Widespread changes in the area, fragmentation, and integrity of natural and semi-natural ecosystems, from topical savannas to alpine wetlands, are a major driver of biodiversity change and loss of ecosystem services worldwide. These changes are expected to further accelerate under future land-use and climate changes, potentially leading to widespread environmental degradation and socioeconomic hardship. Accordingly, the current draft of the monitoring framework for the Post-2020 Biodiversity Framework of the UN Convention on Biological Diversity prominently features indicators on ecosystem extents, and GEO BON has identified data products on the “Ecosystem Extent” Essential Biodiversity Variable a as key priority for global biodiversity monitoring. The IPBES Assessments recently reported that most natural and seminatural ecosystem types are globally and regionally declining. Yet, for most ecosystems and world regions, systematic data on direction and magnitude of changes are lacking, leaving the generality of declines unclear.

We will present a series of global, high-resolution (sub-pixel areas at 1-km) gridded data cubes that capture the annual areas of occupancy for each of ~60 standardized terrestrial, freshwater, and coastal marine ecosystem types over nearly three decades. To achieve this high thematic detail, we built on decades of global Earth Observation and environmental mapping efforts by integrating >40 global satellite-based and in-situ datasets, covering land cover, land use, hydrology, climate, coastal and stream topography, soil, and various other environmental dimensions. The ecosystem types are mapped via a combination of deductive rulesets and inductive modelling using thousands of samples and gridded predictors. Comprehensive validation against millions of in-situ records at multiple spatial grains showed high overall accuracies, however, with strong variation among major ecosystem classes. The depicted ecosystem types conform to the habitat classification scheme of the International Union for Conservation of Nature’s Red List of Threatened Species, assuring interoperability with ongoing assessments of the species habitat preferences. Exemplary application fields include monitoring of species habitats to support conservation interventions, testing hypotheses on biodiversity-change drivers and mapping of global ecosystem services. The data cubes will soon be published as open-access products following FAIR principles.

Together, these data cubes provide an unprecedentedly comprehensive record of global multi-decadal dynamics in ecosystem extents. We will present results of a global analysis of ecosystem transitions and net changes. Artificial ecosystems like pastures, plantations, and mining areas increased substantially over the past decades, mainly at the expense of sub-/tropical lowland moist forests. Yet, surprisingly, global and regional extents of different (semi-)natural ecosystems did not decline systematically over recent decades. Instead, they exhibit regional gains and losses, and common shifts between upward and downward trends, but no net loss on average. This apparent paradox is explained by few, globally extensive and declining ecosystems making room for areal increases in many less extensive (semi-)natural ecosystems, notably including most wetland types, as well as in artificial systems. Our results may inform final discussions within the UN Convention on Biological Diversity (CBD) on ecosystem targets and baselines for the Post-2020 Framework, highlighting among other things, the need for highly targeted ecosystem restoration efforts, while the new data can help improve post-2020 ecosystem monitoring especially for countries with limited national capacity.

The availability of high resolution images from space, driven by Sentinel-1, Sentinel-2 and Landsat satellites, was a major step for the creation of global land cover maps with a high spatial resolution. The land cover is obviously a valuable information for biodiversity study, but 1) its rigorous definition at different scales is challenging and 2) land cover classification systems often lack thematic precision to qualify biotopes according to their ecological functions. One of the main issues is the difficulty to describe the spatial relationship between the different land cover elements in a structured and transparent way,. We therefore make the hypothesis that an object-based approach generates a consistent spatial support for the analysis of the land cover from an ecosystem mapping perspective.

The image-objects are obtained using the multiresolution segmentation algorithm applied on a Sentinel-2 mosaic stacked with ancillary information derived from a digital elevation model (potential solar illumination and topographic position index). The potential of this hybrid source of information was previously demonstrated with very high spatial resolution images over Belgium, increasing the homogeneity of the ecological functions and improving the quality of habitat suitability models derived from it. The new prototype based on Sentinel-2 currently covers Southern Europe (up to latitude 52°N) and is expected to be extended to all the continent in the second quarter of 2022.

This innovative geographic dataset - created in the frame of the European Research Infrastructure Consortium for biodiversity and ecosystems functions (Lifewatch) - is called ecopatch because it is primarily composed of ecologically consistent spatial units. The spatial precision of the boundaries corresponds to the size of the smallest Sentinel-2 pixels (10m). The delineation including topographic variables provides additional consistency to the delineation of spatial units because ecosystems function also depends on the topographic position (e.g. soil type, topographic position...). Each of the 7.7 million polygons is then described with biotic and abiotic attributes mostly provided as continuous variables. The pixel-based land cover proportions at different scales (inside the polygons and in arbitrary buffer zones) is the main characteristic, but topographic indices, climatic variables and contextual variables (e.g. night lights, distance to roads or distance to forest boundaries) provide additional information that can be used to better quantify the services provided by the patches. Multiple categorical legends are derived from these variables, which enhance the interoperability of the dataset and provide additional information about the ecosystems. The results of the object-based validation are not yet available.

Blue carbon ecosystems like seagrasses, mangroves, and tidal flats provide globally significant
yet vastly underestimated and impacted ecosystem services to humans, economies,
sustainability, biodiversity and ecosystems, the so-called natural climate solutions (NCS).
Accelerating climate change, biodiversity loss, eutrophication, coastal development, and uneven
levels of protection are all placing significant stress on the extent, condition, services and financial
benefits of these coastal ecosystems, worldwide. Contemporary approaches are urgently needed
to reverse and halt this accelerating reduction of our global natural capital to avoid crossing tipping
points and cascading impacts across the interconnected environmental and artificial ecosystem.

Ecosystem Accounting (EA) presents a new holistic, comprehensive approach to streamline
physical, monetary, and thematic evaluation for natural ecosystems, integrating their immense
value and services into national and international policy making, funding, and data-driven action.
The conceptualization and adaptation of the System of Environmental-Economic Accounting
(SEEA) EA by the United Nations Statistical Commission signals a new era of holistic, integrated
statistical, technological, ecological, and financial solutions for the conservation, restoration and
protection of ecosystems like seagrasses and mangroves. Although a very promising systems approach, EA approaches and applications for coastal ecosystems are still very much in their infancy, if not non-existent.

In parallel, unprecedented advances in Earth Observation during the last decade—cloud
computing, artificial intelligence (AI), big satellite data from the Sentinel. Landsat, and
PlanetScope satellite sensors—are offering a vast, transparent and homogeneous knowledge
base for our Earth’s natural capital. Integrating global field and citizen observations in synergistic
Earth System Science and Earth Observations approaches comprise very promising big data
paradigms that can transform data and evidence to sustainable insights and solutions.
Amalgamating those big data paradigms in ecosystem accounting frameworks can lay the
foundations of the next generation of large-scale decision support systems for natural
ecosystems. These support systems will transform physical ecosystem accounts reliant on
ecological knowledge into economic units, measurable targets, and thematic accounts organized
around environmental policies. This realization will in turn supercharge targeted financing,
policies, and actions across the entire terrestrial, coastal and marine realm.

Here we present a novel coastal ecosystem accounting prototype, built end-to-end within the
cloud platform of the Google Earth Engine after six years of research and development in coastal
aquatic Earth Observation. Our prototype harnesses the powerful parallel processing of the Earth
Engine cloud, the full public archive of the European Union Copernicus Sentinel-2 satellites,
globally aggregated reference datasets, and AI-guided big data analytics to map the ecosystem
extent, condition, and services of seagrass ecosystems. More specifically, our modular apparatus
synthesizes analysis-ready data cubes using terabytes of 10 m satellite images, whose pixels are
then transformed into a multitude of seagrass-related thematic (e.g., subtidal areal extent,
turbidity, carbon stocks and sequestration, and biodiversity) and continuous accounts (e.g., extent
probability, bathymetry, water quality). We showcase recent national seagrass EA applications,
including their accuracies and uncertainties, across largely-uncharted underwater biomes, the
Western Indian Ocean, the entire Mediterranean, and the Caribbean, spreading across more than
30 countries and 300,000 sq km. We targeted these coastal biomes due to their vast blue carbon
and coastal biodiversity stocks, notable lack of spatially explicit information, and uneven uptake
in global funding mechanisms and policy agendas.

Our introduced EA prototype and its coastal applications concern only baseline mapping of
biophysical stocks of a single ecosystem. In the next phase of research and development, we aim
to integrate fit-for-purpose ecological modelling, economic evaluation techniques, and remotely
sensed change detection into our Earth System Science-Ecosystem Accounting prototype. We
envisage that this amalgamation of legacy and modern mapping, modelling, and observations will
enable policy makers and governments to see the forest for the trees concerning the world’s blue
carbon ecosystems. This will ultimately safeguard balanced blue carbon conservation and
restoration financing, accounting, decision making, and resilience within and beyond the 21st
century.

Some authors have identified five forms of ecosystem integrity: Ecosystem integrity of wilderness; Ecosystem functional and structural integrity; Ecosystem stability and resilience; Ecosystem condition and Ecological quality and value. Here, we propose that Explainable Artificial Intelligence (XAI) can be a useful approach to assess an integral measurement of the condition of ecosystems, combining information from all these forms of ecosystem integrity. Given the different reasoning types offered by methods such as Bayesian networks, it is also possible to infer the values between each one of these forms of ecosystem integrity.
There are many proposals to assess ecosystem or ecological integrity. Most of them adopt an “expert opinion” stance to define the metrics. We departed from this approach and designed a data-driven approach based on Bayesian networks. There is substantial data to support such an approach worldwide. Of course, one of those sources is remote sensing in its various current flavors. In addition, in many countries, there is “in situ data” that provides further sources of evidence, as is the case of Mexico through the INFYS database (National Inventory on Forest and Soils). Thus, we used both satellite and INFYS data to produce an index of ecosystem integrity based on Bayesian networks training and computation. Our conceptual approach was based on a three-tier model that considered that ecosystem condition (a hidden tier in our model) has influence (a likely causal relation) on the actual values we observe on reflectance and in situ variables, accounting for both structural and functional attributes (a detection tier in our model). At the same time, these values we observe are influenced by the ecological context where the observation is conducted (a contextual tier in our model).
In addition to these three tiers, we realize that socioecosystemic interactions have a strong influence on ecosystem condition. This time, we model this influence by including indicator variables on land use. We added them as influencing the ecosystem condition directly. This model defines a comprehensive representation of ecosystem integrity as a joint probability distribution of the ecological process that underlies the dynamics of ecosystem integrity in space and time. Following this approach, the model of Mexico we developed used both Landsat and MODIS remote sensing and INFYS data to produce a wall-to-wall estimate of ecosystem integrity for the whole Mexican conterminous territory.
There is mounting interest in evidence-based decision-making, and sustainability is no exception. A significant development in this direction is the SEEA EA fostered by the UN. We have proposed the three-tier model of ecosystem integrity as an approach to produce ecosystem condition accounts. Although we realize the mathematical background of Bayesian networks could be complex, we have also found that they are powerful artificial reasoning devices to support decision-making. The data-driven approach is also a potent way to deal with what is a daunting challenge to experts when they have to devise decision rules on selected variables (which should include a whole set of multiple combinations among them) to account for different levels of ecosystem condition. Bayesian networks are represented by two paramount components: 1) a qualitative depiction of the influence pattern describing the subject matter, and 2) a set of conditional probability tables accounting for the precise probabilistic interaction among the variables. This segmentation of information provides a great opportunity to make Bayesian networks relatively easy to communicate and thus provide an AI explainable device for public access, probably at a comparable level of complexity to an expert opinion ranking device. Therefore, the modeling approach we used provides both a practicable way to assess ecosystem condition and advances towards an explainable AI commitment. We consider that it provides both an enabler approach and infrastructure for practitioners' adoption in natural capital accounting.
We tested this approach for the case of Mexico and calculated the Ecosystem Integrity Index (EII) that estimates the condition of the terrestrial ecosystem per 250 m. The EII based on the three-tier model that we described above, explicitly integrates causally, indicators of ecosystem integrity of various types. The EII was used within the SEEA EA framework in collaboration with national agencies from Mexico, specially INEGI. We produce estimates for 2004 and 2018. Then, analyzed changes among ecosystems along time, as required for the accounting system.
In addition to the accounting contribution itself, we also found it interesting to produce another Bayesian network that interconnects the various ecosystemic frameworks that were under discussion through the process of conceptual integration of the Mexican accounts. In this network, we identified links between vegetation types classification (sensu INEGI and CONAFOR), life zones (sensu Holdridge), Ecoregions (akin to IUCN ecosystem types), land use type (sensu Inegi). We also found it useful to link Human Footprint assessments that were available for Mexico, besides our Ecosystem Integrity Index. The network was trained at the level of pixels 250m in size. This Bayesian network demonstrated to be an interesting device to communicate and even explain to different audiences the results of that condition account for the various ecological scenarios, and at the same time explore the “condition mapping” on those representations. Depending on the level of detail with which the “evidence” was provided to this network, it was able to present a forecast from the pixel to the regional levels.
Our experience on using a Bayesian networks approach in this fashion, suggests that it is a profitable way to 1) produce reliable data-driven assessments of ecosystem condition, 2) it can be expanded to facilitate communication and even explanation of condition accounting, and 3) we are convinced that the Bayesian approach could be expanded to support ecosystem services accounting at the level of ecosystem assets.

Using a continental-scale geo-spatial accounting system to support EU decision-making

Authors

Eva Ivits, Jan Bliki, Jan-Erik Petersen, Roger Milego, Manuel Loehnertz, Emanuele Mancosu, Mirko Gregor, Iñaki Diaz de Cerio, Mikel Garzia, Oskar Esparza

Land use and land use change are fundamental for sustainable resource use and the delivery of ecosystem services, including the provision of food, nutrient cycling and climate change mitigation through carbon sequestration. Land resources are part of our shared natural capital and must be well managed to maintain a healthy environment and human well-being. As such, only if land use and its impacts are properly addressed is progress towards sustainable development in Europe possible. Land-use related policies require the development of harmonised datasets, transparent methodologies and easily interpretable statistics. Land and ecosystem accounts fit the bill, describing how land resource stocks change over time.

The EEA has developed an Integrated Data Platform for transparent, repeatable and efficient land and ecosystem accounting supporting EU policy making with interactive dashboards, which also allow query and access to the accounting databases.

The EEA’s Integrated Data Platform (ETC/ULS, 2020) is made up of geospatial data architecture and IT system infrastructure to enable the consistent and systematic production of land and ecosystem accounts. In detail, the Spatial Data Infrastructure organises and catalogues geospatial data with an interactive interface. The thematic node, ´Environmental Accounting Reference Layers’, assembles harmonised and quality assured datasets. Interactive dashboards enable the interlinkages between geospatial data that describe our natural capital to be understood. Another element of the IDP is the Joint Environmental Data Infrastructure (JEDI, ETC/ULS, 2020), which uses cloud infrastructure to transfer geospatial data into dimensions. Using data cubes, the dimensions of land cover datasets may be integrated with other dimensions, such as administrative boundaries, biogeographical regions or bio-physical and socio-economic data.. For the integration, uniform reference grids are used with 1 km, 100m and 10m cells as the accounting units, following INSPIRE directive specifications.

A data cube can be read using visualisation software that allows for user driven queries and gives tabular, chart or map results.

With the increasing number, spatial and temporal resolution of Earth observation products, improved capacity for big data processing is becoming a must. To provide policy support, the management and processing of such datasets must follow state-of-the-art standards and become automated, efficient and up-to-date. The EEA contributes to the EU’s environmental knowledge base and international environmental accounting frameworks by further developing its geospatial environmental accounting system.

The European Commission is proposing an amendment on European environmental economic accounts to include ecosystem accounting in their national reporting system. Ecosystem accounts present data on the extent, the condition of ecosystem assets and on the services they provide to society and the economy. As such they address EU policy needs such as the new ‘Green Deal’ , Sustainable Development Goals (SDG), Biodiversity Strategy for 2030, Nature Restoration Law, Pollinators Initiative, etc. The United Nations System of Environmental Economic Accounting – Ecosystem Accounting (SEEA-EA) framework provides guidelines on how these accounts should be set up. To implement these guidelines, EUROSTAT is extending their Integrated System of Natural Capital (INCA). Based on the piloting work done in the period 2016-2020, the developed methods will be further improved, integrated and transferred into a user-friendly tool which allows member states regular and timely production of ecosystem accounts. Through the hierarchical setup of the tool the pilot INCA accounts at EU wall-2-wall scale are improved and extended.
To ease the use, the INCA tool is programmed in Python and will be available as plugin for QGIS - a free and open-source geographic information system software – using the PyQGIS plug-in architecture. The QGIS plugin allows the processing and combination of statistical and earth observation (EO) data from various data streams and at different scales. Since the tool takes away the hassle of data preparation and data processing via Extract, Transform and Load (ETL) workflows, the Member States can generate ecosystem services accounts more efficiently, in line with the SEEA EA framework. For more experienced users, the modular setup of the tool allows batch processing and therefore automatization to conduct for example sensitivity analysis.
During our presentation we will showcase the tool prototype including the already fully implemented workflows for the accounting of the ecosystem services. We will show the wall-to-wall generation of the accounts at Member State level to discuss the advantages and limitations of the automatic accounting system.