Limits to Extrapolating Biodiversity Patterns From EO: Implications for Model-Based EBV Estimation and Corporate Reporting

The widespread degradation of nature has increased pressure on corporations and financial institutions to report and mitigate their impacts on biodiversity. Despite the increasing abundance and accessibility of biodiversity data and measurement technologies, there remains a lack of datasets suitable for deriving comprehensive summary measures, such as Essential Biodiversity Variables (EBV’s), that can underpin robust impact assessments and reporting aligned with the Kunming-Montreal Global Biodiversity Framework (GBF). One way to assess biodiversity risk and impact at scale could be to link local biodiversity data with large-scale measurements of environmental variables, for instance from satellite Earth Observation (EO) platforms, to provide model-based estimates of biodiversity feeding into GBF indicators. However, extrapolation of patterns of biodiversity is a notoriously difficult task, and is often associated with high errors when predicting patterns to new spatial locations. A review of datasets and tools currently used by corporations and financial institutions shows that extrapolation to local sites from global datasets, or using only proxies, is the currently dominant approach. Here, we test the reliability of such assessments by combining high resolution EO time series data that capture seasonal dynamics with large scale biodiversity time series data from two countries, using machine learning algorithms to predict five EBVs. We show that while reasonable predictive performance can be achieved at sites with local data, performance declines considerably when modelling measures at new sites. We draw on these results to argue that biodiversity patterns are hard to generalise to sites where no local data has been collected, and conclude with a proposed biodiversity data hierarchy framework focusing on data quality. Because biodiversity risk and impact reporting is still evolving, corporations now have a critical window of opportunity to showcase measurable incremental improvements in data sources and data quality that underpin the Global Biodiversity Framework impact assessments and reporting.