Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics REPORT NO: P170838 June 2021 © 2021 The World Bank 1818 H Street NW, Washington DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved. This work is a product of the staff of The World Bank. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of the executive directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory, or the endorsement or acceptance of such boundaries. Photography Shutterstock & Flickr. Rights and Permissions The material in this work is subject to copyright. Because the World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to the work is given. Please cite this work as follows: “World Bank. 2021. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics. © World Bank.” All queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics i Contents List of Figures............................................................................................................................................................................................... iii Authorship...................................................................................................................................................................................................... iv Acknowledgments....................................................................................................................................................................................... v Acronyms and Abbreviations................................................................................................................................................................ vi Executive Summary.....................................................................................................................................................................................1 1 Introduction........................................................................................................................................................................................... 6 1.1 Background................................................................................................................................................................................... 6 1.2 Objectives..................................................................................................................................................................................... 10 2 Innovative Technologies for Improving the MRV Process .............................................................................................. 11 3 Identified Technological Challenges..........................................................................................................................................18 4 Recommendations for Rolling Out Innovative Technologies and Building Enabling Environments to Overcome Challenges.................................................................................................................................. 21 4.1 Short-Term Recommendations (1-2 years)....................................................................................................................... 23 4.2 Long-Term Recommendations (3-5 years)........................................................................................................................ 27 5 References............................................................................................................................................................................................31 Appendix A: Workshop Participants List............................................................................................................................................ 34 Appendix B: Glossary of Terms............................................................................................................................................................. 38 Appendix C: Other Platforms................................................................................................................................................................. 40 Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics ii List of Figures Figure 1-1 Example of a Pantropical Biomass Map................................................................................................................................................... 8 Figure 1-2 Traditional M & MRV Approach for Results-Based Payments and Time Frames....................................................................... 9 Figure 2-1 Spaceborne Satellite Platforms and Sensors Relevant to the Measurement of Aboveground Biomass and Its Dynamics........................................................................................................................................................................................... 12 Figure 2-2 Percentage of Biomass Change, Derived from L-band SAR............................................................................................................ 14 Figure 2-3 Aboveground Biomass (AGB) Stocks for 2010 from the ESA DUE GlobBiomass Project...................................................... 15 Figure 4-1 Implementation Framework...................................................................................................................................................................... 22 Figure 4-2 Potential Biomass Reference Measurement Sites to Set Up to Increase Confidence in Estimates................................... 24 Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics iii Authorship This report was prepared by: GMV Aerospace and Defense Marta Gómez Giménez, María Julia Yagüe Ballester, Beatriz Revilla Romero, Ana Sebastián López, Elsa Carla de Grandi, Omjyoti Dutta, Erika Pinto Bañuls, Antonio Franco Nieto, Patricia Pérez Ramírez, Ángel Fernández Carrillo, Antonio Tabasco Cabezas The University of Edinburgh Iain McNicol, Edward Mitchard, Sam Staddon Insight Centre for Data Analytics (University College Cork) Eduardo Vyhmeister, Gabriel González-Castañé Cyprus University of Technology Phaedon Kyriakidis Aristotle University of Thessaloniki Nikos Nanos Federica Chiappe Consulting Federica Chiappe Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics iv Acknowledgments The preparation of this report was tasked by the World Bank to GMV, leader of the FOCASTOCK consortium. This final report is the result of the joint cooperation among the World Bank, the FOCASTOCK consortium (GMV Aerospace and Defense, the University of Edinburgh, Insight Centre for Data Analytics at the University College Cork, Cyprus University of Technology, Aristotle University of Thessaloniki, and Federica Chiappe Consulting), and a group of international experts. Particular thanks to Andrés Espejo, Senior Carbon Finance Specialist at the World Bank; Elitsa Peneva-Reed, Carbon Science Coordinator at the World Bank; and World Bank experts and facilitators Edward Carr, Erick Fernandes, Keith Garret, Nagaraja Rao Harshadeep, and Rama Chandra Reddy. This final report has benefited from the advice of 86 world experts who provided key inputs during the international consultation workshop organized as part of this project and the collation of this report. The complete list of experts is found in Appendix A. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics v Acronyms and Abbreviations AD activity data AfriTRON African Tropical Rainforest Observation Network AGB aboveground biomass AI artificial intelligence ALOS Advanced Land Observing Satellite ALOS PALSAR Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar BGB belowground biomass C carbon CC cloud computing CCI Climate Change Initiative CEOS Committee on Earth Observation Satellites COP Conference of Parties CTFs Center for Tropical Forest Science DLR Deutsches Zentrum für Luft- und Raumfahrt ECV essential climate variable EF emission factor ENVI Environment for Visualizing Images EO Earth observation EPFL École Polytechnique Fédérale de Lausanne ER emissions reduction ESA European Space Agency ESA ROSE-L European Space Agency Radar Observation System for Europe in L-band FAO Food and Agriculture Organization of the United Nations FCPF Forest Carbon Partnership Facility ForestGEO Forest Global Earth Observatory FREL Forest reference emission level FRL Forest reference level GCOS Global Climate Observing System GEDI Global Ecosystem Dynamics Investigation Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics vi GEE Google Earth Engine GEF Global Environment Facility GEO Group on Earth Observations GFOI Global Forest Observations Initiative GHG greenhouse gases GS geostatistics ICAO International Civil Aviation Organization ICESat Ice, Clouds, and Land Elevation Satellite IPCC Intergovernmental Panel on Climate Change ISRO Indian Space Research Organization JAXA Japan Aerospace Exploration Agency LiDAR Light Detection and Ranging or Laser Imaging Detection and Ranging M & MRV monitoring and measurement, reporting, and verification MAAP Multi-Mission Algorithm and Analysis Platform NASA National Aeronautics and Space Administration NISAR NASA-ISRO Synthetic Aperture Radar NFI National Forest Inventory NICFI Norway’s International Climate and Forest Initiative OBIWAN Online Biomass Inference using Waveforms and iNventory PAMs policies and measures RAINFOR Amazon Forest Inventory Network REDD+ Reducing emissions from deforestation and forest degradation in developing countries RS remote sensing SAOCOM Satélite Argentino de Observación COn Microondas SAR synthetic aperture radar SEOSAW A Socio-Ecological Observatory for Southern African Woodlands SEPAL System for Earth Observations, Data Access, Processing & Analysis for Land Monitoring SOC soil organic carbon UAV unmanned aerial vehicle UNFCCC United Nations Framework Convention on Climate Change WWF World Wide Fund for Nature Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics vii EXECUTIVE SUMMARY Forest-related greenhouse gas (GHG) emissions, · Decreasing the time needed to generate estimates, emission reductions, and enhanced removals (carbon because MRV systems can become operational in sequestration) are estimated by measurement, months, not years, and there is a much smaller reporting, and verification (MRV) systems, usually time lag between the end of a monitoring period, based on a combination of remote sensing data, field and the availability of data. or in situ measurements, and modeling approaches. In this context, the World Bank launched a study to assess Operationalizing the MRV process, is lengthy, however, the readiness of various innovative technologies— often taking years even in countries with currently high including remote sensing (RS), geostatistics (GS), capacities for such a task, and once it is operational, artificial intelligence (AI), and cloud computing (CC)— it relies on a complex, nonstandardized, uncertain, and to identify how these can be combined and leveraged lengthy process of integrating remote sensing and in to foster a next-generation MRV, which would help situ measurements. This negatively affects the ability to unlock climate finance and enable governments to address the drivers of these emissions, and at the and stakeholders to monitor the implementation of same time apply and access climate finance in a timely environmental policies and assess the status of the manner. world’s forests. Under the United Nations Framework Convention on The study began with a review of the current and Climate Change (UNFCCC), the lack of consistency limits potential innovative technologies in order to gain a the comparability between countries and makes the comprehensive understanding of the readiness of these reconciliation of national reports and global estimates technologies, and the challenges to rolling out their that are needed for the 2023 Global Stocktake under implementation. the Paris Agreements difficult. Moreover, the ongoing costs of MRV systems can be high, while the accuracy of Following the review of the technologies, the World the estimates is often low, and thus not able to unlock Bank hosted a virtual two-day international workshop the full potential of climate finance. Traditionally, MRV of experts with the objective of deepening the overall processes have been based on land use and land cover understanding of existing gaps through discussions change (LULCC) approaches, which are heavily reliant of the methodological issues and limits, as well as the on satellite optical imagery. New developments in disruptive technologies and data management tools technology are improving our capabilities for mapping that could contribute to overcoming these obstacles. carbon (C) stocks, and C-stock change with improved During the workshop, specific sessions also covered accuracy. In particular, biomass, which can be obtained the policy and institutional barriers that will need to be through in situ measurements, remote sensing, and addressed in order to deploy these technologies and models, is an essential climate variable (ECV) that offer solutions on how to disrupt the MRV process. provides a direct measurement of C changes and impacts on other ECV, such as land cover. Upcoming As a result, a set of the main technological challenges, and satellites and the ever-falling costs of airborne data recommendations for overcoming them, were identified. (especially from drones) will result in unprecedented The technological challenges can be grouped into four availability of data to support biomass estimation. The areas: data availability and access; processing and combination of innovative approaches and increased computational performance; uncertainty management; availability of data is expected to overcome several and standardization and protocols. major challenges to estimating C stocks by: · Enabling the monitoring of C stocks with increased frequency (<1 year frequency, compared to the current lower frequency of the reporting cycle); · Standardizing C-stock estimation so that data from different sources are compatible and can be easily integrated, and uncertainties can be quantified; and Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 1 The major challenges identified are related to the comprehensive methodological frameworks would lack of free and open access to data, tools, and cloud contribute to overcoming some of the challenges and computing systems. There is a need for transparent achieving the main goal of this analysis, which is to methods and algorithms for turning remotely sensed improve the MRV process, reducing the time needed variables into accurate aboveground biomass (AGB) for MRV implementation, and consequently speeding estimates. Establishing sustainable long-term in up the mobilization of GHG emission reduction-based situ monitoring networks is a challenging task with payment in the short term; and to foster sustainability, multiple contributing factors. To start with, the lack and build an operational service to carry out carbon of secured funding exacerbates the issues with the stock based finance at global scale in the long term. limited availability of representative data and metadata To do so, we have outlined a nonexhaustive list of to estimate and reduce uncertainty, and the difficulty recommendations for the short term (1—2 years) of establishing standardized and universally accepted and for the longer term (3—5 years) and have also measurement protocols. Government reluctance to identified potential coordinators and actors to the best share and exchange local data that are currently located of our knowledge, and based on the state-of-the-art on national servers with centralized cloud computing review and the feedback received from experts. It is systems, and the low bandwidth in many regions, which clear from both the review and the discussions held makes building distributed systems challenging, are during the virtual workshop that a “one size fits all” additional complications. Finally, innovative methods method for AGB mapping and monitoring is unlikely can be difficult to implement owing to the lack of to be achievable, or even desirable, given the different communication among domains, which hinders the requirements, geographic locations, and types of forest integration of GS and AI solutions into the process of under observation; therefore, a new generation of MRV estimating C stock and dynamics. processes will need to be flexible in order to enable its A set of recommendations was developed to make adaptation to various conditions. The recommendations better use of current and upcoming remote sensing we have devised allow for this flexibility and, if technologies through the smart combination of implemented, will lead to an enhancement and geostatistics and AI, deployed to the cloud, and simplification of the MRV process by setting up and anchored on traditional forest inventory data sets. running systems in individual countries. Implementing the proposed technologies into Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 2 Short Term Areas of Needed Improvement Potential Coordinators and Actors (1—2 years) Data Availability and Access Data Availability and Access • Seek international partnerships with remote • Group on Earth Observations (GEO). sensing data providers; support existing efforts • e agencies: European Space Agency (ESA), in in situ data collection; and build on existing National Aeronautics and Space Administration infrastructure so as to facilitate reliable, free, and (NASA), Japan Aerospace Exploration Agency open access to data, algorithms, and centralized (JAXA). cloud computing services through sustainable funding. • Committee on Earth Observation Satellites (CEOS). • Foster partnerships with research groups and institutions developing and maintaining forest • In situ data collection networks and coordination plot networks and support them in additional data mechanisms: GEO-TREES, AfriTRON, CTFS- collection or access (that is, build a Global Forest ForestGEO, ForestPlots.net, RAINFOR, SEOSAW. Biomass Reference System). Processing and Computational Performance Processing and Computational Performance • Support the integration of new approaches (such • RS, GS, AI, and CC research centers and groups. as AI and GS) into traditional ones. • Support the convergence of techniques between research groups (RS, AI/computer vision, GS, CC) as this will enable the enhancement of the developed tools. Uncertainty Management Uncertainty Management • Pilot the implementation of geostatistical (GS) and • RS, GS, AI, and CC research centers and groups, AI solutions through demonstration activities and national agencies. pilots to link in situ and RS data, harnessing the potential of CC. • Include estimates of error propagation from the input data to the final output in MRV systems. Standardization and Protocols Standardization and Protocols • Establish a common understanding of how • CEOS, GFOI. data will be used and processed to address • International Organization for Standardization various needs, through the collection of users’ (ISO), Cloud Security Alliance (CSA), European requirements. AI Alliance, World Economic Forum Global AI • Develop standards and protocols for data Action Alliance. collection and development of the components of the system. • Promote data protection and security protocols for data migrations and protection. Enabling Environments Enabling Environments • Support data policies in situ and RS data) for • GEO, FAO, GFOI, space agencies, plot networks, access and sharing. national authorities, research groups. • Engage with stakeholders to inform them about • Financial institutions, donors. how local data will be used to build confidence throughout the MRV system. • UNFCCC, IPCC, World Bank • Create mechanisms for incentivization, such • Climate AI, European AI Alliance, World as rewards for establishing public-private Economic Forum Global AI Action Alliance, partnerships to promote communication and Global Partnership on AI, International collaboration among relevant institutions and Association of Mathematical Geosciences stakeholders. (IAMG), geoENVia. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 3 Areas of Needed Improvement Potential Coordinators and Actors • Create the necessary financial support mechanisms seeking public and private investments, such as impact investing, blended finance, and voluntary markets. • Create SMART KPIs (Specific, Measurable, Attainable, Relevant, Time-bound Key Performance Indicators) to monitor system implementation. • Designate a perceived neutral entity that coordinates these actions, especially on the quality control of the in situ data, making it available and accessible to stakeholders (similar to the World Meteorological Organization network). • Establish communication among experts and users, which would help to generate confidence and encourage data sharing as well as building on ongoing efforts. Long Term Data Availability and Access Data Availability and Access (3—5 years) • Support plans for the future follow-up to the • Space agencies (NASA, ESA). GEDI and BIOMASS missions, and/or plans for • GEO, CEOS. alternative missions with similar characteristics. • Continue supporting and reinforce the continuation of international partnerships by making satellite data publicly available and the long-term maintenance of a Global Forest Biomass Reference System. Processing and Computational Performance Processing and Computational Performance • Build distributed systems with local micro- • National AI and CC research centers, national clouds in regions without local data migration and local authorities. possibilities. • Promote interoperability between local data centers and central servers. Uncertainty Management Uncertainty Management • Continue exploring innovations such as the • RS, GS, and AI research centers and groups. automation of processing through GS with “meta- models.” • In situ data collection networks and coordination mechanisms: GEO-TREES, • Enhance quantification of spatial patterns from AfriTRON, CTFS-ForestGEO, ForestPlots.net, , training images and combine input data for AGB RAINFOR, SEOSAW. estimation using high spatial resolution satellite imagery, possibly selected via AI methods. • Estimate the impact of overestimating and underestimating carbon stocks on results-based payments. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 4 Long Areas of Needed Improvement Potential Coordinators and Actors Term (3—5 years) Standardization and Protocols Standardization and Protocols • Establish an international framework for adopting • International alliances and partnerships, and standard data management and processing organizations for standardization. approaches deployed in cloud computing systems located in different regions. Enabling Environments Enabling Environments • Analyze whether new data and tools could create • RS, AI, research centers and groups. ethical issues, keeping in mind the risk of “dual • World Bank, FAO. uses” that do not occur in current approaches. • National governments and offices. • Invest in research, training, and knowledge generation in user countries. • Financial institutions, donors. • Support policy frameworks for AI solutions and cloud security. • Foster collaboration among space agencies, international organizations, and governments. • Carry out a full data and capacity-building needs assessment, based on identified target audiences and stakeholders in specific countries before developing a complete strategy to build distributed systems. • Allocate funding to support the regular acquisition of unmanned aerial vehicles (UAVs) and LiDAR data through CEOS and the private sector. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 5 1. INTRODUCTION 1.1 BACKGROUND assurance of environmental integrity. In 2005, the parties to the United Nations Framework Financing mechanisms are provided by various Convention on Climate Change (UNFCCC) began to institutions: For example, the World Bank’s Climate formally set up a framework for financially incentivizing Change Fund Management Unit includes two funds— emissions reduction due to deforestation and forest the Forest Carbon Partnership Facility (FCPF), and the degradation through conservation, the sustainable Initiative of Sustainable Forest Landscapes (ISFL)— management of forests, and enhancement of forest both of which aim to pilot results-based payments and carbon (C) stocks in developing countries. market-based mechanisms of land use interventions at a large scale. These funds have capitalized more In 2013, the Warsaw Framework for REDD+, which than $1 billion dollars for result-based financing was adopted at the 19th Conference of the Parties against emissions reduction (ER) units. In particular, (COP19), provided a comprehensive methodological the FCPF is an ambitious program working with 47 and financing guideline for implementing activities REDD+ country participants and 17 donors; it includes for Reducing Emissions from Deforestation and a Readiness Fund and a Carbon Fund, both focused on forest Degradation (REDD+). According to the Warsaw the implementation of REDD+ programs. Framework, activities aimed at reducing greenhouse gas (GHG) emissions from deforestation and forest To obtain these funds, countries need to first define degradation, and fostering sustainable management their forest reference levels (FRLs) and forest practices in developing countries are to be implemented reference emission levels (FRELs) (FCPF 2020). Parties in three phases (UNFCCC 2021): are required to assess FRELs and/or FRLs—which measure the amount of emissions from deforestation (i) Development of national strategies, policies, and and forest degradation—as well as removals due to measures (PAMs), and identification of capacity the enhancement of C stocks in a given area within a building needs (readiness phase); reference period. Actual results are then compared with (ii) Implementation of demonstration activities, the assessed FRELs in order to mobilize payments for national PAMs, strategies, or action plans that could actions that prove consistency between the FRELs and involve further capacity building and technology FRLs; include transparent information that will allow development; and for recalculation of estimates; and provide a description of the National Forest Monitoring System (NFMS) (FAO (iii) Monitoring and assessing the performance of PAMs 2013). at the national scale, allowing countries to obtain results-based payments. Under the UNFCCC, as well as under other standards, it is required that the methodologies for estimating In 2015, Article 5 of the Paris Agreement, which was GHG emissions are consistent with the guidelines adopted by 196 parties at COP 21, highlighted the developed by the Intergovernmental Panel on Climate pivotal role of results-based financing mechanisms Change (IPCC), and that they comply with the following in reducing GHG emissions, deforestation, and forest principles (FAO 2013): degradation. Although REDD+ is not considered a market-based mechanism (one in which credits are  Adequate to represent C-stock changes generated and transacted to compensate for GHG (representing land use classes and conversions); emissions) under Article 5, it is expected to be part of  Consistent over time (without discontinuities in the Article 6 transactions. Moreover, voluntary markets time-series data); and offsetting programs (for example, ICAO’s CORSIA)  Complete (all the land of a country should be include the generation of credits from REDD+. These included); and market transactions require more robust monitoring,  Transparent (data, tools, and methods should all reporting, and verification (MRV) systems, and be thoroughly described). Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 6 The IPCC has identified five types of carbon pools: (i) resulted in MRV frameworks that are primarily based on aboveground biomass (AGB); (ii) belowground biomass optical data. However, biomass, which can be obtained (BGB); (iii) dead wood; (iv) litter (that is, dissolved through in situ measurements, remote sensing (RS), organic matter; and (v) soil organic carbon (SOC), which and models, is an essential climate variable (ECV) that can be measured and reported as part of national GHG provides a direct measurement of carbon changes and inventories (FAO 2013). impacts on other ECV, such as land cover (FAO 2009). As discussed during the virtual workshop, the RS of When submitting their national GHG inventories, vegetation methods based on optical and C-band SAR parties are encouraged to report on as many of their data sets are not ideal for applying existing biomass significant C pools as possible, according to their mapping algorithms. Fundamentally, these data sets national circumstances, and with methodological tend to describe only the top of the canopy, and they consistency. AGB is the most visible and dynamic pool, struggle to obtain information on forest biomass, and a key component in C inventories, representing particularly in dense forests. In addition, persistent 30 percent of the total terrestrial ecosystems (Kumar cloud cover over the tropics hinders the use of optical and Mutanga 2017). SOC is also an important pool, imagery. especially in some regions. For example, in peatlands it is estimated that the carbon stored in soils could be Over the past decade there have, for the first time, twice as much as that stored in all the world’s forests emerged continuous RS-based maps of aboveground (UNEP 2019), and peat C is released rapidly following forest carbon storage (Saatchi et al.. 2011; Baccini et al.. drainage and/or clearance of the overlying forest. In 2012; Avitabile et al.. 2016). These were made possible moist, tropical forests, SOC represents less than 50 through an innovative spaceborne LiDAR sensor called percent of the total C stock (that is, terrestrial living ICESat that was launched by NASA in 2003, which, plant material—(AGB and BGB)—and soil carbon stock) despite its primary mission being about the thickness (Scharlemann et al.. 2014). SOC and BGB are difficult of ice, collected sparse footprints across the globe from to monitor via satellite-based approaches (FAO 2009); 2003 to 2009, giving information on tree density and thus, BGB is usually inferred from the AGB, via ratios or height that could be related directly to forest biomass specific functions. (Lefsky 2010). These individual LiDAR footprints could not themselves be used to create biomass or biomass The choice of methodologies for collecting data and change maps because they only sampled a tiny fraction compiling GHG inventories follows the MRV approach, of 1 percent of the world’s land surface. However, they which is based on three “pillars” (FAO 2013): could be used to train machine-learning algorithms  Satellite Land Monitoring Systems, which estimate based on other optical and radar remote-sensing layers the activity data (AD);1 to effectively fill in the gaps, and create biomass maps  Terrestrial forest inventory, such as a National without the cost and logistical challenge of collecting Forest Inventory (NFI), which estimates emission LiDAR data across a whole region, for example, factors (EF);2 and Colombia (Asner et al.. 2012).  GHG inventory, which combines ADs and EFs to estimate GHG emissions and removals. Traditionally, the production of AD relied upon small ground-data sets and classical classification algorithms, partially due to the lack of satellite imagery, and sufficient computing power to process these images. EFs have been often estimated through costly traditional forest inventories, which may or may not be repeated and provide estimates at a coarse scale. The availability of free optical data, and the possibility of implementing straightforward methods to derive AD from land use and land cover change (LULCC) have 1 Activity data (AD) refers to trends in land use change (i.e., area changes) 2 The emission factor (EF) provides an estimation of carbon stocks (i.e., emissions or removals per activity data) such as emitted due to deforestation. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 7 Figure 1-1 Example of a Pantropical Biomass Map. Source: Saatchi et al.. 2011 While these biomass maps represent a massive step  Costly in terms of the time and money required forward for carbon accounting and tropical ecology, for field campaigns. MRV systems are complex, and enabling for the first time an accurate estimate of the they depend on the existence of sustained capacity biomass in each region, and biomass gradients, they and capabilities in key institutions during the have very high uncertainty on a pixel level. Indeed, reporting period (5 years); and evidence from independent in situ plots suggests that  Uncertain: The sources of uncertainty are they did not even correctly map large-scale regional related to the following: the quality and suitability gradients (Mitchard et al.. 2014), and did not agree of satellite data; data pre-processing and post- with each other well in many areas. In particular, it has processing; the definition of land cover classes; in become clear that repeating the same method each situ data measurement; and emissions calculations year will not lead to reliable biomass change estimates: using an integration of AD and EF, which Since individual pixels have errors of 30—40 percent, oversimplifies reality. changes related to forest growth or degradation (which Some of the challenges associated with the estimation typically have smaller percentage changes) will not of GHG emissions for results-based payments for be captured this way. One study attempted to repeat REDD+ could be overcome by using and combining new these methods (Baccini et al.. 2017), creating annual technologies and other data sets that present better maps of forest carbon stock change over a decade, relationships with forest structure and biomass, such but the results are not widely considered credible in as long-wavelength SAR missions (L -and P-band SAR). the scientific community (Hansen et al.. 2019), and it Innovative approaches will help: is accepted that extrapolated maps based on passive optical remote sensing will not produce accurate  Reduce the cost of national or subnational-wide enough change data for this method to work. Therefore, ground-data surveys by a fraction of the original innovative methods and data should be used to provide cost. accurate AGB estimations and carbon stock changes.  Substantially decrease the time needed However, the production of biomass maps and the to implement the MRV cycle (from years/months estimation of C-stock change based on RS data have not to weeks). (Figure 1-2). been taken to an operational stage, and MRV systems  Improve deforestation and afforestation rely on traditional methods that combine AD and EF. estimates, and therefore GHG emissions Therefore, the mechanism for calculating emissions for and removals through satellite-based AGB MRV is: measurements, and information about the associated uncertainty.  Slow due to a lack of automation, computing power, adequate infrastructure, know-how, In order to analyze the possibilities for overcoming and standardization. Even if MRV systems the existing challenges that are hindering the current are operational and sustainable, the time for REDD+ MRV approach, the World Bank hired a conducting the measuring and reporting varies consortium led by GMV Aerospace and Defence (GMV) from 3 to 16 months, depending on the country, and to assess the readiness of innovative technologies and for verification, 6—12 months is required; approaches in fields such as geostatistics (GS), artificial Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 8 intelligence (AI), and cloud computing (CC) to boost the Payments” held November 16—17, 2020. An online use of remote sensing of vegetation to estimate forest survey was also conducted to receive feedback and carbon stocks and dynamics. The consortium, led by recommendations on how to improve and disrupt the GMV and formed by the University of Edinburgh, the current MRV process (Figure 1-2). Insight Centre for Data Analytics-University College The workshop participants analyzed five key areas Cork, and GS consultants from the Cyprus University of that play a relevant role in building a REDD+ enabling Technology and Aristotle University of Thessaloniki, and environment for rolling out innovative technologies: enabling environments (provided by Federica Chiappe Consulting Ltd.) performed a state-of-the-art review i. Policy, regulatory, and institutional frameworks on current and potential innovative technologies. The ii. Finance and economics results of this review were shared during a virtual iii. Technology and markets International Workshop of Experts on “Disrupting iv. Information and capacity Carbon Stock Dynamics Estimation for Results-Based v. Social, cultural, and behavioral factors Figure 1-2 Traditional M & MRV Approach for Results-Based Payments and Time Frames VALIDATION 1. Project validation MEASUREMENT 2. Measurement and estimation of funds RESULTS-BASED FINANCE M+R: 5. Payments of carbon 3–16 months credits / label issuance REPORTING 3. Periodic emissions report VERIFICATION 4. Verification against an international standard V: 6–12 months Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 9 1.2 OBJECTIVES MRV process, and map AGB, using RS-based technologies. The objective of this report is to assess the feasibility of iv. Define an implementation framework leading improving and enhancing the role of RS and emerging to the disruption of the MRV process based on technologies such as GS, AI/machine learning, and CC, recommendations for rolling out innovative in order to estimate GHG emissions and to achieve more technologies and identified enablers (institutional rapid mobilization of results-based payments. frameworks, mandates, and incentives) to ensure The specific objectives of the report are to: the operationalization of RS-based technologies and processing approaches. i. Analyze the results of a state-of-the-art review on innovative technologies, and feedback collected The report is structured as follows: First, we present during the virtual International Workshop of the results of the state-of-the art review and the Experts on “Disrupting Carbon Stock Dynamics feedback received from experts. Second, we identify Estimation for Results-Based Payments.” the challenges associated with the implementation of ii. Gain a comprehensive understanding of the these technologies and provide recommendations for readiness of RS of vegetation, and GS, AI, and CC ways to disrupt the MRV process. Finally, we discuss technologies for disrupting the MRV process and the enabling environments that will be required in overcoming the aforementioned challenges. order to potentially reduce the time needed for MRV iii. Present the identified frontier technologies and implementation, and therefore speed up mobilization of state-of-the-art approaches for disrupting the results-based payments. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 10 2. INNOVATIVE TECHNOLOGIES FOR IMPROVING THE MRV PROCESS Remote sensing (RS) refers to a technology that employs However, it is important to acknowledge that, active or passive sensors that can scan the Earth’s because of the wide range and sensitivity of surface and process the data captured to infer spatially available sensors, a “one size fits all” method of continuous, meaningful data, and information that is AGB mapping and monitoring is unlikely to be directly usable for understanding and monitoring (at achievable. Moreover, methods will depend on user various scales) many of the natural and anthropogenic requirements, such as the geographic extent, the type activities taking place on our planet. of vegetation, the AGB densities under consideration, and the objective of the report—for example, whether We are now entering a golden age of RS, with a plethora the need is for the most recent up-to-date AGB estimate, of spaceborne, airborne, and ground-based platforms or for a long-term AGB trend for establishing baselines. and sensors that are either currently operating or From our state-of-the-art review and the subsequent scheduled to become operational within the next 1—5 discussions held during the International Workshop years (Figure 2-1). The launch of new missions by of Experts on “Disrupting Carbon Stock Dynamics space agencies is causing an unprecedented increase Estimation for Results-Based Payments,” it is clear in imagery availability and revisit rates. The availability that achieving global and temporally consistent carbon of new spaceborne platforms specifically designed for stock estimations annually, and with errors below 20 forest aboveground biomass (AGB) mapping has the percent, as required by the Intergovernmental Panel on potential to greatly enhance our capacity to develop Climate Change (IPCC) and the RS community,3 remains a monitoring system capable of reporting changes a significant technological and logistical challenge, within the time frame of the monitoring, reporting, and particularly in the short to medium term (2-3 years). verification (MRV) process. 3 The target set by the IPCC and GCOS (Global Climate Observing System) is for a global and temporally consistent AGB monitoring system, with data sets generated annually at a -100-50 meter resolution, with relative errors < 20 % in areas with AGB densities >50 Mg ha1-, and a fixed error of 10 Mg ha1- in lower-density areas. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 11 Figure 2-1 Spaceborne Satellite Platforms and Sensors Relevant to the Measurement of Aboveground Biomass and Its Dynamics (1—100 m) (100—1,000 m) (1,000+ m) Note: This summary of spaceborne satellite platforms and sensors that are either dedicated to the measurement of AGB and its dynamics, or have been demonstrated as useful in its derivation, either in isolation, or as part of a multi-sensor approach, is based on our state-of-the- art review. The launch date of future missions should be considered nominal, and subject to change. The $ symbol indicates that the full data catalog is not free to access; $* indicates free products are available. The spatial resolution refers to the scale at which these products are typically aggregated for wider use. RS systems can be located on the ground, but in this and LiDAR, which emits laser beams, are examples of report, when discussing RS, we will refer to either active sensors. Active sensors are key to monitoring airborne sensors (onboard a plane or unmanned aerial forests because they have the capacity to penetrate vehicle, or UAV), or spaceborne sensors (onboard the forest canopy, and in some cases clouds, which is satellites). These systems are defined depending on the useful since most tropical forests are located in areas source of the energy they detect. Passive sensors detect with frequent continuous cloud cover, which represents the radiation that objects naturally emit or reflect (for a major problem for passive sensors. SAR can penetrate example, reflected sunlight). Active sensors emit their clouds and provide volume and height estimates to own source of energy, and thus operate independently measure biomass and LiDAR instruments are able to of solar illumination, which is then scattered from accurately map the 3D structure of stands of trees, the target and received back at the sensor. Synthetic even identifying the size and shape of individual trees; aperture radar (SAR), which emits microwave pulses, however, unlike SAR, they cannot penetrate clouds. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 12 A major feature of SAR systems is that radiation NASA’s Global Ecosystems Dynamics Investigation penetrates more or less through the canopy, depending (GEDI), onboard the International Space Station on the specific wavelength at which the system is (Dubayah et al.. 2020), and their Ice, Clouds and Land operating. For instance, X-band (short wavelength) Elevation Satellite (ICESat-2). The GEDI mission is radiation interacts with the surface of the forest canopy the first spaceborne LiDAR specifically tasked with and is backscattered by small-scale components such collecting data on tree canopy height, canopy cover, and as foliage and small branches, whereas P-band (long various other metrics of the vertical forest structure, all wavelength) radiation penetrates deeper into the within 25-meter footprints, much smaller than ICESat’s canopy and is scattered back by larger components, 70 meters. A global network of coincident in situ field such as large branches, tree stems, and the surface and airborne LiDAR data sets will be used to develop of the terrain. Since most of the biomass is contained and refine calibration models for converting GEDI- in the stems and largest branches, P-band SAR is the derived metrics of forest structure to AGB density, preferred sensor for mapping AGB; however, there has both at the footprint level (25 meters) and as part of a never been a spaceborne SAR sensor operating at this continuous, but coarser-resolution 1-kilometer product wavelength, meaning that for many years AGB has been (Duncanson et al.. 2020; Patterson et al.. 2019). These modeled instead, with relative success, using L-band data will only ever cover a small percentage of the (the next shortest wavelength from P) and C-band world’s surface (4 percent for GEDI), and thus need (Bouvet et al.. 2018; McNicol et al.. 2018; Rodríguez- to be combined with other RS data, namely SAR, and Veiga et al.. 2019). The challenge with C-band and to a lesser extent multispectral optical imagery, to L-band SAR systems is that when used in isolation, their extrapolate the data contained in these small footprints signal saturates at relatively low AGB levels,4 estimated to the wider region. Such a multisensor approach— to be at around 50 Mg ha-1 for C-band, and 150 Mg ha-1 leveraging discrete LiDAR samples as a basis for for L-band, values that that preclude the measurement creating wall-to-wall data sets—has formed the basis of of AGB in most intact tropical forests, where densities several national and regional products created over the exceed 200 Mg ha-1. This means that above this point, last decade, including the benchmark pantropical AGB differences in AGB are no longer captured. Conversely, maps of Saatchi et al.. (2011) and Baccini et al.. (2012). P-band airborne data is shown to respond to AGB With a new generation of platforms and sensors now changes in forests well over 200 Mg ha-1, and indeed available, innovative methods and data are emerging, above 500 Mg ha-1 in French Guiana (Minh et al.. 2016). including those that combine GEDI with SAR data, such as TanDEM-X, to produce contiguous forest biomass ESA’s BIOMASS mission, which will operate on P-band, maps at both 1-kilometer and 1-hectare resolution, is scheduled to launch in 2023, and is expected to with the latter achieving accuracies ranging from 11 to be a game changer in this regard, by improving AGB 27 percent (Qi et al.. 2019). Despite clear promise, the estimates, particularly over tropical areas (which have challenge for MRV is that TanDEM-X, along with many high AGB densities) and overcoming the saturation other SAR data, is currently a commercial product, issues with shorter wavelength systems. It is worth meaning that the costs associated with obtaining a noting that spaceborne SAR sensors do not measure global annual data set is likely to hinder widespread biomass or carbon stocks directly, but rather fusion attempts in support of this process. (See Figure parameters that correlate with biomass, such as forest 2-2.) structure and volume, and canopy height. Overall, SAR and LiDAR sensors have proven to be more suitable The vast amount of optical satellite imagery now for accurate AGB modeling because of their capability available, which is in many cases free of charge, has for penetrating the canopy and providing information the potential to contribute to the overall process of about the forest 3D structure. Optical sensors, which integrating innovative approaches analyzed in this are limited to measurements of the visible (2D) surface, report, possibly by discovering new patterns and can provide no information on the vertical structure or correlations between satellite imagery and forest density of the forest. AGB (Figure 2-2). Indeed, several recent studies have already explored the possibility of combining LiDAR Along with the ESA BIOMASS mission, two spaceborne with freely available optical data from Planet, Landsat, missions are now collecting LiDAR data at global scales: and Sentinel using machine-learning approaches, both to estimate AGB (Csillik et al.. 2019) and, more 4 Estimated to be around 50 Mg ha1- for C-band, and 150 Mg ha1- for L-band, values that preclude the measurement of AGB in most intact tropical forests, where densities exceed 200 exceed Mg ha1-. 200 Mg ha1- Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 13 e vast amount of optical satellite imagery now available, which is in many cases free of charge, has th tential to contribute to the overall process of integrating innovative approaches analyzed in this repor ssibly by discovering new patterns and correlations between satellite imagery and forest AGB. (Figure 2 several recent Indeed,prevalently, to measure studies have tree height already (Ploton explored et al.. 2017; andthethe possibility of combining evidence obtained LiDARpanel from the expert with freely Lang et al.. 2019), a reliable proxy for AGB for which that SAR is the remote sensing technology with the ailable optical data from Planet, Landsat, and Sentinel several calibration models exist (Asner and Mascaro using machine-learning approaches, both t greatest maturity in terms of readiness to address the timate AGB2014;(Csillik Jucker etet al al.. 2019), 2017). and Despite more these prevalently, advances, these to measure challenges of tree height the MRV (Ploton process, et al 2017; particularly sensors Lang et a methods proxy 19), a reliable remainfor AGB very much forinwhich several calibration the proof-of-concept models operating exist at L-band (Asner (Figure and 2—2), which Mascaro 2014; Jucke have provided al 2017). Despite stage these and require advances, independent these before validation methods they remain very relatively much accurate in the of estimates proof-of-concept AGB in areas with low stage an can be considered suitable for wider application. to moderate AGB density (0—150 Mg ha-1) (Bouvet et al.. quire independent validation before they can be considered as suitable for wider application. Howeve However, the main barrier to developing reliable and 2018; McNicol et al.. 2018). e main barrier developing toconsistent temporally reliable monitoring and temporally strategies based on consistent monitoring strategies based on LiDAR i at our current both areas Crucially, these are about comprise likely90 topercent of futur LiDAR is spaceborne that our current LiDAR sensors, spaceborne LiDARGEDI and ICESAT-2, sensors, of which underpin the land surface globally, which shows the significant GEDI andare apping efforts, not operational ICESat-2, both of which aresatellites, and have potential likely to underpin no guarantee of long-term coverage. Airborne dat of the SAR data sets for operational AGB future mapping efforts, are not operational satellites, llection using aircraft, while capable of replicating discrete LiDAR coverage at the regional to nationa mapping in many regions. However, tropical forests and and have no guarantee of long-term coverage. ales, is prohibitively expensive ($200-500 per km ), and Airborne data collection using aircraft, while capable of 2 is unlikely other areas with be densities toAGB part of any long-term that are monitorin considerably gime. replicating discrete LiDAR coverage at the regional to greater than 150 Mg ha will remain challenging to -1 measure due to the saturation of the SAR signal. There national scales, is prohibitively expensive ($200—$500 s therefore clearkilometer), per square from both and the literature is unlikely of anyand the evidence obtained from the expert panel, tha review to be part is no clear and readily available method for operational R is the sensing technology remotemonitoring long-term regime. with the greatest maturity large-scale AGB in termsand mapping of monitoring readiness to address within the th next 2 years, at least until the launch of BIOMASS. allenges of the MRV process, particularly sensors operating at L-band (Figure 2-2), which have provide It is therefore clear from both the literature review atively accurate estimates of AGB in areas with low to moderate AGB density (0-150 Mg ha ) (Bouvet e -1 2018; McNicol et Figure 2-2 al 2018). Percentage of Biomass Change, Derived from L-band SAR Source: McNicol et al.. 2018. Percentage Figure 2-2 Note: of biomass Areas where change, derived seasonal differences dataL-band from prohibited analysisSAR (McNicol are shown et al in white. 2018). SAR Areas = synthetic whereradar. aperture seasonal differences prohibited data analysis are shown in white. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 14 Current state-of-the-art regional to global mapping capable of supporting quantification of AGB change in efforts necessarily rely on freely available RS data, support of MRV in the short term (1—2 years).5 The prominent among which are the ALOS PALSAR annual data sets for 2010 and 2017 have already been created mosaic products available from the Japan Aerospace and are available for public use. However, per-pixel Exploration Agency (JAXA), which are delivered fully uncertainties on these products are high, around 40— processed with a latency period of 1—2 years and are 50 percent—values which, at present, should preclude available through Google Earth Engine (GEE) (Bouvet their use in AGB change mapping.6 Furthermore, as et al.. 2018; McNicol et al.. 2018; Santoro et al.. 2020). previously noted, despite their global coverage, these The ESA Climate Change Initiative (CCI) ESA DUE Biomass CCI data sets cannot be used to estimate AGB GlobBiomass project targets the creation of consistent in forests with densities >150—200 Mg ha-1 due to the global AGB maps at 1-hectare resolution for three saturation of L-band SAR. epochs (2010, 2017, and 2018), which are theoretically Figure 2-3 Aboveground Biomass (AGB) Stocks for 2010 from the ESA DUE GlobBiomass Project Note: The ESA DUE GlobBiomass project is a precursor to the ESA Biomass CCI project (Santoro et al.. 2020). The 47 countries delineated in red form part of the Forest Carbon Partnership Facility (FCPF), of which 33 have at least 75 percent of forested areas with AGB densities < 150 Mg ha-1. 5 To generate maps of AGB and AGB change, the project uses data from past, current, and future satellite missions, including optical sensors (for example, Sentinel 2A/B), C-band (Sentinel 1A & B), and L-band (ALOS-2 PALSAR-2) SAR data, and spaceborne LiDAR (for example, NASA's GEDI). https://climate.esa.int/en/projects/biomass/about/ 6 Reducing the uncertainties on per-pixel estimates may be possible if aggregated to coarser resolution, assuming that errors are random, and the product does not contain systematic biases. However, such data sets are more difficult to validate using standard comparisons, as there exists little ground data with a spatial resolution larger than 1 hectare. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 15 There are other open issues related to the quality of monitoring, will provide data free of charge, and this the input SAR data itself that may be contributing to will result in a new generation of AGB data sets capable increasing the uncertainties in the AGB estimations. of supporting the MRV process in the mid to long term One of the most obvious is that the “mosaics” comprise (5—10 years). Longevity in free L-band data is likely images collected throughout the year, meaning that to be provided by the ESA ROSE-L mission, which is they are subject to the variable influence of surface considered a high-priority candidate in the Copernicus moisture, which can enhance the backscatter intensity program, and is scheduled for launch toward 2027. in dry conditions. Mitigating the seasonal and These data, combined with inputs from ALOS-4, weather-related effects on SAR data will likely require PALSAR-3, and other L-band platforms, including the a complete regeneration of the products; however, Earth observation satellite constellation of Argentina’s to date, the cost of acquiring the necessary number space agency, SAOCOM (Satélite Argentino de of ALOS PALSAR images required to generate the Observación COn Microondas; Spanish for Argentine mosaics over large areas for a given year has been Microwaves Observation Satellite), have the potential considered prohibitively expensive ($2,300 for a single to contribute to a long-term AGB monitoring system 70 x 70 kilometer acquisition). Certainly, for global capable of annual reporting at global scales should mapping, cost may not be an issue if implemented by a suitable solution to the commercial restrictions be groups of nations, for example the 47 countries that found. Figure 2-1 shows the type of access, availability, are part of the Forest Carbon Partnership Facility and the expected launch timeline of these missions. (FCPF). New opportunities may also arise with the recent announcement that JAXA will provide free Together with new satellite missions, and new and open access to the ScanSAR observation data developments in data science, artificial intelligence from ALOS and ALOS-2 (60- to 100-meter resolution); (AI) and geostatistics (GS) are also improving the however, the time frame of this release has yet to capability for obtaining more accurate predictions be established. The launch of NASA-ISRO’s synthetic and quantifying uncertainty. Their implementation aperture radar (NISAR), an L-band mission scheduled in MRV frameworks, which is not yet complete and for launch in 2022 at the earliest, will change this comprehensive, could help to mobilize greenhouse situation, by distributing L-band SAR data for free gas (GHG) emission reduction-based payments. For and providing AGB data with an accuracy of up to 20 example, geostatistical methods (including model- percent. based inference) represent relatively mature technologies in the forestry domain, but for the As a result of these open issues, the currently most part they have not yet found their way into an available global AGB products have a high level of operational MRV context because of their complexity. discrepancies (Saatchi et al.. 2011; Baccini et al.. 2012). However, promising data sets for AGB estimation have Currently, the linking of ground-based AGB to RS been recently made available, and more will shortly be data in an MRV context is typically achieved through made available through various space missions that rather simple models that often yield poor predictive are soon to launch. As shown in Figure 2-1, many new performance. Nevertheless, GS offers a multitude satellite missions (SAR, LiDAR, and optical) will render of methods and algorithms for integrating data of abundant high- and medium-resolution data during multiple variables and improving spatial prediction/ the next decade. However, none of these sensors or mapping, including: methods will be capable of accurately mapping AGB i. Spatial regression models (linear or nonlinear) i across all vegetation types in all regions, at least that account for spatial correlation, differences in with a spatial and temporal resolution sufficient for data resolution, and measurement error. the detection of small patterns and changes in AGB, ii. Regression models, combined with advanced ii including those associated with forest degradation spatial interpolation methods, which address the and growth. This was confirmed by the experts issue of spatial misalignment. participating in the workshop, who discussed the need iii. Spatiotemporal geostatistical models accounting iii to base operational AGB monitoring on a combination for the temporal dimension of the data. of sensors. iv iv. Computational procedures for handling large data The upcoming BIOMASS and NISAR missions (in 2022— sets. 2023), specifically designed around AGB mapping and v v. Bayesian extensions of all of the above. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 16 Geostatistical regression models (see ii. above) have behavior could be accomplished using AI solutions. direct links to model-based inference and hybrid Similarly, patterns and correlations within the inference, which are widely employed in the forestry collected information would be performed by context. Moreover, GS provides a comprehensive various machine-learning approaches that would framework for error and/or uncertainty modeling/ help reduce the burden of dimensionality in the propagation. Geostatistical simulation, in collected information, and at the same time particular, extends classical nonspatial simulation facilitate handling information from annexed to account for spatial and/or spatiotemporal approaches (for example, cloud computing processing). correlation, as well as different data resolutions, At the current stage, various AI applications already and constitutes an all-around approach for spatial exist in several platforms that help bring tools and error propagation, particularly when results need scientific communities together: GEE,7 EO-learn (open- to be reported at various spatial resolutions (pixel source Python library),8 Radiant MLHub,9 Open Data versus regional versus national). Spatiotemporal Cube,10 and ENVI in the Cloud,11 among others. Apart geostatistical models for AGB (see iii above) are being from all of the work performed so far, efforts have to actively developed, particularly for combining time- be made to facilitate the application of AI technologies series analysis with spatial statistics for understanding within the MRV framework. temporal patterns and actual changes that occurred Cloud computing (CC) is a mature technology that between the times of image acquisition. Finally, is shifting the paradigm in processing large data satellite-based AGB estimation could benefit volumes and ensuring scalability. The foundational from the advent of multiple-point GS, whereby technologies and systems enabling cloud services spatial patterns are first “learned” from training are consolidated and standardized. CC underpins images and are then “exported” to the ground a vast number of services and information backups via geostatistical simulation, and fused with that allow large enterprises to host all their data and actual data to provide realistic models of spatial run their applications in the cloud. It is based on the heterogeneity and complexity. This is a very active concept of dynamic provisioning, which is applied area of research, which is being applied in fields such not only to services but also to computing capability, as soil science and engineering, recently in combination storage, networking, and information technology (IT) with RS techniques, which could improve the accuracy infrastructure in general. Resources are made available of AGB estimates. through the internet and offered on a pay-per-use basis Geostatistical approaches and AI solutions can from CC vendors. However, there are currently very few be considered as relatively mature technology end-to-end examples of AI algorithms that are employed that could be used at the various stages of to derive AGB from satellite data and are deployed in a AGB estimation from space, mainly to support cloud environment. image processing and pattern recognition within As more satellite sensors are launched, the availability the remotely sensed information. The algorithms of data will progressively increase. Some advances that can be considered for modeling AGB dynamics already exist, such as cloud computing platforms that and classification purposes would have to be, of provide global maps, and some progress has been made course, tuned to the specific application. Given toward the automation of AGB estimation. However, the need for image processing and recognition of there is still a long way to go before the efficient patterns within RS data, it is easily predicted that the combination of these technologies will allow us to build various methodologies involved in image processing a wall-to-wall AGB processing chain that speeds up MRV would be incorporated into AGB estimations. This implementation and mobilizes emissions reduction- implies that deep neural networks and the various based payments. In the following section of this report, approaches used for its performance improvement we highlight the main challenges of the presented would be applied. AI would not only be used for technologies, which have been grouped into four topics: image processing. Other tasks, such as inter/ data availability and accessibility; processing and intra-annual dynamic estimation, uncertainties computational performance; uncertainty management; estimation, data filtering, data processing/ and standardization and protocols. curation, and data insight/variables/system 7 Google Earth Engine (GEE) is a cloud computing platform for processing and analyzing satellite imagery and other Earth observation (EO) data. 8 EO-learn is a collection of Python packages that fuse AI and remote sensing techniques, and have been developed to seamlessly access and process spatiotemporal image sequences. 9 Radiant MLHub is an open library for ready-to-use, open-source geospatial training data, and advanced machine-learning applications on EO. 10 Open Data Cube is an open-source geospatial data management tool in which data is organized as a multidimensional array of values. 11 ENVI in the Cloud provides users with the full functionality of software packages like ENVI, in a powerful cloud-hosted IT environment. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 17 3. IDENTIFIED TECHNOLOGICAL CHALLENGES Data Availability and Access to be required in areas that are undergoing forest disturbance or encroachment (Herold et al.. 2019). The first of these challenges is the need for free and However, securing the necessary funding to open access to remote sensing (RS) data and maintain and remeasure these ground networks algorithms that can be produced in a timely and is often difficult, and it represents a challenge to cost-effective manner, within the time frame of the the establishment of long-term in situ monitoring monitoring, reporting, and verification (MRV) cycle. networks. This is critical to ensuring reproducibility. The open- data policy adopted by ESA and NASA has gone some A global biomass reference system would also require way toward addressing this requirement; however, the capacity to handle and process immense amounts the data sets on which the most promising results of data from in situ, airborne, and spaceborne are based, including TanDEM-X and ALOS PALSAR, are sensors. The challenge here is not only the required not available free of charge, while those that are, for computational power but also the ability to efficiently example C-band SAR and optical data from the Sentinel integrate these data sets into methodological missions and Landsat, are not necessarily the most frameworks. To do so, there are two challenges that ? appropriate data sets for aboveground biomass (AGB) have to be overcome. First, geostatistics (GS) and mapping and monitoring (CEOS 2021b). Other useful artificial intelligence (AI), like all analytical frameworks, satellite data sets, while open, were not available in rely on representative reference data, such as the past—for example, NASA’s GEDI spaceborne LiDAR common sites and plots where different methods can was only given its two-year slot on the International be tested and/or validated. Therefore, the limited Space Station starting in late 2018, though its timeline availability of representative data (both in terms has fortunately been extended. ESA’s BIOMASS mission of quantity and quality), along with the scarcity will likely only operate for four years, and there is of metadata about their generation, constitutes no plan for a successor satellite. In some ways, this a key barrier toward the application of GS in an makes these data sets even less useful for operational MRV context, since it hinders confidence building MRV systems than commercial data sets, since once in comparative studies as well as uncertainty they are no longer operational there is no possibility of reduction when communicating results to experts gaining access to the data, even if budget is available. and stakeholders. Second, data collection should meet the calibration and validation requirements Another challenge is the need for transparent set for all of the domains (that is, RS, GS, and AI) methods and algorithms for converting RS that depend on the application and expected functions variables to AGB estimates, which requires and outputs of the methodological approach (image accurate in situ data for calibration and processing, regressions, time-series analysis, etc.) validation. A smaller number of scientific in situ plots are available, and are very useful for calibration and Processing and Computational validation, but that is not their primary purpose, so Performance they are often not located where they would ideally Another barrier is related to scaling up from local be for this purpose, and are not measured repeatedly. estimations to the national scale and beyond. Different Remeasurement of AGB in situ plots should be done model parameterizations, such as stratification by at least every 2—5 years in order to account for forest type, are required for scaling GS applications changes and update calibration models and validation to large areas, along with computational methods for databases, with more frequent observations likely Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 18 inverting very large covariance matrices. Different nonforest), positional errors, attribute errors, model parameterizations also entail decisions that temporal uncertainty, and completeness. are critical in model development, rendering the Finally, the lack of familiarity with geostatistical task of automation rather challenging. Additional methods of the people involved in operational improvements are needed to determine optimum MRV highlights the need for such methods to be methods for integrating (with linear or nonlinear appropriately communicated. models) in situ and RS data in AI models, while explicitly accounting for spatial correlation; accommodating the Overall, geostatistical methods are well developed temporal dimension of the data and models of AGB and can contribute to improving uncertainty change; and considering differences in data resolution, estimations. Yet there are two main areas that measurement error, and spatial misalignment between require improvement: different data sources, as well as the complexity of different environments when scaling up (spatial i. Accounting for different sources of uncertainty, heterogeneity), among other issues. Geostatistical including spatial data quality (along with methods also include model-based inference methods uncertainty/vagueness in definitions—for example, that are widely employed in forestry. Additionally, the forest versus nonforest), positional errors, attribute development of new algorithms and solutions errors, temporal uncertainty, and completeness; and may require not only in situ data collection but ii. Identifying optimal GS simulation methods also well-defined user requirements in order to for spatial error propagation and expanding develop solutions that fit the objectives, scale, classical Monte Carlo methods13 to include and expected performance of the devised tool. spatial correlation and differences in data Therefore, the time planned for building a monitoring resolution. This would be useful because system should also consider the time frame needed to simulation results may apply to a fine spatial collect and implement user requirements. resolution, and subsequently be aggregated to a coarser scale—for example, to the country level—for The availability of large volumes of data requires reporting purposes. powerful computing systems with efficient computational performance, and the resources for Regarding specific AI solutions, the main limitation that massive data storage. Centralized clouds are frequently is clearly jeopardizing a “relatively” fast implementation used for storing and processing RS data, reducing in the MRV processes is the lack of communication computational time and cost. Such methods have between the RS and AI domains, which is hindering already been used to estimate AGB in Sub-Saharan the adequate integration of AI solutions into the Africa using very high-resolution satellite imagery.12 carbon stock estimation process. Uncertainty Management Standardization and Protocols Differences of calibration between observations of Despite the efforts of Open Science,14 the lack of open the same sensor type in time and space are difficult methodologies and algorithms is still a common to correct. Ensuring satellite data comparability will issue that remains unresolved and is hindering require separate calibrations based on data that standardization. In addition, the lack of broadly is collected concurrently with the period of image accepted guidelines in developing tools for specific acquisition. National Forest Inventories (NFIs) have tasks (for example, defining system variables) and the potential to contribute to this; however, at the data management (that is how, and in which format international level, concerns and issues related products will be accessed, stored, and shared) can to plot size, data access, standardization, and pose challenges to the reliability and consistency of measurement protocols, or lack thereof, can create open processing systems. additional challenges and uncertainties. Moreover, Cloud computing (CC ) is a well-developed technology difficulties arise when attempting to account for error for managing resources using standard protocols and and uncertainty in various MRV steps, including the producing scalable products. There are International quality of spatial data (along with uncertainty or Organization for Standardization (ISO) standards for vagueness in definitions—for example forest versus cloud interoperability, although the level of 12 “Counting Trees and Shrubs in the Sub-Sahara Using Cloud Computing”, https://www.nccs.nasa.gov/news-events/nccs-highlights/counting-shrubs-trees-using-cloud-computing. 13 Monte Carlo methods are one of the two methods to combine uncertainties accepted by the IPCC and is based on numerical simulations that draw pseudo-random samples from probability density functions representing the population of each parameter involved in the estimation. 14 Open Science is a new approach to the scientific process based on cooperative work and new ways of sharing knowledge by using digital technologies and collaborative tools. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 19 implementation and application worldwide is campaigns). Within this architecture, data can be not known. Potential barriers highlighted during processed and analyzed independently in local data the virtual workshop were related to massive data centers that connect to the core data center. However, storage and governmental reluctance to share and bandwidth resources and connectivity issues in migrate data from national servers and research some regions of the world may pose challenges to centers to private cloud providers located in foreign building these systems. countries. Moreover, dependency on a particular In summary, the main challenges to be considered cloud provider could, on the one hand, constrain the while building these systems are: modification of components; and on the other hand, reduce interoperability issues. · Data storage, sharing, and migration from local servers to a centralized system; At a global scale, one could also employ a client- · Processes distributed in different clouds could server model of distributed computing through edge produce interoperability issues; and computing, through which multiple local users share · Absence of internet connectivity: Establishing their computing resources to be run as one system. computing system architecture based on high data Edge computing is an emerging cloud architecture transfer remains a challenge in some parts of that can contribute to solving some national data- the world because of connectivity and bandwidth sharing issues (for example, NFI, soil sampling, field limitations. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 20 4. RECOMMENDATIONS FOR ROLLING OUT INNOVATIVE TECHNOLOGIES AND BUILDING ENABLING ENVIRONMENTS TO OVERCOME CHALLENGES In general, technological advancements have been First, we have identified four areas for improving MRV relatively quick, because of widespread confidence processes, especially C-stock measurements, from a in their methods and the resultant data, as well as technological perspective: data availability and access; in their ability to improve carbon measurements and processing and computational performance; uncertainty MRV processes. However, the enabling environment management; and standardization and protocols. has not always been conducive to change, and the Second, we have highlighted appropriate enabling policies that support technological advancement have environments for rolling out these innovative progressed at a much slower rate, and consequently technologies. We have proposed five categories: have not kept up with this speed. There is an increased policies and regulations; institutions and stakeholders; risk that since the scientific community has greater capacity and information; finance and sustainability; faith in the data and tools than policy makers do, as and social, cultural, and behavioral factors. this gap widens, challenges to the uptake and use of new technologies could intensify further. Therefore, it was reiterated in the virtual workshop discussions that technological advancements in REDD+ monitoring, reporting, and verification (MRV) cannot function without positive enabling environments; the latter are key to implementation. Implementing the proposed technologies into comprehensive methodological frameworks would contribute to overcoming the challenges and achieving the main goal of this analysis, which is to improve the MRV process, reducing the time of MRV implementation, and consequently speeding up the mobilization of greenhouse gas (GHG) emission reduction-based payments in the short term; and to foster sustainability and build an operational service to carry out carbon stock-based finance at a global scale in the long term. To do so, we have outlined recommendations for the short term (1—2 years) as well as for the long term (3—5 years) to the best of our knowledge, based on the state-of-the-art review and the feedback received from experts (Figure 4-1). Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 21 Figure 4-1 Implementation Framework SHORT TERM LONG TERM Data Availability and Access Data Availability and Access • Seek international partnerships that foster free and • Support future satellite missions with similar INTERMEDIATE open access to data, algorithms, and centralized characteristics. OUTCOMES cloud services through sustainable funding. • Ensure and reinforce international partnerships to • Support forest inventory updates and new make satellite data publicly available, and increase initiatives to build a Global Forest Biomass the Global Forest Biomass Reference System. Reference System. Processing and Computational Performance Processing and Computational Performance • Support the integration of new approaches • Build distributed systems with local micro-clouds in (biomass estimation, hybrid modeling, data regions without local data migration possibilities. augmentation) into traditional ones. • Promote interoperability between local data centers • Support the convergence of techniques among and central servers. research groups. Uncertainty Management Uncertainty Management • Support demonstration activities and pilots to link • Develop meta-models. in situ and remote sensing data. • Combine input data selected by AI solutions. • Include estimates of error propagation from the input data to the final output in MRV systems. • Estimate the impact of overestimating and/or underestimating carbon stocks on results-based payments. Standardization and Protocols Standardization and Protocols • Establish a common understanding of how data will • Establish an international framework to adopt be used and processed. standard data management and processing approaches deployed in cloud computing systems • Promote data standards, protection, and security. located in different regions. • Support data policies (for both field and remote • Analyze whether new data and tools could create sensing data) regarding access and sharing. ethical issues in the future, keeping in mind the risk of “dual uses.” ENABLING • Incentivize stakeholders to promote public-private partnerships. • Support policy frameworks for AI solutions and ENVIRONMENT cloud security. • Create key performance indicators. • Foster collaboration among space agencies, • Select a neutral entity to coordinate actions. international organizations, and governments. • Establish cross-communication among experts • Assess capacity-building needs. and users. • Invest in research, training, and knowledge transfer. • Seek public and private funding sources. • Allocate funding to support regular acquisition of • Engage with stakeholders to build confidence UAVs and LiDAR through CEOS. throughout the MRV system. • Improve the MRV process to reduce the time • Foster sustainability. GOAL required for implementation. • Speed up the mobilization of greenhouse gas • Build an operational and sustainable service to IMPACT emission reduction-based payments. carry out carbon stock-based finance at a global scale. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 22 4.1 SHORT-TERM RECOMMENDATIONS mapping and monitoring in support of MRV (for (1—2 YEARS) example, tropical dry forest) that lack the resources for such inventories (Figure 4-2). For example, citizen INTERMEDIATE OUTCOMES science and open, free tools for data collection such as OpenForis15 could contribute to increasing in situ  Data Availability and Access data collection, processing, and sharing in regions Ensuring reliable remote sensing (RS) data inputs with limited resources. We estimate that the cost will require establishing international partnerships of remeasuring a single 1-hectare plot is $2,000— with RS data providers; supporting existing efforts $4,000 if conducted by a local team. This will require in in situ data collection; and building on existing institutional coordination and significant financial and infrastructure. For example, the intergovernmental administrative support from organizations such as Group on Earth Observations (GEO) is playing a key the World Bank, and national governments that have role in promoting open access to data, information, and both the necessary capacity and the resources to knowledge. GEO has contributed to increasing data assist in these efforts and can help develop a Global availability for users and reducing the cost barrier Forest Biomass Reference System. through agreements with key data providers (NASA, Open-data access and suitable data (in terms of ESA, DLR, JAXA) to make some of their data free to use quality, acquisition time, and quantity) for model and share. Further efforts to ensure data availability calibration and validation will directly impact the should warrant that existing agreements—for example, development and performance of geostatistical with JAXA—are fulfilled, and new arrangements (GS) models and artificial intelligence (AI) solutions. with other commercial service providers are sought The deployment of algorithms and AGB estimates (GEO 2020). An example of such an agreement is the using big data should be carried out through cloud collaboration between Kongsberg Satellite Services computing (CC) systems to speed up the MRV (KSAT), Planet, and Airbus with Norway’s Ministry of processing chains. Centralized and distributed clouds Climate and Environment, to provide access to high- have the potential to speed up AGB estimates; improve resolution imagery for monitoring the tropics in order MRV implementation; allow reproducibility; and to curb deforestation. The contract awarded was reduce implementation time. During the course of the approximately €37 million euros, which is comparable virtual workshop, experts stated that the availability to the sum estimate for the development of a Global of open-source clouds is an important requirement Forest Biomass Reference System (CEOS 2020). for establishing a methodological framework using Ground-data sampling campaigns should foster CC systems. Therefore, short-term actions should partnerships with research groups and institutions focus on promoting open clouds—which would help that are developing and maintaining the forest plot generate trust, and encourage data sharing—as well networks responsible for in situ data collection and as building on existing platforms. curation—such as ForestPlots.net, SEOSAW, AfriTRON, CTFS-ForestGEO, and RAINFOR—to make some of their forest inventory data publicly available and contribute to enhancing them with data from other regions. Granting access to repeat inventories conducted in the periods 2005—2010, 2015, and more recent plots, which will ensure an overlap with key spaceborne missions, would be particularly relevant for calibrating and validating RS-based methodologies. Emphasis should be placed on 1-hectare sampling plots to ensure spatial matching between RS pixels and in situ samples. In addition, we recommend addressing efforts toward not only updating out-of-date inventories in areas with limited capacity for establishing remeasurement programs, but also supporting new initiatives to monitor forest structure in regions with clear opportunities for aboveground biomass (AGB) 15 http://www.openforis.org Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 23 Figure 4-2 Potential Biomass Reference Measurement Sites to Set Up Increase Confidence in Biomass Estimates Note: These will need to be re-censused since they will be outdated by 2022/2023 (CEOS 2020).  Processing and Computational Performance information. Finally, convergence of techniques between research groups (RS, AI/computer vision, The short-term recommendations on the automation GS, CC) would enable the enhancement of the tools that of analysis through GS approaches are based on have been developed. These interactions will need to selecting state-of-the-art, and preferably open-source, be fostered by end users who are interested in such GS software that should be enhanced in computational applications. efficiency, parallelized for speed, and ported onto CC systems to address large-scale applications. There are Operational and in-development examples of centralized also additional developments and trends in the field platforms for retrieving satellite data, such as the of AI (for example, hybrid modeling frameworks) System for Earth Observations, Data Access, Processing that could be incorporated in future methodological and Analysis for Land Monitoring (SEPAL), which is frameworks to design AI components with high hosted by the United Nations’ Food and Agriculture performance, and with a broader range of applicability, Organization (FAO), harnesses cloud-based capabilities in order to improve implementation of the MRV process. and modern geospatial data infrastructures like Google Input data aided by automatic and self-evolving AI Earth Engine (GEE), allowing users to access and configuration components could be useful for building process satellite imagery to monitor forests. NASA and dynamically modified systems, constraints, parameters, ESA are currently building the Multi-Mission Algorithm and estimations, leading to an enhanced system. Data and Analysis Platform (MAAP), which has features augmentation could be used as a promising approach similar to GEE, and the Copernicus Data and Information for increasing the supply of information to AI algorithms Services platform (DIAS), which are also targeted as long as they do not incorporate further uncertainties toward monitoring forests. MAAP aims to improve the into the system. This implies that methodologies such ease of access to huge amounts of data that are or will as image inversion/flipping could be included, but be available using CC resources. This platform will those based on synthetically created information should enable users to develop code, analyze results, and share be implemented only with caution. Environmental and process global-scale data, which in turn should and land-based conditions are considerably dynamic. increase reproducibility and transparency. However, This variability has to be included in nontraditional there are still issues surrounding the availability of components in order to produce trustworthy solutions. input data, both for creating and parameterizing existing Furthermore, AI is well recognized for making inferences models for predicting AGB, or AGB change, many of based on “what is currently true,” but if the situation which require regional calibration, or lack up-to-date is changing dynamically, the system has to develop information. Therefore, the first step required is the alternatives to cope with recognized uncertainties establishment of partnerships to ensure RS and in situ or incorporate them into the system as a new set of data access and their integration into open-access cloud Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 24 systems. This system will ideally retrieve RS and in situ regulations through policies and professional data from various providers and will use AI solutions initiatives will help create a common understanding and geostatistical approaches to estimate global and of how data will be used. annual carbon maps in tropical regions (see examples in Appendix C). The need to store large volumes of ENABLING ENVIRONMENTS data (for example, satellite imagery of time-series)  Policies and Regulations makes centralized clouds the best option for avoiding big data migration among servers. Also, The implementation of REDD+, globally, is guided by the lack of stable connectivity and bandwidth a series of overarching policies, including the Paris in many parts of the world makes the option of a Agreement and Agenda 2030. Still, to enable the centralized cloud system the most plausible one in development of a Global Forest Biomass Reference the short term. System, and to leverage the discussed technologies, specific policy and regulation requirements regarding  Uncertainty Management the use of these technologies need to be taken into It is recommended that the implementation of consideration. These include: geostatistical methods be approached through · Revisions to the decision 11/CP19 on the modalities demonstration activities and pilots to link in situ for forest-monitoring systems under the Warsaw and RS data. These methods should account for spatial Framework regarding the need to use the latest correlation; differences in spatial resolutions (and technologies, and cooperation between national and possible misalignments) between in situ and remotely international agencies in order to harmonize data sensed data; errors in reference and other data; and needs for reporting to the UNFCCC. Some steps spatial heterogeneity in attribute values when scaling have already been taken, such as the introduction to large regions, due to different biomes. It is also of guidance for the use of allometric models and recommended to employ geostatistical simulation biomass maps that were included in the 2019 algorithms to propagate spatial uncertainty Refinement to the 2006 IPCC Guidelines for National stemming from various sources and steps in the GHG Inventories (IPCC 2019); and MRV process to the final carbon estimates reported. · Development of policies to improve compliance and  Standardization and Protocols voluntary carbon standards to encourage the use of disruptive technologies for forest monitoring. Data migration and storage should be accompanied by privacy and security protocols (for example, different The following actions are recommended in order to create levels of access to the platform, and the adoption of an environment that can promote the development of a international regulations) to establish standard data global forest biomass monitoring system: protection guidelines, including security strategies, · Adoption of standard operating procedures for in depending on the system architecture and the license situ and remote sensing data for forest monitoring. to use national data sets and added-value products · Development of policies for aligning in situ and RS developed in the platform. data. Improved model performance will go hand in hand with · Development of a policy for data access and sharing extended periods of development and testing phases among public and private institutions. following trustworthy AI solutions, which should be · Development of legal provisions on the use and “lawful (respecting all applicable laws and regulations), validity of spatial data. ethical (respecting ethical principles and values), and robust (both from a technical perspective while  Institutions and Stakeholders taking into account its social environment)” (European Institutions and stakeholders play a key role in providing Commission 2019). In particular, human intervention in the regulations, funding mechanisms, and training supervised approaches will likely be required for reasons activities and education needed to implement new of assessment and transparency (because people technologies and ensure future sustainability. They might not believe the full black box). Therefore, broadly need to be incentivized to collaborate and communicate, accepted guidelines should be added to the development while also being held accountable for outcomes. and deployment of any AI component. Promoting Incentivization can be achieved through financial and developing data management protocols and rewards, but also, and very importantly, through Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 25 well-developed key performance indicators. must be established between key private and public Incentives will be particularly important for piloting players in order to enable knowledge transfer new technologies at the national and international and develop operational solutions following the levels. Setting up an environment in which people recommendations proposed in this report. Some and institutions can contribute to something that of the key players in the field of standardization and they consider to be useful will result in a win-win software development, guidelines, and initiatives are situation. Furthermore, if a clear purpose and the International Organization for Standardization (ISO), framework for sharing data is provided, it can help ClimateAI, the European AI Alliance, the World Economic create openness as people become more willing to Forum Global AI Action Alliance, the Global Partnership share data. Providing rewards for contributing data on AI, the Partnership on AI, the International Association is a great example of incentivization. The reward of Mathematical Geosciences (IAMG), and geoENVia.16 does not need to be financial, but it can still be very The development of AI solutions in a cloud environment interesting for countries if, for example, they can should also consider the following key relevant industry access additional data, and if their institutions partners with specific products: Google, Microsoft, IBM, trust the organization holding the data. This could Amazon, and the Environmental Systems Research be fostered through public-private partnerships. Institute (ESRI). However, open source (data, algorithms, and cloud)  Capacity and Information is crucial to overcoming data-sharing reluctance. Cloud computing as both infrastructure and Additionally, the architectural design and the combination platform is the core system through which the other of software and hardware needed to process big data three technologies could be implemented through should be selected in such a way as to improve the several major steps. computational cost of the analyses, not only to improve MRV implementation, but also to reduce the associated Cloud-based data catalogs have to be compiled or carbon footprint derived from it. Open-source tools retrieved from public and private Earth observation usually include tutorials and documentation for self- (EO) data providers (NASA, ESA, JAXA, DLR). learning. However, designing field data sampling Therefore, partnerships with these agencies and the development of new algorithms tailored and intergovernmental institutions, and other to monitoring specific environmental issues will international organizations such as GEO and FAO likely require technical training. As with geostatistics, (for example, the Global Forest Observation Initiative) the removal of barriers in the use of AI solutions may should be fostered. be achieved by supporting communication between In situ data sets have to be used in combination with RS experts and users working in these domains. data and highlighted technologies in order to calibrate  Finance and Sustainability and validate models for biomass estimation. However, there is a lack of data in many tropical regions because Financing could be made available to open satellite of remoteness, lack of capacity, paucity of data, or armed data archives and could even fund new missions as conflicts (Rodríguez-Veiga et al.. 2017). Therefore, carbon well as in situ data campaigns. This would remove maps in many regions will rely on a combination of barriers in the way of developing a consistent innovative technologies and allometric models that are carbon measurement strategy. All in all, setting up a representative of the forests of that region. Updating Global Forest Biomass Reference System (for example, forest inventories and supporting existing and new a network of 100 biomass reference measurement initiatives for in situ data collection will increase the sites, plus 210 additional distributed sites) will require quantity of calibration and validation samples. There significant investment—about 34 million euros, or $41 are international initiatives already in place to million for in situ data collection, personnel time, airborne coordinate in situ data collection and harmonization campaigns, data curation, and processing over a 5-year that could contribute to establishing partnerships period (CEOS 2020). Other estimates indicate that in for migrating in situ data to this system—for order to fully re-census 600 1-hectare plots across all example, AfriTRON, CTFS-ForestGEO, ForestPlots, and four tropical continents would require 9 million euros RAINFOR. Remote sensing and in situ data could be ($11 million) per remeasurement cycle (Chave et al.. integrated into AI solutions and geostatistical models 2019). run in a cloud environment. But to do so, collaboration 16 ClimateAI (https://climate.ai/), European AI Alliance, the World Economic Forum Global AI Action Alliance (https://www.weforum.org/projects/global-ai-action-alliance), the Global Partnership on AI (https://gpai.ai/), the Partnership on AI (www.partnershiponai.org), the International Association of Mathematical Geosciences (IAMG -- https://www.iamg.org/), with key players in geostatistics among its membership; and geoENVia (https://geoenvia.org), an association promoting the use of geostatistics for environmental applications. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 26 For any system to be implemented, and to be 4.2 LONG-TERM RECOMMENDATIONS sustainable beyond the time frame of an intervention, (3—5 YEARS) adequate financing is needed. In general, financing for digital MRV will come from demonstrating its INTERMEDIATE OUTCOMES value, and both public and private sources will be  Data Availability and Access required. Long-term remote sensing (RS) data acquisition should · Public Sources focus on exploring opportunities and supporting plans There are numerous potential sources of public finance for future follow-up to the GEDI and BIOMASS to support innovation in forest MRV. At the multilateral missions, and/or plans for other missions with levels, the Green Climate Fund (GCF) and the Global similar characteristics. By implementing this Environment Facility (GEF) play a strong role and recommendation, the availability of input data will continue to do so. In addition, there is the need with similar features will ensure the relevance of for continued support from bilateral funding such as the geostatistical and AI algorithms developed Norway’s International Climate and Forest Initiative and tailored to these satellite data sets once those (NICFI), the United States Agency for International “explorer” missions reach the end of life. It would be Development (USAID), and the United Kingdom, among essential to ensure and reinforce the continuation others. of international partnerships, making satellite data publicly available. · Private Funding Regarding access to in situ data, our long-term Impact investment, blended finance, and carbon credits recommendations are focused on maintaining and are all possible sources of finance. The voluntary market increasing the Global Forest Biomass Reference might serve as a useful platform (and source of financing) System (a network of 100 biomass reference for pilots, while the cost of maintaining technologies over measurement sites, plus 210 additional distributed the long term needs to be factored into the products in sites), which as discussed earlier, would require order to ensure the sustainability of the solutions. In this about 34 million euros for in situ data collection, regard, technology companies have a strong role to play personnel time, airborne campaigns, and data curation in providing computational power, software, and training and processing over a five-year period (CEOS 2020). capacity.  Processing and Computational Performance  Social, Cultural, and Behavioral Factors To resolve cases in which in situ data-sharing challenges Effective implementation relies on confidence being cannot be easily overcome, edge computing has been built throughout the MRV system. This means that the proposed as an emerging cloud computing architecture stakeholders and/or beneficiaries need to know that their for applications at large scale. Within this architecture, data will not be used against them. The system must be nonsharable in situ data could be processed and reliable, and it must ensure continuity over a number of analyzed in local data centers in a similar manner years so that users and data providers consider it worthy to the one in a centralized cloud in this case, each local of investing the required resources. data center could act as a micro-cloud where analysis and results would be computed locally. Distributed systems enable quicker responsiveness and processing, lower network traffic, and facilitate real-time monitoring. Local data centers should therefore be created to store privacy-protected data and run relevant AGB- related analytics. In addition, interoperability with the centralized cloud needs to be maintained.  Uncertainty Management As previously recommended, once a well-established and sustainable plot network has been developed, long-term measures to automate processing through geostatistics could evolve toward “meta- Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 27 models” and linking GS (or other) model parameter and tools creating ethical issues in the future should estimates across different plots, conditions, data types, be analyzed, and thought should be given to possible accuracies, and modeling approaches to actual carbon “dual uses” before the technology and data become predictions and their quality; this would improve feature operational. For example, while satellites do not engineering, machine-learning approaches, and model recognize facial images, drones can and do; this generates development in GS and AI. In addition, multiple-point even higher risks and highlights the need for policies geostatistics could enhance the quantification on spatial data privacy. In addition, the application of of spatial patterns from training images for AGB the selected algorithms should follow broadly accepted estimation by combining various sources of satellite guidelines. Many organizations have started to create AI imagery, possibly selected via AI solutions. regulations to avoid negative consequences related to the use of AI: For example, the European Commission Quantifying the (possibly monetary) impact of has issued an Ethics Guide based on seven requirements: overestimating versus underestimating the true transparency, explainability, safety, fairness, human carbon amount is recommended; this will set the rights, privacy, and security (European Commission 2019). stage for evaluating the suitability of data sources for It should be noted that negative consequences are linked computing emission reduction-based payments and to someone in the value chain being able to make certain improving uncertainty management. In addition, the kinds of decisions, while ethical risks are simply related development of a set of prototype case studies to the use of very high-resolution imagery. Confidence involving GS models will help showcase the benefit can be built through policy frameworks such as the of improving uncertainty management in the MRV Cloud Security Alliance (CSA), with various levels of cycle . certification and trust.  Standardization and Protocols  Institutions and Stakeholders Establish an international framework to adopt Coordination between satellite and ground inventory standard data management and processing in cloud systems at the technological level is one of the key computing systems located in different regions. needs, enabling near real-time data collection; the Some specific areas for which guidelines should be reduction of uncertainty around the interpretation of that considered include resilience to attack and security; data; and therefore, higher confidence in the estimates accuracy, reliability, and reproducibility; data protection from satellite-based methods. This will require close and security (authentication, encryption); quality and data collaboration among the space agencies, national integrity; traceability, explainability, accessibility, and governments, and international agencies that universal design; stakeholder participation; auditability; are funding the in situ data collection programs. readiness for tackling specific problems; and sustainable International initiatives such as the Global and environmentally friendly solutions for fostering Forest Observations Initiative (GFOI) can play computational efficiency. a role in engaging countries, always taking into ENABLING ENVIRONMENTS consideration each country’s circumstances.  Policies and Regulations On another front, long-term recommendations include stakeholders’ engagement at early stages of Exploring the interaction of the above actions with AI deployment tools so that they understand the some of the more specific REDD+ policies will also be limitations of the system (independently of the extent of important. For example: the foreseen application of the system) and promoting the incorporation of scientific advances ready for · Country-level REDD+ readiness policies, strategies, operationalization (for example, incorporate auto, self- and regulations; evolving/automatic configuring AI components once they · Other sector-specific policies and regulations; are considered mature for deployment). · Free and prior informed consent; and · Environmental and social safeguards. Moreover, as increasing amounts of data become available, facilities provided by the private cloud sector Furthermore, the ethics of data and usage is always will become more important for data storage and data a concern. Although the current resolution of satellite processing and analytics. The success of innovations data does not specifically enter into the ethical risk zone in the development of a global forest biomass (for example facial recognition) for non-MRV purposes, monitoring system will require collaboration it is recommended that the possibility of new data Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 28 between space agencies, international institutions investment in research, training, and knowledge such as the FAO and the World Bank, and national transfer across the industrial, academic, and end- partners. Participants in the workshop recommended user communities. that to increase confidence in MRV data, there needs There are capacity-building needs associated with to be a perceived neutral entity that takes charge of all of the presented technologies: the process and, specifically, of the quality control of the in situ data, making it available and accessible · Remote sensing. Capacity building is currently to the global community, similar to the World leaning on IPCC Good Practice Guidance for Land Meteorological Organization network. An important Use, Land-Use Change and Forestry (GPG-LULUCF); driver will be increasing local capacity for data analysis Agriculture, Forestry and Other Land Use (AFOLU) and allowing the analysis to be co-located with data guidelines; and GFOI Methods and Guidance on the collection, while supporting country-level policy makers. use of biomass maps. However, these guides are very generic and hard to apply given the highly Greater decentralization seems to spur the use of new diverse country contexts. While most countries data and technologies. However, with these opportunities now maintain a National Forest Inventory (NFI), there are also challenges for REDD+ countries. there is great variation among the countries, due to Implementation in countries with higher degrees of different definitions of forests and biomass, varying decentralization tends to generate better outcomes, as data availability, and the reliability of in situ data, as the challenges for digital MRV and potential solutions well as different country needs. Initiatives such as tend to be locality-specific and require knowledge of NASA and ESA’s collaboration on the Multi-Mission the places in which they are to be implemented. Also, Algorithm and Analysis Platform (MAAP) requires places with greater decentralization often have greater in situ data in order to develop the algorithms and local capacity and the autonomy to deliver. However, validate the products, without which data cannot be decentralization efforts require higher degrees and trusted by policy makers. Local institutions need volumes of capacity to be built, especially with newer to be willing to share their data, but open-source technologies, but usually there are only a limited solutions are usually hard to enforce, especially amount of people at the local level and they are already when incentives are insufficient, or lacking overburdened with an increasing number of mandates altogether. and responsibilities.  Capacity and Information · Artificial intelligence. In the AI realm, technologies that are able to accelerate and cover topics There are significant needs in data collection, storage, concerning policy (including the sharing of retention, security, interoperability, portability, and information) and human resources, such as training interconnectivity between systems. To address and tackling current needs in the sector, should these needs, capacity development is the main be prioritized. recommendation that was suggested by many of the participants in the workshop. Capacity needs to be · Cloud computing (CC). In the CC space, the need developed or improved, especially at the national and to move computation closer to data generation subnational levels. The selection and tuning of AI models will create capacity challenges. Even when there are both highly skilled tasks involving experts who may are policy frameworks (such as the Cloud Security require capacity building in forest and environmental Alliance, data requires a full process before it is departments in order for local implementation to be ready for analysis: It needs to be harmonized and effective, without having to outsource these services to placed in the cloud; then training samples must external experts (who may themselves not understand be generated and quality controlled; finally, data the local environmental context). It is recommended sets need to be trained using machine learning that a full data and capacity-building needs or AI, and methods for post-processing the data assessment based on the identified target audiences need to be developed. Therefore, there is a clear and stakeholders be carried out before developing need for standard operating procedures on the a complete strategy. This assessment could start preparation of remote sensing and in situ data with lessons learned from previous programs by NASA, from various sources, and its sharing across ESA, and other national agencies. In addition, cutting- platforms. Critically, local expertise needs to be edge approaches, including the development of built, and stakeholders need to be consulted, which new algorithms and methodologies, will require Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 29 will require financial commitments. Finally, but  Social, Cultural, and Behavioral Factors importantly, the allocation of resources, capacity building, and training activities are needed to At the global level, lessons learned from the build, maintain, and operate data centers within mobilization of demand for deforestation-free supply distributed systems in some regions. chains can also be drawn, and synergies derived for implementation in the carbon measurement  Finance and Sustainability and MRV space. In fact, investors managing It is important to ensure sustainability in the approximately $6 trillion are increasingly demanding longer term, which can only happen if there is an a wide range of research, analysis, benchmarking ongoing, adequate understanding of the value tools, and best-practice materials that will help of the technologies and the embedding of these them understand and manage the risks and technologies in systems owned by key decision- opportunities associated with intensive agriculture. makers in the countries. Funding should be Such powerful investor groups wield powerful allocated to support the regular acquisition influence levers on national policy.17 Therefore, of unmanned aerial vehicles (UAVs)/aerial engagement with them is key. LiDAR in situ data through the Committee on Earth Observation Satellites. The cost of such acquisitions will vary; however, a reasonable guide would be $250–$600 per square kilometer, with UAV acquisitions tending toward the lower end, and airborne (that is, airplanes) the upper end.  17 https://www.fairr.org/research/ Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 30 5. REFERENCES Albinet, C. et al.. 2019. “A Joint ESA-NASA Multi-Mission Algorithm and Analysis Platform (MAAP) for Biomass, NISAR, and GEDI.” Surveys in Geophysics. 40 (4): 1017—27. doi: 10.1007/s10712-019-09541-z. Asner, G. P. et al.. 2012. “High-Resolution Mapping of Forest Carbon Stocks in the Colombian Amazon.” Biogeosciences. 9 (7): 2683—96. doi: 10.5194/bg-9-2683-2012. Asner, G. P. and J. Mascaro. 2014. “Mapping Tropical Forest Carbon: Calibrating Plot Estimates to a Simple LiDAR Metric.” Remote Sensing of Environment. 140: 614—624. doi: 10.1016/j.rse.2013.09.0 Avitabile, V. et al.. 2016. “An Integrated Pan-Tropical Biomass Map Using Multiple Reference Datasets.” Global Change Biology. 22 (4): Art. no. 4. doi: 10.1111/gcb.13139. Baccini, A. et al.. 2012. “Estimated Carbon Dioxide Emissions from Tropical Deforestation Improved by Carbon- Density Maps.” Nature Climate Change. 2 (3): 182—85. doi: 10.1038/nclimate1354. Baccini, A., W. Walker, L. Carvalho, M. Farina, D. Sulla-Menashe, and R. A. Houghton. 2017. “Tropical Forests Are a Net Carbon Source Based on Aboveground Measurements of Gain and Loss.” Science. 358 (6360): 230—234. doi: 10.1126/science.aam5962. Bouvet, A. et al.. 2018. “An Above-Ground Biomass Map of African Savannahs and Woodlands at 25 m Resolution Derived from ALOS PALSAR.” Remote Sensing of Environment. 206: 156—173. doi: 10.1016/j.rse.2017.12.030. CEOS. 2020. “Global Aboveground Biomass Product Validation Best Practice Protocol.” doi: 10.5067/doc/ceoswgcv/ lpv/agb.001. CEOS. 2021a. “CEOS Data Policy.” https://ceos-cove.org/en/data_policy/ (accessed Jan. 27, 2021). CEOS 2021b. “CEOS FDAI: Future Data Architectures Inventory.” http://ec2-3-208-162-171.compute-1.amazonaws.com/fdai/ (accessed Jan. 28, 2021). CEOS 2021c. “Home | GCOS.” https://gcos.wmo.int/en/home (accessed Feb. 15, 2021). Chave, J. et al.. 2019. “Ground Data Are Essential for Biomass Remote Sensing Missions.” Surveys in Geophysics. 40 (4): 863—80. doi: 10.1007/s10712-019-09528-w. Csillik, O., P. Kumar, J. Mascaro, T. O’Shea, and G. P. Asner. 2019. “Monitoring Tropical Forest Carbon Stocks and Emissions Using Planet Satellite Data.” Scientific Reports. 9 (1): 1—12. doi: 10.1038/s41598-019-54386-6. Dubayah, R. et al.. 2020. “The Global Ecosystem Dynamics Investigation: High-Resolution Laser Ranging of the Earth’s Forests and Topography.” Science of Remote Sensing. 1 (January), Art. no. doi: 10.1016/j.srs.2020.100002. Duncanson, L. et al.. 2020. “Biomass Estimation from Simulated GEDI, ICESat-2 and NISAR Across Environmental Gradients in Sonoma County, California.” Remote Sensing of Environment. 242 ( March), Art. no. doi: 10.1016/j. rse.2020.111779. European Commission. 2014. Open Science: Shaping Europe’s Digital Future. https://ec.europa.eu/digital-single- market/en/open-science (accessed Feb. 15, 2021). Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 31 European Commission. 2019. Ethics Guidelines for Trustworthy AI: Shaping Europe’s Digital Future. https:// ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai (accessed Feb. 22, 2021). FAO. 2009. “Assessment of the Status of the Development of the Standards for the Terrestrial Essential Climate Variables BIOMASS.” Accessed Feb. 09, 2021. http://www.fao.org/3/i1238e/i1238e00.htm. FAO. 2013. National Forest Monitoring Systems: Monitoring and Measurement, Reporting and Verification (M & MRV) in the Context of REDD+ Activities. Geneva: UN-REDD Programme Secretariat. http://www.fao.org/3/bc395e/bc395e.pdf FCPF. 2020. FCPF 2020 Annual Report. Accessed Jan. 27, 2021. https://www.forestcarbonpartnership.org/ document/fcpf-2020-annual-report. GCOS. 2021a. “Essential Climate Variables.” https://gcos.wmo.int/en/essential-climate-variables/about accessed Feb. 15, 2021). Hansen, M. C., P. Potapov, and A. Tyukavina. 2019. Comment on “Tropical Forests Are a Net Carbon Source Based on Aboveground Measurements of Gain and Loss.” Science. 363 (6423): 3629. Jan. 2019, doi: 10.1126/science. aar3629. Herold, M. et al.. 2019. “The Role and Need for Space-Based Forest Biomass-Related Measurements in Environmental Management and Policy.” Surveys in Geophysics. 40 (4): 757-78. Jul. 2019, doi: 10.1007/s10712-019- 09510-6. IPCC. 2003. IPCC Good Practice Guidance for Land Use, Land-Use Change and Forestry. Japan. https://www.ipcc-nggip.iges.or.jp/public/gpglulucf/gpglulucf_files/GPG_LULUCF_FULL.pdf IPCC. 2019. “Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories—IPCC. Accessed: Feb. 23, 2021. Available: https://www.ipcc.ch/report/2019-refinement-to-the-2006-ipcc-guidelines-for-national- greenhouse-gas-inventories/. Jucker, T. et al.. 2017. “Allometric Equations for Integrating Remote Sensing Imagery into Forest Monitoring Programmes.” Global Change Biology. 23 (1): 177—190. doi: 10.1111/gcb.13388. Kumar, L. and O. Mutanga. 2017. “Remote Sensing of Above-Ground Biomass.” Remote Sensing. 9 (9): doi: 10.3390/ rs9090935. Lang, N., K. Schindler, and J. D. Wegner. 2019. “Country-Wide High-Resolution Vegetation Height Mapping with Sentinel-2.” Remote Sensing of Environment. 233 (July): 111347. doi: 10.1016/j.rse.2019.111347. Lefsky, M. A. 2010. “A Global Forest Canopy Height Map from the Moderate Resolution Imaging Spectroradiometer and the Geoscience Laser Altimeter System. Geophysical Research Letters. 37 (15). doi: https://doi. org/10.1029/2010GL043622. Le Toan, T. et al.. 2011. “The BIOMASS Mission: Mapping Global Forest Biomass to Better Understand the Terrestrial Carbon Cycle.” Remote Sensing of Environment. 115 (11): 2850-60. doi: 10.1016/j.rse.2011.03.020. Mitchard, E. T. A. et al.. 2014. “Markedly Divergent Estimates of Amazon Forest Carbon Density from Ground Plots and Satellites. Global Ecology and Biogeography. 23 (8): 935-46. doi: 10.1111/geb.12168. McNicol, I. M., C. M. Ryan, and E. T. A. Mitchard. 2018. “Carbon Losses from Deforestation and Widespread Degradation Offset by Extensive Growth in African Woodlands.” Nature Communications 9: Art. no. 3045, 1-19. doi: 10.1038/s41467-018-05386-z. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 32 Minh, D. H. T. et al.. 2016. “SAR Tomography for the Retrieval of Forest Biomass and Height: Cross-Validation at Two Tropical Forest Sites in French Guiana.” Remote Sensing of Environment. 175: 138-147. doi: 10.1016/j. rse.2015.12.037. Patterson, P. L. et al.. 2019. “Statistical Properties of Hybrid Estimators Proposed for GEDI: NASA’s Global Ecosystem Dynamics Investigation.” Environmental Research Letters. 14 (6). doi: 10.1088/1748-9326/ab18df. Ploton, P. et al.. 2017. “Toward a General Tropical Forest Biomass Prediction Model from Very High Resolution Optical Satellite Images.” Remote Sensing of Environment. 200 (August), 140—153, doi: 10.1016/j.rse.2017.08.001. Qi, W., S. Saarela, J. Armston, G. Ståhl, and R. Dubayah. 2019. “Forest Biomass Estimation Over Three Distinct Forest Types Using TanDEM-X InSAR Data and Simulated GEDI LiDAR Data.” Remote Sensing of Environment. vol. 232 (July), 111283. doi: 10.1016/j.rse.2019.111283. Rodríguez-Veiga, P., J. Wheeler, V. Louis, K. Tansey, and H. Balzter. 2017. “Quantifying Forest Biomass Carbon Stocks from Space.” Current Forestry Reports. 3 (1): 1-18. doi: 10.1007/s40725-017-0052-5. Rodríguez-Veiga, P. et al.. 2019. “Forest Biomass Retrieval Approaches from Earth Observation in Different Biomes.” International Journal of Applied Earth Observation and Geoinformation. vol. 77 (May), 53-68. doi: 10.1016/j. jag.2018.12.008. Rosen, P. et al.. 2017. “The NASA-ISRO SAR (NISAR) Mission Dual-Band Radar Instrument Preliminary Design.” 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 3832-35. Fort Worth, TX. USA, 2017. doi: 10.1109/IGARSS.2017.8127836. Saatchi, S. S. et al.. 2011. “Benchmark Map of Forest Carbon Stocks in Tropical Regions Across Three Continents.” Proceedings of the National Academy of Sciences. 108 (24): 9899-9904. doi: 10.1073/pnas.1019576108. Santoro, M. et al.. 2020. “The Global Forest Above-Ground Biomass Pool for 2010 Estimated from High-Resolution Satellite Observations.” Earth System Science Data Discussions. July, 1—38. doi: 10.5194/essd-2020-148. Scharlemann, J. P., E. V. Tanner, R. Hiederer, and V. Kapos. 2014. “Global Soil Carbon: Understanding and Managing the Largest Terrestrial Carbon Pool. Carbon Management. 5 (1): 81-91. doi: 10.4155/cmt.13.77. Stavros, E. N. et al.. 2017. “ISS Observations Offer Insights into Plant Function.” Nature Ecology & Evolution 1 (7): 194. doi: 10.1038/s41559-017-0194. UNEP. 2019. “Peatlands Store Twice As Much Carbon As All the World’s Forests.” UN Environment. http://www. unenvironment.org/news-and-stories/story/peatlands-store-twice-much-carbon-all-worlds-forests (accessed Jan. 27, 2021). UNFCCC. 2021. “What Is REDD+?” https://unfccc.int/topics/land-use/workstreams/redd/what-is-redd (accessed Jan. 27, 2021). UN-REDD. 2009. Measurement, Assessment, Reporting and Verification (MARV): Issues and Options for REDD. Draft Discussion Paper, United Nations Collaborative Programme on Reducing Emissions from Deforestation and Forest Degradation in Developing Countries (UN-REDD). Geneva. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 33 APPENDIX A: WORKSHOP PARTICIPANTS LIST The names of presenters and experts providing substantial feedback have been bolded and underscored. Name Affiliation Sector Aguilar-Amuchastegui Naikoa World Wide Fund for Nature (WWF) Remote Sensing, Geostatistics, and REDD+ MRV Asiyanbi Adeniyi BIOSEC Policy Matters Lancaster Environment Centre at the Remote Sensing, Artificial Intelligence, Machine Atkinson Peter University of Lancaster  Learning and/or Big Data and Geostatistics Artificial Intelligence, Machine Learning and/or Big Baydin Atilim Gunes University of Oxford Data Benito Pablo Llopis South Pole Policy Matters and Project implementation Remote Sensing, Artificial Intelligence, Machine Benjamins Richard Telefónica Learning and/or Big Data and Policy matters Brooks Chris Oxford Policy Management Policy Matters Cabezas Antonio Tabasco GMV Aerospace and Defense Remote Sensing Remote Sensing, Artificial Intelligence, Machine Camara Gilberto Group on Earth Observations Learning and/or Big Data, Geostatistics, Cloud Computing and Policy Matters Remote Sensing, Artificial Intelligence, Machine Camps-Valls Gustau University of Valencia Learning and/or Big Data and Geostatistics  Carr Edward Clark University Climate Change Castren Tuukka World Bank Forest Data Chiappe Federica Federica Chiappe Consulting Policy Chua Darryl Temasek Policy Matters Cooke Katherine Oxford Policy Management Policy Matters Laboratory of Geo-Information Science Remote Sensing, Artificial Intelligence, Machine De Bruin Sytze and Remote Sensing at Wageningen Learning and/or Big Data and Geostatistics University De Grandi Elsa Carla GMV Aerospace and Defense Remote Sensing Di Gregorio Monica University of Leeds Policy Matters Remote sensing, Artificial Intelligence, Machine Disney Mat University College London Learning and/or Big Data Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 34 Artificial Intelligence, Machine Learning and/or Big Dumra Bidyut DBS Bank Data Duncanson Laura University of Maryland Remote Sensing Dutta Omjyoti GMV Aerospace and Defense Remote Sensing / Cloud Computing Espejo Andrés World Bank Carbon Finance / Forestry Fernandes Erick World Bank Sustainable Development Flasher Joe Amazon Cloud Computing Remote Sensing, Artificial Intelligence, Machine Fleming Sam Earth Box Learning and/or Big Data and Cloud Computing Garcia Monica Technical University of Denmark Remote Sensing Garret Keith World Bank Sustainable Development Indian Institute of Technology Remote Sensing, Artificial Intelligence, Machine Ghosh Soumya Kanti Kharagphur Learning and/or Big Data and Cloud Computing Giménez Marta Gómez GMV Aerospace and Defense Remote Sensing Remote Sensing, Artificial Intelligence, Machine Giuliani Gregory University of Geneva / UN Grid-Geneva Learning and/or Big Data and Cloud Computing Gonzalez-Castañé Gabriel Insight Centre for Data AI/Machine Learning Remote Sensing, Artificial Intelligence, Machine Gorelick Noel Google Learning and/or Big Data, Geostatistics, Cloud Computing, and Policy Matters Haas Oliver DBS Bank Finance VTT Technical Research Centre of Remote Sensing, Artificial Intelligence, Machine Häme Tuomas Finland Learning and/or Big Data Harshadeep Nagaraja-Rao World Bank Sustainable Development Laboratory of Geo-Information Science Remote Sensing, Artificial Intelligence, Machine Herold Martin and Remote Sensing at Wageningen Learning and/or Big Data and Policy Matters University Howard Luke Plan Vivo foundation Voluntary Carbon Market Iversen Peter UNFCCC Policy Matters Jiménez Julián Gonzalo World Bank Climate Change Food and Agriculture Organization of the Jonckheere Inge Remote Sensing and Cloud Computing United Nations (FAO) Artificial intelligence, Machine Learning and/or Big Kubasiak Anna Microsoft Data Kyriakidis Phaedon Cyprus University of Technology Geostatistics Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 35 Lacoste Alexandre Element AI Artificial Intelligence Lynch Jim Earth-I Remote Sensing and Policy Matters Remote Sensing, Artificial Intelligence, Machine Institute of Earth Surface Dynamics at Mariethoz Gregoire Learning and/or Big Data, Geostatistics and Cloud the University of Lausanne Computing Martínez Carlos López Polytechnic University of Catalonia Artificial Intelligence and Big Data McNichol Iain University of Edinburgh Remote Sensing Mendes Flavia Remote Sensing Solutions GmbH Remote Sensing Mitchard Edward University of Edinburgh Remote Sensing Nanos Nikos Aristotle University of Thessaloniki Geostatistics Remote Sensing, Artificial Intelligence, Machine Nuño Bruno Sánchez-Andrade Microsoft Learning and/or Big Data; Geostatistics, Cloud Computing and Policy Matters Remote Sensing, Artificial Intelligence, Machine Nussbaum Madlene Bern University of Applied Sciences Learning and/or Big Data and Geostatistics Japan Aerospace Exploration Agency Ochiai Osamu Earth Observation (JAXA) Olofsson Pontus Boston University Earth & Environment Remote Sensing and Geostatistics Paganini Monica World Bank Policy Matters Parisa Zack SilviaTerra Forest Data Pascual Adrián Arizona State University Remote Sensing and Geostatistics Patenaude Genevieve University of Edinburgh Remote Sensing and Cloud Computing Peneva-Reed Ellie World Bank Remote Sensing / Climate Remote Sensing, Artificial Intelligence, Machine Ploton Pierre Institute of Research for Development Learning and/or Big Data and Geostatistics; Earth Observation Prados Ana I. University of Maryland Baltimore County Remote Sensing and Policy Matters Ramage Steven Group on Earth Observations Earth Observation Remote Sensing, Artificial Intelligence, Machine Ramoelo Abel South African National Parks Learning and/or Big Data and Geostatistics Artificial Intelligence, Machine Learning and/or Big Rana Omer University of Cardiff Data, and Cloud Computing Reddy Rama Chandra World Bank Climate Change Reed Bradley United States Geological Survey Remote Sensing Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 36 Romero Beatriz Revilla GMV Aerospace and Defense Remote Sensing Sadler Marc World Bank Climate Funds Schneider Fabian Jet Propulsion Laboratory (JPL) Remote Sensing Scipal Klaus European Space Agency (ESA) Remote Sensing Sebastian Ana GMV Aerospace and Defense Remote Sensing Remote Sensing, Artificial Intelligence, Machine Shapiro Aurelie World Wide Fund for Nature (WWF) Learning and/or Big Data Sinha Chandra Shekhar World Bank Climate Change Staddon Sam University of Edinburgh Remote Sensing Swedish University of Agricultural Remote Sensing, Artificial Intelligence, Machine Stahl Göran Sciences; Department of Forest Resource Learning and/or Big Data and Geostatistics Management Stein Alfred University of Twente Geostatistics CCST do Instituto Nacional de Pesquisas Tejada Graciela Remote Sensing and Earth System Science Espaciais (INPE) Brazil Thau Dave World Wide Fund for Nature (WWF) Big Data and Artificial Intelligence Tolosana Rafael University of Zaragoza Cloud Computing École polytechnique fédérale de Remote Sensing, Artificial Intelligence, Machine Tuia Devis Lausanne (EPFL) Learning and/or Big Data Verhegghen Astrid Joint Research Centre (JRC) Remote Sensing and Cloud Computing Swiss Data Science Center, ETH Zürich Remote Sensing, Artificial Intelligence, Machine Volpi Michele and EPFL Learning and/or Big Data Vyhmeister Eduardo Insight Center for Data AI/Machine Learning Williams Mathew University of Edinburgh Remote Sensing Yagüe Julia GMV Aerospace and Defense Remote Sensing Remote Sensing, Artificial Intelligence, Machine Zhang Yujia Cornell University Learning and/or Big Data and Geostatistics Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 37 APPENDIX B: GLOSSARY OF TERMS Aboveground biomass (AGB) is “all living biomass above the soil including stem, stump, branches, bark, seeds, and foliage” (IPCC, 2003 p. 557, G.2). Belowground biomass (BGB) is “all living biomass of live roots. Fine roots of less than (suggested) 2 mm diameter are sometimes excluded because these often cannot be distinguished empirically from soil organic matter or litter” (IPCC, 2003 p. 558, G.3). Biomass is “the organic material both aboveground and belowground, and both living and dead, e.g., trees, crops, grasses, tree litter, roots etc. Biomass includes the pool definition for above- and below-ground biomass” (IPCC, 2003 p. 558, G.3). Carbo pool is “the reservoir containing carbon.” (IPCC, 2003 p. 559, G.4). Carbon stock is “the quantity of carbon in a pool” (IPCC, 2003 p. 559, G.4). Dead wood “includes all non-living woody biomass not contained in the litter, either standing, lying on the ground, or in the soil. Dead wood includes wood lying on the surface, dead roots, and stumps larger than or equal to 10 cm in diameter or any other diameter used by the country” (IPCC, 2003 p. 562, G.7). Essential climate variable (ECV) is a “physical, chemical or biological variable or a group of linked variables that critically contributes to the characterization of Earth’s climate” (GCOS 2021). Forest reference level (FRL) is “a benchmark for emissions from deforestation and forest degradation and removals from sustainable management of forest and enhancement of forest C stocks” (FCPF 2020, p. 25). Forest reference emission level (FREL) is “a benchmark for emissions exclusively from deforestation and forest degradation” (FCPF 2020, p. 25). The Global Climate Observing System (GCOS) is a “co-sponsored programme which regularly assesses the status of global climate observations and produces guidance for its improvement. It is co-sponsored by the World Meteorological Organization (WMO), Intergovernmental Oceanographic Commission of UNESCO (IOC-UNESCO), United Nations Environment Programme (UN Environment), and International Science Council (ISC). GCOS expert panels maintain definitions of Essential Climate Variables (ECVs). They identify gaps by comparing the existing climate observation system with these ECVs. ECVs are the observations required for systematically observe Earth`s changing climate. The expert panels regularly develop plans on how to sustain, coordinate and improve physical, chemical and biological observations” (CEOS 2021c). Litter “includes all non-living biomass with a diameter less than a minimum diameter chosen by the country (for example 10 cm), lying dead, in various states of decomposition above the mineral or organic soil. This includes litter, fumic, and humic layers. Live fine roots (of less than the suggested diameter limit for belowground biomass) are included in litter where they cannot be distinguished from it empirically” (IPCC, IPCC p. 567). Measurement is “the processes of data collection over time, providing basic datasets, including associated accuracy and precision, for the range of relevant variables. Possible data sources are in-situ measurements, field observations, detection through remote sensing and interviews” (UN-REDD 2009, p. 3). Monitoring a) “is a function of the National Forest Monitoring Systems, which is primarily a domestic tool to allow countries to assess a broad range of forest information, including in the context of REDD+ activities. Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 38 The monitoring function can be implemented through a variety of methods and serve a number of different purposes, depending on national circumstances” (FAO, 2013 p. vi). Monitoring b) is “the need for periodic information on the results obtained through national policies and measures” (FAO, 2013 p. 5). Open science refers to “the way research is carried out, disseminated, deployed and transformed by digital tools and networks. It relies on the combined effects of technological development and cultural change towards collaboration and openness in research” (European Commission 2014). Reporting is “the process of formal reporting of assessment results to the UNFCCC, according to predetermined formats and according to established standards, especially the IPCC Guidelines and GPG. It builds on the principles of transparency, consistency, comparability, completeness and accuracy” (UN-REDD 2009, p. 4). Soil organic matter “includes organic carbon in mineral and organic soils (including peat) to a specified depth chosen by the country and applied consistently through the time series. Live fine roots (of less than the suggested diameter limit for belowground biomass) are included with soil organic matter where they cannot be distinguished from it empirically” (IPCC, IPCC p. 574, G.19). Verification is “the process of formal verification of reports, for example the established approach to verify national communications and national inventory reports to the UNFCCC” (UN-REDD 2009, p. 4). Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 39 APPENDIX C: OTHER PLATFORMS An inventory of platforms can be found on the CEOS webpage (CEOS, 2021 b). Here, we highlight the following examples. Multi-Mission Algorithm and Analysis Platform (MAAP) for Biomass, NISAR, and GEDI NASA and ESA are currently collaborating to build the Multi-Mission Algorithm and Analysis Platform (Albinet 2019). Exponential data growth is a significant factor in the earth sciences and carbon monitoring community with the launch of several high-data-volume missions, including ESA BIOMASS (Le Toan et al.. 2011); NASA-ISRO SAR (NISAR) (Rosen et al.. 2017), and NASA Global Ecosystem Dynamics Investigation (GEDI) (Stavros et al.. 2017), as well as complementary and/or similar missions. Both ESA BIOMASS and NASA-ISRO SAR (NISAR) have planned launches for 2022, while NASA GEDI was launched in December 2018 for a two-year mission, with data being made available to users from late 2019. This platform will enable users to develop code, analyze results, and share and process global-scale data. It is similar to Google Earth Engine (GEE) but targeted to forest communities, which will hold all of the satellite and in situ data using cloud computing resources. However, challenges remain on how to turn observations into actionable products, given that satellites can only measure structure; they cannot measure biomass directly. Where conversions have been carried out, measurements have been affected by large uncertainties, and most contain regional biases, making their use as evidence for result-based payments unsatisfactory. In every case, ground observation data is needed to develop the algorithms and validate the products, without which, data cannot be trusted for policy making. The success of this will require collaboration between space agencies, institutions such as FAO and the World Bank, and national partners; space agencies alone cannot solve this challenge, as they do not have either the expertise or the mandate to establish such a system. GEDI’s OBIWAN One of the objectives of the GEDI mission is to produce estimates of mean biomass with uncertainty on 1 x 1 kilometer grid cells. Using the GEDI data set and footprint-level biomass library, OBIWAN (Online Biomass Inference using Waveforms and iNventory)18 will provide biomass estimates over areas defined by users. Users will be returned a standard carbon report documenting the statistical estimator used, along with query-specific information about sample number and model parameters. OBIWAN is expected to provide critical emission factors for forests under both the REDD+ and IPCC reporting frameworks in many parts of the world. OBIWAN, the first space-based carbon density estimates, is characterized by its level of statistical rigor, and the spatial resolution required for market-based and international carbon accounting. SEPAL The System for Earth Observations, Data Access, Processing and Analysis for Land Monitoring (SEPAL) is an open- source, cloud computing platform developed for the automatic monitoring of land cover. It combines cloud services such as GEE and Amazon Web Services Cloud (AWS) with free software, including geospatial services. The main focus of this platform is on building an environment with previously configured tools and on managing the use of computational resources in the cloud to facilitate ways to search, access, process, and analyze Earth observation data, especially in countries that have difficulties with internet connection and few computational resources. It works as an interface that facilitates access and the integration of other services. 18 https://climate.esa.int/sites/default/files/D1_S1_T6_Healey.pdf Assessment of Innovative Technologies and Their Readiness for Remote Sensing-Based Estimation of Forest Carbon Stocks and Dynamics 40 www.forestcarbonpartnership.org