Earlier this year, Microsoft’s AI for Earth and the GEO BON Secretariat launched the program “Essential Biodiversity Variables in the Cloud”, a new US$1 million grant program providing financial support and Microsoft Azure credits to monitor Earth’s biodiversity. The call was particularly designed for advancing research and applications that leverage cloud-scale computation to expand the geographical and temporal coverage of biodiversity information.

The program successfully attracted 60 proposals aiming to contribute developing the Essential Biodiversity Variables and derived biodiversity change indicators. We are pleased to announce the five selected projects that cover a diversity of world regions and biodiversity monitoring applications and that will receive direct funding, Microsoft Azure cloud credits, and free Esri Licenses:

Using AI to validate and downscale ecosystem-related Essential Biodiversity Variables (EBVs) in mountain environments

Ruth Sonnenschein (PI – EURAC Research, Italy)

The Group on Earth Observations Global Network for Observations and Information in Mountain Environments (GEO-GNOME) is a GEO Work Programme Initiative1 that seeks to connect and facilitate access to diverse sources of mountain observation data and information at different scales. At a recent GEO-GNOME workshop2, experts on mountain ecosystems were convened to identify and prioritize Essential Biodiversity Variables (EBVs) that are considered relevant for mountains, using the GEO BON EBV framework as a starting point. For the EBV class on Ecosystem Structure, the EBVs ‘Ecosystem extent’ and ‘Ecosystem fragmentation’3 were identified as priorities to monitor and better understand changes in mountain ecosystems and their species-level biodiversity. Both EBVs require the mapping of ecosystem occurrences and ecosystem distributions to determine the area occupied by ecosystems (Ecosystem Extent) and how and where those areas are reduced due to natural and human-caused disturbances (Ecosystem Fragmentation).

While a map of global mountain ecosystems at a spatial resolution of 250m is available, its utility for regional or local scale applications is not yet demonstrated for two main reasons: 1) the accuracy of this global characterization in a local setting is unknown due to a lack of valuation, and 2) the relatively coarse spatial resolution of the data product may preclude its use at a local scale. We propose the use of artificial intelligence (AI) to address these two problems of validation and coarse spatial resolution, by outlining a project that exploits both the advanced feature extraction capabilities provided by AI-based algorithms and the computational potential of cloud based-platforms to derive accurate, high-resolution maps of mountain ecosystem extents. Via the incorporation of the latest remote sensing data, we seek to produce these maps through time, thereby enabling a comprehensive assessment of ecosystem change and fragmentation.

AI for the Belize National Marine Habitat Map

Arlene Young (PI – Coastal Zone Management Authority &Institute, Belize), Chantalle Samuels, Andria Rosado, Emil Cherrington, Robert Griffin.

For this project, we propose to use Microsoft Azure for machine learning-based mapping of the Essential Biodiversity Variable (EBV) of ecosystem extent and fragmentation. The project’s geographic focus will be Belize’s coastal and marine ecosystems, with particular attention focused on coral reefs, seagrass pastures, and mangrove ecosystems. These ecosystems are recognized for their blue carbon focus and potential to contribute to Belize’s climate change mitigation efforts. Via this proposed work, Belize’s 1997 30m Landsat-based National Marine Habitat Map will be updated, using a combination of 3m Planet Scope and 10m Sentinel-2 imagery.

As such, the data will provide updated estimates of the status of Belize’s major coastal and marine ecosystems. In addition to helping inform Belize’s Nationally Determined Contribution (NDC) to the Paris Agreement, the data will also support the country’s implementation of Sustainable Development Goal 14 and will be integrated into the revision of the national Integrated Coastal Zone Management Plan. Capacity building will also be a key focus of the project through knowledge-transfer workshops which will help to sustain and extend the technical capacity of CZMAI in the field of cloud computing.

AMAZECO: Covering the Amazon with an Ecosystem Structure EBV product combining satellite and airborne LIDAR

Ruben Valbuena (PI, Bangor University, UK), Eric B. Görgens (co-PI, Universidade Federal dos Vales do Jequitinhonha e Mucuri, Brazil), Carlos A. Silva (co-PI, University of Florida, USA).

Ecosystem vertical profiles (EVPs) characterize the vertical distribution of sessile biological entities in an ecosystem, which affects the number and variety of potential niches and identifies critical aspects of ecosystem state. In Valbuena et al. (2020) we advocate for a standardization of ecosystem morphological traits derived from EVPs characterized by LIDAR, so that they can become useful to inform ecosystem structure EBVs. These traits should focus on being relevant to the ecosystem, and not on the means for measuring them. Thus, the goal of this project is to demonstrate that we can deliver platform-independent EVPs from both satellite and airborne LIDAR sensors, and provide the means for a global ecosystem structure LIDAR product that can be crowdsourced through national BONs.

This will be enabled by high performance computing (HPC) workflows for common satellite/airborne LIDAR derivation of ecosystem traits, which we will produce and implement into a first prototype product covering whole of the Amazon with traits produced from combined satellite and airborne LIDAR. The satelliteLIDAR will be obtained from the currently operational global ecosystem dynamics investigation (GEDI) mission. The airborne LIDAR workflow will make use of an unprecedentedly extensive dataset of 906randomly located transects sizing 375 ha each, from the ‘improving biomass estimation methods for the Amazon’ (EBA) (Görgens et al. 2020), plus data from the Sustainable Landscapes Brazil (SLB) project (Longo et al. 2016).

The product will consist of a multilayered raster data product with LIDAR measures of EVPtraits–ecosystem height, cover, and structural complexity, including estimations of their uncertainties and a demonstration of how airborne LIDAR can be used to improve those over a satellite product. The code developed will be made publicly available for other GEO BON members and organizations, with procedures incorporated as a function in the rGEDI package (Silva et al. 2020), and HPC pipelines enabling national BONs to compute these EVP traits locally, or nationally, using globally consistent protocols that comply with the standards established for the EBV portal.

Extracting the signal of change in community-composition EBVs from big unstructured species-occurrence datasets through Azure-enabled spatiotemporal analytics

Simon Ferrier and Andrew Hoskins (PIs, CSIRO, Australia).

Big-data initiatives such as the Global Biodiversity Information Facility (GBIF) have revolutionized access to data on known occurrences of species across space and time. Growth in the volume of these data is continuing at a rapid pace, thanks in large part to the advent and popularity of smartphone apps for recording species observations, through citizen-science initiatives such as iNaturalist and eBird. The highly unstructured nature of many species-occurrence datasets has long presented a major challenge for any attempt to extract useful information on biodiversity change, as records distributed across time have most often been generated not through repeated sampling of the same spatial locations, but rather through largely opportunistic or ad hoc observation. A recent extension of CSIRO’s generalised dissimilarity modelling (GDM) technique – obs-pair GDM – addresses this challenge head-on by allowing change in the species composition of communities across both space and time to be detected and quantified through analysis of unstructured species-occurrence data.

The proposed project will make this analytical capability more widely available for use around the world, through the Azure cloud-computing service. The developed capability will allow ingestion and analysis of spatiotemporal species-occurrence data and environmental layers from a wide variety of sources, varying freely in spatial and temporal extent and resolution, and in biological scope. Outputs generated by the application of this capability to any given dataset will be structured as a Beta Diversity EBV hypercube, thereby addressing an important element of the Community Composition EBV Class which has received little attention to date. The project will demonstrate the value of this solution through initial application to national datasets for two contracting countries – Australia (working in collaboration with the Atlas of Living Australia) and Bolivia (working in collaboration with NatureServe and the EcoHealth Alliance).

Bioacoustics and Machine Learning for Automated Avian Species Monitoring in Global Biodiversity Hotspots

Naomi Bates (co-PI, Future Generations University, USA,) and Sebastian Herzog (co-PI, Asociación Armonía, Bolivia).

Time-series of species presence across multiple ecosystems can be used as a biodiversity change indicator for informed decision-making in conservation efforts. Bioacoustic monitoring with machine learning (ML) analysis provides species presence/absence tools. These tools: 1) measure complex biodiversity change; for 2) scientists and conservation managers to understand behavior, diversity, and habitat preference; providing 3) a scalable and locally relevant global framework towards GEO BON’s mission of ‘improving the acquisition, coordination and delivery of biodiversity observations’.
The Songs of Adaptation project of Future Generations University maintains a monitoring network of biodiversity hotspots in Nepal, Bolivia, Uganda, and the United States. Bioacoustics and climatic data are collected across mountain transects (or other ecosystem gradients), aligned with GEO BON Essential Biodiversity Variable candidate: species distribution (SD EBV). As local temperatures, weather patterns, and ecosystems change, a gradual soundscape shift is hypothesized as species move to different elevations or ecosystems in response to changing climate conditions.

Combining the big data management tools of Azure and ML, open source scalable tools will be created for the GEO BON community and beyond. The impact of the proposed work will be two-fold: open source tools and data for the GEO BON community, and specific species insights for species in biodiversity hotspots in Nepal, Bolivia, USA, and Uganda. Starting with this now proven framework for using bioacoustics as a tool for avian species population surveys and monitoring, the vision is within reach to create tools for stakeholders to track climate change impacts for locally driven decision-making and informed adaptation.

Five grantees for the first EBVs on the cloud call