We are beginning to understand that nature is the foundation upon which the global economy is built. Today, nature degradation and exceeding planetary boundaries in relation to natural capital is recognised as a systemic risk to economic stability and resilience (Steffen et al. 2015) with an estimated $44 trillion of economic value threatened by biodiversity decline and ecosystem collapse – equal to over half of the world’s total GDP (World Economic Forum, 2021). Steffen also emphasises the impact of humanity on biodiversity loss. A steep acceleration in degradation is already occurring: around a million species are currently at risk of extinction due to human activity, with declines in biodiversity threatening over 80% of UN Sustainable Development Goal targets (IPBES, UN, 2019). The European Union recently put in place the first phase of its Sustainable Financial Disclosure Regulation, requiring that asset-management companies provide statutory information about their impact on society and environment. Such reporting will require objective data, information and assessment of the impacts of investee company operations in all their aspects.
In the wake of the United Nations Climate Change Conference (2021), the next 12 months are critical for international environmental policy as governments and businesses orient their activities to achieve global commitments. A critical issue facing global leaders of the private and public sector alike is engineering the systemic change required towards a more sustainable relationship between humanity and our planet; one key lever is changing business decision-making. Commitments, including the Glasgow Financial Alliance for Net Zero (GFANZ), which saw over $130 trillion of private capital committed to transforming the economy towards net-zero, underscore the increasing relevance of private sector investment alongside public policy in achieving positive change (Gfanzero.com, 2021). The financial aspects of GHG emission management and security of biodiversity are also closely linked.

Yet although climate is starting to feature in business strategies, disclosures and investment decisions, biodiversity is still largely absent. The situation is changing as governments and businesses begin to recognise the importance of the issue. For example, the 2020 World Economic Forum Global Risks Report ranks biodiversity loss and ecosystem collapse as one of the top five threats humanity will face in the next decade, based on its potential impact on supply chains, food and health. There is both regulatory and internal pressure on companies to report and reduce their impacts on nature (e.g. through the Taskforce on Nature-related Financial Disclosures, and outputs resulting from the Dasgupta Review. Consumer research in 2019 by FNZ, a major investment platform, revealed that 78% of end-investors are keen to know the environmental and societal impact of their investments; understanding their impact on nature was among their top three concerns.

Unfortunately, biodiversity impact measurement lags far behind other areas, including climate change, with limited data to support investment decisions available, and no trusted standard for biodiversity performance measurement that spans the universe of holdings considered by investors in global capital markets. Yet for planning and action to be effective, the development of science-based metrics enabling investors to integrate nature into their risk and due diligence decision-making is crucial. Measuring biodiversity losses and gains, particularly where data are limited, would support tracking and reporting of the net outcomes of human development and nature restoration. This could guide decision-making at the international and national policy levels, as well as internally by corporates and investors (Bull et al. 2020). Satellite data can be a vital component in providing objective and accessible evidence of the impact of companies operations on biodiversity through, for example, the provision of data from which Essential Biodiversity Variables (e.g. Pereira et al 2013, Pettorelli et al 2016) can be defined and measured.

Leclere et al. (2020) demonstrate that if nature is to be restored, conservation action needs to be supplemented by rapid shifts towards sustainable supply chains; Williams et al (2020) give a similar message focussed on agriculture. Pertinent to reducing future pandemic risk, over half of emerging human infectious diseases are linked to land conversion for agriculture (Rohr et al. 2019), and anthropogenic changes to ecosystems are key drivers of zoonotic disease risk (Allen et al. 2016). This research and more suggests that a robust, standardised approach to measuring and reporting corporate impacts on biodiversity is urgently required.

Sustainability reporting is increasingly used by both investors and customers to gain a better understanding of a company’s non-financial risks and opportunities, and informs investment decision making. However, very few financial institutions manage biodiversity or nature-related risks (Refinitiv 2020), though (Mace et al. 2018) argue that finance and business sectors could become drivers of positive change for biodiversity impact across all aspects of investment. Existing research on corporate accounting and accountability suggests that corporate biodiversity accountability is in its infancy (Adler et al. 2017; Addison, Bull, and Milner-Gulland 2019). Yet systematically mispricing nature has resulted in misallocation of capital, especially in land-based sectors, and has exposed the financial sector to nature-related risks (Refinitiv 2020).

It is widely accepted that measuring changes in biodiversity is more complex than measuring climate change (Dasgupta 2020). However, for corporates and financial institutions, there is an increased interest in the role that AI could play in integrating data on both biodiversity impacts and broader ESG data itself into the investment process, improving sustainability assessment and acting as a catalyst for sustainable investing at scale (Generation Investment Management 2019). Investors are now increasingly looking to apply AI to sustainability reporting, as some investment funds and consultancies, including pension funds such as APG and Pension Danmark, already use machine learning in ESG investing (Dohle 2020). In the context of retail investor decision-making, technological innovations from fund platforms will deliver this information into the hands of public long-term savers for the first time at scale.

With the arrival of GPS, satellite remote sensing, and personal computers, the last three decades have witnessed rapid advances in the field of spatially-explicit ecological modelling. (Roberts et al. 2010). The increasing availability of digital images and other satellite based data, coupled with sophisticated artificial intelligence (AI) techniques, presents an exciting opportunity for biodiversity researchers to create new datasets of species observations (August et al. 2020; Garske et al. 2021). But with this innovation has come complexity. Understanding the complex relationships and processes that regulate the world’s balance of biodiversity impacts and dependencies is critical to our understanding of how to mitigate the grand challenges of environmental sustainability. Studies have considered that machine learning can help with this complex problem. (Talbert et al. 2017)
For example, developments in earth observation data (Talbert et al. 2017) coupled with location data for companies and their supply and value chains can be combined with both machine learning and biodiversity science, surfacing new insights into the impacts and dependencies of companies on nature. This enables impact, risk and value-at-risk to be calculated, alongside relevant qualitative detail such as information on controversial activities. Probabilistic causal chain models may then be applied to link the activities of specific businesses or sectors to biodiversity impact in specific locations (with associated uncertainties). With the integration of additional datasets to link supply chain and spatial data with impact, financial institutions may be able to measure the biodiversity impact and risk associated with a given investee company, alongside making comparisons with peers in a given sector and geography. This can be used for a range of applications, including credit risk pricing, disclosure and reporting, portfolio management, risk management, portfolio stewardship.
The convergence of sustainability and digital imperatives is beginning to gain traction in the private and public sectors (George et al. 2019) but has yet to galvanize systematic and rigorous academic research. Exploratory research has seen entrepreneurial actors employing digital technologies to tackle crucial sustainability challenges. They have done so, not only through technological innovation, such as developing digital sustainability tools, but also by developing business models that infuse innovations with new purpose (George et al. 2019). Trends now combine in the actions of entrepreneurial actors employing digital technologies to tackle crucial sustainability challenges through technological innovation (George et al. 2012, 2019).
In tackling the challenge of accurately mapping nature and biodiversity impacts for financial decision-making, the engineering of highly integrated tools and frameworks will be crucial for the implementation of appropriate investment decisions for both public and private actors. This paper will examine the opportunities for novel combinations of earth observation data and machine learning in this domain, identify challenges encountered with efforts, and chart a path forward. If incorporated into our $120trillion global capital markets, it is these tools which could engender a paradigm shift in capital allocations at scale - one that is net-positive for nature. The outcome is beneficial for all: an improvement in biodiversity and more sustainable, and hence financially successful, decision making in companies.

Detailed and complete data on physical assets are required in order to adequately assess environment-related risk and impact exposure and the diffusion of these risks and impacts through the financial system. Investors need to know where the physical assets (e.g. power plant, factory, farm) are located of companies in their portfolios, and what their polluting characteristics are. This is essential to manage these environment-related risks and to channel investments to more sustainable alternatives. At present, data on physical assets is typically incomplete, inaccurate or not released in a timely manner. As a result, key stakeholders including asset owners, asset managers, regulators and policymakers are frequently forced to make crucial decisions with incomplete information.

Accurate and comprehensive global asset-level databases are a prerequisite for meaningful innovation in green and digital finance. They provide the link between the financial system and the “real economy” and allows the wealth of EO datasets and insights that we have available to be made actionable for sustainable finance decision making.

We created a framework to derive a global database of pollutant plants, such as cement, iron and steel, which represent about 15% of the global CO2 emissions. Our solution makes use of state-of-the-art deep learning architectures coupled with Earth observation data. In this talk we will show the technical journey, from the initial feasibility study to the deployment at scale, which brought to the definition of 3,117 cement production assets, accounting for ~90% of global cement production capacity, and 1,598 iron and steel assets, accounting for ~70% of global crude steel production. Specifically, the solution is based on a two-staged approach. Convolutional neural networks are initially used on open access EO data in both a computer vision approach with Sentinel-2 data and a physics-aware one with Landsat-8. This gives a first broader indication of the location of the assets. The precise characterisation is then performed with high-resolution EO imagery from the WorldView missions through collective intelligence. This approach has been tested against a pure machine learning one on both medium-resolution and high-resolution EO data and found of comparable or better accuracy. Several parameters are inferred with this approach: the precise localisation of the assets, their production capacity, the plant type and the production type. An extensive manual validation concluded that the production capacity has been inferred with continent statistics within the 5% error. The training dataset has been built by using public data from Open Street Map and dedicated databases, e.g. Global Cement Directory, and it has been thoroughly verified and refined. We also identified asset ownership through company disclosure of major producers and further verification with the OpenPermID database.

The datasets are now publicly available through the GeoAsset project of the Spatial Finance Initiative, established by The Oxford Sustainable Finance Group, Satellite Applications Catapult, and the Alan Turing Institute: https://www.cgfi.ac.uk/spatial-finance-initiative/geoasset-project/geoasset-databases/

Environmental, Social and Governance (ESG) data are the basis for monitoring impacts or risk, and how they change over time at the asset level and at aggregated levels across enterprises, portfolios or administrative areas. We elucidate to how Earth Observation data and derived products such as Copernicus services along with other geospatial information can be used to assess potential physical, transition or liability ESG impacts and risks.
We suggest a geospatial-based approach to provide ESG relevant insights into a specific asset and call it spatial finance. We focus on spatial ESG data as the basic (fundamental) level. Through combining earth observation (EO), Geographic Information Systems (GIS) with artificial intelligence (AI) we enhance the availability of information in financial systems, and change how risks, opportunities and impacts are measured and managed by financial institutions and civil society. This data renders possible to assess the asset against observational data, such as environmental, climate, governance and social variables. This way, financial decisions shall become more transparent.
Sub-asset monitoring (e.g. power smart meters) or voluntary reporting (e.g. ESG company reports) can be integrated for even greater insights. Merging results of multiple assets at the subsidiary, parent company, portfolio or national, or sector level, provides insights at scales relevant to different financial applications, ranging from project finance to sovereign debt.
Continuous advancements in data processing technologies such as AI, specifically machine learning, computer vision and edge computing, will enable dynamic and timely spatial analysis, both at a global scale and at local to regional scales. The current ESG data affordances require already some modelling, but predictive modelling capabilities will expand spatial finance applications to new markets. The authors believe that a harmonisation of the methodologies and increased availability of asset-level data will overcome the current ESG data bottleneck and will allow translating existing environmental and climate datasets into suitable indicators for financial sector use.
The EU taxonomy regulation and green banking regulations from bodies like EBA accelerate the developments in this field. The EU taxonomy is a classification system, establishing a list of environmentally sustainable economic activities. It shall play an important role helping the EU scale up sustainable investment and implement the European green deal. The EU taxonomy provides companies, investors and policymakers with appropriate definitions for which economic activities can called environmentally sustainable. It shall protect private investors from greenwashing, help companies to become more climate-friendly and help shift investments where they are most needed.
In addition to the legislative stimuli, technologies mature and more sophisticated applications become available and the amount of studies based on spatial analysis expand rapidly. In principle, asset managers and banks are already able to make decisions and to report on ESG risks in a granular way but so far not in a harmonized way. A comparison between different rating systems shall become possible when all technical annexes of the EU taxonomy regulation will be available.
In the full version of this article, we will report on a study in Austria that creates methods and prototype tools to assess environmental exposures at asset (parcel) level. We exploit the increased temporal and spatial resolutions and the availability of EO to assess Tier 3 indicators on soil sealing, biodiversity and climate change risks. We link EO techniques to the locations of private and company assets and develop methods for assessing the exposure to deforestation, biodiversity loss and water risk over time and across sectors of activity. Tier 3 is defined as high-level asset-level insights covering the full extent of the specific assets, resting primarily on the comparison of asset datasets (location of factories, farms, mines) against spatial datasets and remote sensing insights sourced independently. Insights generated by GIS define the proximity of commercial assets to other assets such as World Heritage Sites or Protected Areas, rivers, wetlands, endangered species, and indigenous areas. Secondly, we combine this data with other modelled layers and EO products to estimate risk and impact aspects of the asset, such as water risk of the site, nearby deforestation or carbon emissions. In addition, we show exemplarily utilization from Tier 4 datasets for Tier 3 and vice versa, while aggregating or dis-aggregating or disclose insights of the status of an asset. We use a range of data from freely available open datasets to highly detailed data from commercial companies. It turns out that sustainable investors rely on the quality and relevance of environmental (the “e” in ESG) data but the availability of some observational data is the bottleneck in such a methodology. As of today, even the most transparent companies disclose almost no data on their environmental impact and exposures. Therefore, combining explicitly spatial data with spatial data science will help to replace self-disclosure based environmental sustainability data with a fact-based, objective, reliable and timely assessment. We conclude that the tremendous improvement and increased availability of EO data offer options to derive essential data at asset level addressing EU taxonomy aspects such as land use, soil sealing, biodiversity or water data.

From the United Nations Decade on Biodiversity 2011 - 2020, a new consciousness on ecosystem-friendly investments has evolved in the financial world, finding a regulatory manifestation in the EU taxonomy for sustainable finance. New metrics and tools have been developed that allow for the assessment of businesses, economic sectors, or entire portfolios with regard to their surrounding ecosystem services and natural capital assets. Most existing initiatives focus on the impacts of economic activities on ecosystems and local biodiversity. To close the gap, our consortium is developing a prototypical platform that allows investors and financial institutions to assess biodiversity-related risks towards their portfolio, using EO technology and remote sensing based data products as well as available company-specific supply chain and asset level data from publicly listed companies.
Users will be able to upload their portfolio composition, for which different components of biodiversity-related risk are automatically assessed. These are in particular physical risk, reputational risk, regulatory risk, and market risk. This presentation focuses on the physical risk module, where EO data products represent a valuable source of information. The module builds on fundamental work done in the ENCORE project, in which the relationships between drivers of environmental change, natural capital assets, ecosystem services and production processes have been described exhaustively and weighted by economic sector. Additionally, established databases containing information on the status of natural assets, e.g. the WRI Aqueduct 3.0 dataset on the spatial distribution of water-related risks, are used as a baseline. Up-to-date EO data products are then used to assess drivers of environmental change to investigate dynamics and trends, and to identify potential threats to the natural capital assets and the provision of corresponding economy-relevant ecosystem services. Here, we focus on drivers responsible for biodiversity loss, for example by using annual global land cover data from the Copernicus Global Land Service to assess habitat modification due to land cover change. The spatial scale of the analysis is determined by the granularity of available data on publicly listed companies. If asset-level data is available, e.g. coordinates of specific production sites, drivers and ecosystem services can be assessed on a location-specific level, such as catchment areas or certain radii around the locations. For such “spotlight” analyses, further high-resolution options are envisaged that employ Sentinel-1/2 data and derived products. If the available company data does not contain specific location information, relevant statistics can be calculated on country or state level or they can be derived from official sources. Depending on the respective economic sector, an importance weighting of ecosystem services is performed, and an overall indicator for physical risk of the entire portfolio is calculated from all company-specific assets.
The first prototype tool will contain routines for the physical risk calculation for a subset of those drivers and ecosystem services with the highest relevance to biodiversity and economic impact (e.g. the provision of surface water as an ecosystem service is crucial to many production processes). First models of driver and ecosystem service assessment at different spatial scales are currently being developed and consolidated in a proof of concept, taking into account a variety of spatial data sources. First results indicate that continuity and global coverage are the most relevant data requirements to develop and continuously apply a stable methodology. Therefore, existing and future Copernicus data products represent an essential input to this project.
Perspectively, this project will advance the understanding of how biodiversity-related risks affect financial institutions and will thereby a) contribute to the divestment from assets with high biodiversity-related risks; b) incentivize investments in economic activities that substantially contribute to biodiversity protection and ecosystem restoration; and c) provide a starting point for shareholder engagement on biodiversity issues.

Cloud-native geospatial technologies are key to scaling the geospatial data sector, and for the efficient discovery and computation of relevant data. However, they are built in a centralized computing paradigm, with roots still in the client-server model. Data networks designed in this way are brittle - links can break, file content can be changed unbeknownst to users, and data must be held by trusted custodians who may abuse their role as maintainers of server systems holding geospatial data.

A more secure, resilient and user-centric vision for the Internet is evolving with Web3. Decentralized identifiers, peer-to-peer storage and transfer, and content identifiers resolve the most critical fragilities that exist in the incumbent web.

The geospatial data sector is not immune to these problems - in fact, the ecosystem is rife with inefficiencies. Many organizations hold multiple versions of the same reference spatial datasets on various internally-operated servers. Each of these servers requires computing resources and skilled workers to maintain. It seems unlikely that each of these datasets is kept up to date based on update release cycles, meaning the reference data upon which the organizational data is overlaid is not current. Deep architectural wastefulness pervades geospatial data storage systems in the public, private and third sectors worldwide.

The Web3-Native Geospatial vision seeks to integrate the learnings of the incumbent geospatial data practitioners with the principles underpinning the design of the decentralized web.

- Data must persist — we cannot try to resolve a mission-critical dataset a few years after its creation only to receive a 404 error. This is especially true for spatial finance applications where decisions dealing with large monetary values are made based on insights derived from geospatial data.
- Datasets must be verifiable — all parties can have cryptographic confidence when accessing a dataset around the world and into the future that they are looking at the same information as others.
- Data networks must be antifragile, — they must be self-healing and resilient to the failure of any actor.
- The right to privacy must be built in on the technical layer, not added on top of system design at the policy layer.

These are challenging problems to solve, and the community is early in the process of developing the standards, tools and protocols that will unlock the potential of the decentralized, location-based web.

In this talk, the core team from the Astral Protocol will outline their vision for a Web3-native geospatial data architecture, including data storage and computation systems that are decentralized and fault tolerant, and in which all participants are cryptoeconomically-aligned. The Astral team will touch on data storage systems they are designing based on IPFS, Filecoin, Arweave and Ceramic, and will share insights into decentralized geospatial data processing and discovery systems including dClimate, Ocean Protocol and Algovera AI. They will also touch on their early research into privacy-preserving geospatial technologies.

These technologies are designed to interoperate with distributed ledgers, smart contracts and digital currencies, with an eye towards underpinning the emerging regenerative finance ecosystem being built on public blockchains.