990 to 2017, Earth System Science • Urban habitat environmental metrics from the same observational datasets used with asset level Kura, Y., Lamoreux, J. F., Wettengel, W. W., Hedao, P., Kassem, K. Data, 12, 1953–1972, https://doi.org/10.5194/essd-12-1953-2020. • Vegetation height assessment, and potentially, summed relevant asset level insights. R. (2001) Terrestrial ecoregions of the world: a new map of life on 17. See here for more information on STAR Earth. Bioscience 51(11):933-938. Tian, J., Zhu, X., Wu, J., Shen, M. and Chen, J. (2020) Coarse- • Plant area index profile (canopy cover) 18. For example, NASA Global Ecosystem Dynamics Investigation Lidar and the German Aerospace Centre’s high-resolution and wide-spectrum satellite resolution satellite images overestimate urbanization effects on • Habitat structure programme EnMAP Regelink, M., Nie, O., Mare, D. S., Wang, D., and Bavandi. A. vegetation spring phenology. Remote Sensing. 2020, 12, 117. https:// 19. Hilton and Lee (2021) (2022, forthcoming) Banking Sector Risks in the Aftermath of doi.org/10.3390/rs12010117 • Fraction of vegetation cover 20. Within this document we focus on defining ‘environmental’ impact related to Climate-Related Natural Disaster, Policy Research Working Paper • Specific leaf area biodiversity. We do not attempt to discuss these impacts in terms of material, Series, Washington, D.C.: World Bank. UNEP-WCMC (2017) Biodiversity Indicators for Extractive Companies: reputational, regulatory or physical risks. Whilst geospatial approaches can provide robust insights into physical risks such as coastal flooding risk, extreme An Assessment of Needs, Current Practices and Potential Indicator • Chlorophyll content and flux weather risk, real-time weather risk, etc., we do not include them directly as an Rocha, R. and Soares, R. R. (2015) Water scarcity and birth Models. Cambridge, UK, 39pp. area of focus within this document. • Land surface peak (maximum of season) outcomes in the Brazilian semiarid, Journal of Development 21. It should be apparent that looking at values in isolation from one another is not a Vihervaara, P. et al. (2017) How essential biodiversity variables and • Land surface green-up (start of season) robust approach: ecosystems are heavily connected, and subsequently observed Economics, 112, 72–91. values will cascade and impact upon another. However, for a rapid high-level remote sensing can help national biodiversity monitoring. Global screening, the approaches outlined here should suffice and provide a useful • Land surface senescence (end of season) Ruesch, A. and Gibbs, H.K. (2008) New IPCC Tier-1 Global Ecology and Conservation. 10, 43–59. starting point in the concept, and they can of course be further refined. biomass carbon map for the year 2000. Available online from the • Carbon cycle (above-ground biomass) 22. Within this paper, for simplicity, we do not integrate, or discuss climate change data. However, climate change is heavily intertwined with biodiversity loss and Carbon Dioxide Information Analysis Center [http://cdiac.ornl.gov], United Nations Office for Disaster Risk Reduction (2021) GAR Special • Peak season (maximum of season) should be considered alongside environmental metrics. Oak Ridge, Tennessee: Oak Ridge National Laboratory. Report on Drought 2021. Geneva. 23. Here we assume the current data reality where ground data from power plants • Green-up (start of season) themselves, or any asset, isn’t available. As such we have to apply other data Salafsky, N., Salzer, D., Stattersfield, A. J., Hilton-Taylor, C., World Bank; WWF (2020) Spatial Finance: Challenges and approaches to independently gain as much insight as possible. Of course, if highly • Senescence (end of season) granular site level data on their operations were obtainable, these methods would Neugarten, R., et al. (2008) A standard lexicon for biodiversity Opportunities in a Changing World. Equitable Growth, Finance and be unnecessary. conservation: Unified classifications of threats and actions. Institutions Insight, Washington, DC.: World Bank, © World Bank. • Leaf dry matter content 24. The term ‘open’ is used as a generality here; the data layers reviewed have Conservation Biology 22:897-911. doi: 10.1111/j.1523- https://openknowledge.worldbank.org/handle/10986/34894 differing licensing constraints, ranging from fully open for any application, • Ecosystem soil moisture open for non-commercial use, to restricted, requiring written permission for 1739.2008.00937.x use. However, these datasets and their licensing variability could be considered World Economic Forum (2020) Nature Risk Rising: Why the Crisis • Functional diversity indicative of the ‘public’ or ‘open’ environmental geospatial space. Schadewell, Y. and Adams C.I.M, (2021) Forensics meets ecology Engulfing Nature Matters for Business and the Economy, Geneva, 25. The term ‘open’ is used as a generality here; the data layers reviewed have • Species abundance – environmental DNA offers new capabilities for marine ecosystem Switzerland, 36p, differing licensing constraints, ranging from fully open for any application, open for non-commercial use, to restricted, requiring written permission for and fisheries research. Frontiers in Marine Science ., 27 April 2021 • Relative species abundance use. However, these datasets and their licensing variability could be considered | https://doi.org/10.3389/fmars.2021.668822 World Meteorological Organization (2012) Standardized Precipitation indicative of the ‘public’ or ‘open’ environmental geospatial space. • Population density Index User Guide. WMO-No. 1090. Geneva. 26. 76% (105) of the 137 data layers publicly listed, data layers which were unsuitable World Meteorological Organization. (2012): Standardized Precipitation were excluded e.g., ‘demo layers’ – data reviewed as at 11/03/2021 Seymour, M., Edwards, F.K., Cosby, B.J. et al. (2021) Index User Guide,. 27. Jarić, Quétier and Meinard (2020) Environmental DNA provides higher resolution assessment of 28. Tian et al. (2020) riverine biodiversity and ecosystem function via spatio-temporal 29. Battistella et al. (2018) nestedness and turnover partitioning. Communications Biology 4, WWF (2020) Almond, R.E.A., Grooten M. and Petersen, T. (Eds), Living 30. O’Connor et al. (2015) 512. https://doi.org/10.1038/s42003-021-02031-2 Planet Report 2020 - Bending the Curve of Biodiversity Loss. Gland, 31. Skidmore, et al. (2015) Switzerland. WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT 71 32. Biodiversity can be defined loosely as the variety of life found within an area. 60. IUCN Category VI – Protected areas conserve ecosystems and habitats together Throughout this document, we deliberately do not strictly define the term, reflecting with associated cultural values and traditional natural resource management the reality that within the ESG space, there are a wide range of geospatial data systems. They are generally large, with most of the area in a natural condition, products that present, or could potentially provide, insights relevant to biodiversity where a proportion is under sustainable natural resource management and and as such are often communicated around or within the ‘biodiversity’ label. A wide where low-level non-industrial use of natural resources compatible with nature range of indirect proxies such as ‘freshwater’ or ‘legal area delineations’ are used, conservation is seen as one of the main aims of the area (IUCN, 2013). https:// which without actually being a direct measure of biodiversity, may still arguably portals.iucn.org/library/node/30018 provide some useful insight within ESG applications. 61. For example, see the Digital Observatory for Protected Areas However, for reference, IPBES provides an accurate definition: ‘Biodiversity is 62. Provided as an example; more complex assessments are possible, for example it the variability among living organisms from all sources, including terrestrial, is possible to consider the range of a species and the dependence of that species marine and other aquatic ecosystems, and the ecological complexes of which to a specific site – e.g. does it constitute 100% of its range or 1%, and against other they are a part. This includes variation in genetic, phenotypic, phylogenetic and ecological values such as known habitat types and elevation. functional attributes, as well as changes in abundance and distribution over time 63. Beyond geospatial approaches, some actors may have gained insight into supply and space within and among species, biological communities and ecosystems.’ chain footprints using Environmentally Extended Input-Output models and other (Intergovernmental Panel on Biodiversity Ecosystem Service, 2019). methods. 33. Here we define ‘near-real-time’ as updates every month within a month, i.e. 64. Facciolo, De Franchis & Meinhardt-Llopis (2017) January’s results before the end of February. 65. ESG investing refers to an investment process where ESG factors are either used 34. The term ‘open’ is used as a generality here, the data layers reviewed have differing as inputs into financial decision-making (e.g. to identify risk factors or guide licensing constraints, ranging from fully open for any application, open for non- investment allocations) or to measure the outputs of the investment (e.g. emission commercial use, to restricted, requiring written permission for use. However, these reductions or sustainable forest management). We refer the reader to (Gratcheva datasets and their licensing variability could be considered indicative of the ‘public’ et al. (2021)), who discuss the underlying dual materiality concept. or ‘open’ environmental geospatial space. 66. Overall institutional assets are estimated at about US$100 trillion. Current 35. See Appendix 1. estimates of ESG-themed strategies amount to about US$ 40 trillion (Gratcheva et 36. Skidmore, Coops et al. (2021) al. (2021)). 37. Vihervaara et al. (2017) 67. A benchmark index describes the performance of a group of financial securities, such as stocks or bonds. The benchmark’s constituents are weighted to best 38. Such as Ecometrica reflect the group’s overall performance. In the stock market, the S&P500 tracks 39. Such as Earth Knowledge the 500 largest US listed companies, which are weighted based on their market 40. IPBES (2019b) capitalization. ESG integration into an index would adjust the constituents’ weights to also reflect their ESG practices. 41. IPBES (2019a) 68. lanalp et al. (2020)) 42. WWF (2020) 69. (Gratcheva, Emery, and Wang (2021)), (Gratcheva et al. (2021)) 43. World Economic Forum (2020) 70. In addition, ESG providers tend to construct Social and Governance indicators 44. Salafsky et al. (2008) based on similar sources. 45. These types of assessments are viable and scalable. To guide its work and 71. (Dang et al. (2021)) insight, the WWF Conservation Intelligence team has since 2016 run global scale assessments on millions of mining, oil and gas, power, infrastructure and other 72. For further discussion about the difference between environmental and economic assets against 50+ observational datasets every quarter. materiality, we refer to (Gratcheva and Wang (2021)). 46. All the mines reported here are to illustrate the geospatial approach and the 73. (World Meteorological Organization (2012)) complications which may arise. They serve as examples only; the authors are 74. For example, the Palmer Drought Severity Index (PDSI), Integrated Drought Index making no delineation or statement as to the environmental or ESG performance of (IDI), the Next Generation Drought Risk Index (NGDI), the Weighted Anomaly these assets. Standardized Precipitation (WASP). 47. Here we use a simple, generalised area value of 1km² for each mine to illustrate 75. Climate Risk Profile: Brazil (2021): The World Bank Group the method and results in its simplest terms. Using a slightly more sophisticated 76. (United Nations Office for Disaster Risk Reduction (2021)) approach, it is possible to consider assets exactly by their concession areas or exact measured footprint, and/or by an estimated footprint. For example, Theobald et 77. Impulse response graphs are commonly used in economics to assess the effect of al. 2020, using a sampling approach, estimated mines’ footprints into four type of a sudden change (“shock”) in an input variable (e.g. interest rate changes) onto an mines by commodity: 1) coal; 2) hard-rock (bauxite, cobalt, copper, gold, iron ore, output variable of interest (e.g. employment) over an extended time horizon. lead, manganese, molybdenum, nickel, phosphate, platinum, silver, tin, U3 O8, and 78. While the LULC changes over time, units and their most prominent LULC are zinc); 3) diamonds; and 4) other (antimony, chromite, graphite, ilmenite, lanthanides, constant throughout the sample. For farming, which includes pastureland and lithium, niobium, palladium, tantalum, and tungsten). They used an estimated mean cropland (annual, perennial and semi-perennial) as well as combinations thereof area of 1) 12.95km2 for coal; 2) 8.54km2 for hard-rock; 3) 5.21km2 for diamonds; (Souza et al. (2020)), we identify the following twelve units (descending order): and 4) 3.40km2 for others. https://essd.copernicus.org/preprints/essd-2019-252/ Sergipe, Alagoas, São Paulo, Espírito Santo, Paraná, Goiás, Minas Gerais, Rio de essd-2019-252.pdf Janeiro, Mato Grosso do Sul, Santa Catarina, Rio Grande do Sul, Distrito Federal. 48. The IUCN Red List, World Database on Protected Areas and Key Biodiversity Areas 79. (AMIS (2012)) global datasets are available for commercial application via the IBAT Platform. 80. Harvesting seasons for soybeans, corn (first crop) and rice begin in January and 49. Olson et al. (2001) end in May and June (AMIS (2012)). 50. Newbold et al. (2016) 81. We employ the future bias correction from (Teulings and Zubanov (2014)). Without 51. Ruesch & Gibbs (2008) it, we would assume that no droughts happen during the response period, which is unlikely. 52. Hansen et al. (2013) 82. (Gratcheva, Emery, and Wang (2021)) 53. Hansen (2020): SCI (2020 version) 83. (Henley, Arabsheibani, and Carneiro (2009)) 54. Hansen (2020): FSII (2020 version) 84. Global Reporting Initiative (GRI) (2016) 55. Protected Areas (PAs) often exist in the same spatial area with multiple designations covering one area. For example, a National Park boundary may also be designed as 85. Skidmore et al. (2021) a World Heritage Site. 86. A tension arises here: the more sector specific the metric, potentially the more 56. Here we use a simple generalised area value of 1 Km² for each mine to illustrate informative; however, also potentially the harder it will be to scale the metric and the method and results in its simplest terms. Using a slightly more sophisticated to integrate it with other more generic metrics at the portfolio level. approach it is possible to consider assets exactly by their concession areas or 87. Such as the Green Digital Finance Alliance (GDFA) – Open-Source Biodiversity Data footprint, and or by an estimated footprint. See Theobald et al. 2020. Platform Initiative 57. Values may exceed 100% overlap, as multiple protected area designations may 88. Karra et al IGARRS 2021 occupy the same spatial extent. More complex measurements are easily achievable, 89. GBIF such as depth of assets within the conservation area, fragmentation, etc. 90. Data4Nature initiative 58. Results reported here do not in any way report an actual factual position of WWF, or any other of the institutions or authors, on the environmental impact or ESG 91. Wildlife Insights performance of these mines or companies, nor their rankings. Results are not 92. eBioAtlas consistent; they show examples of mines with medium to high scores, excluding 93. Mora et al. (2011) records to randomise results. 94. Fierer (2017) 59. IUCN Category V – A protected area where the interaction of people and nature over time has produced an area of distinct character with significant ecological, 95. Seymour et al. (2021) biological, cultural and scenic value, and where safeguarding the integrity of this 96. Schadewell & Adams (2021) interaction is vital to protecting and sustaining the area and its associated nature 97. Fernández et al. (2021) conservation and other values (IUCN, 2013). https://portals.iucn.org/library/ node/30018 WWF-UK | WORLD BANK | GLOBAL CANOPY : GEOSPATIAL ESG REPORT Why we are here RL To stop the degradation of the planet’s natural environment and to build a future in which humans live in harmony with nature. Why we are here To stop the degradation of the planet’s natural environment and ular to build a future in which humans live in harmony with nature. wwf.org.uk