Sequencing to satellites: Integrating biodiversity surveys across scales

The global decline of natural ecosystems is a consequence of a number of phenomena, including climate change, habitat loss and fragmentation, and the introduction of invasive alien species. As a result, the ecosystem services provided, including carbon sequestration, nutrient cycling through the trophic pyramid, and pollination and seed dispersal, are at risk. Mitigation strategies to halt and reverse these trends can be devised but must be informed by spatially and temporally accurate information on the location and causes of ecosystem change. This type of scalable ecological modelling directly supports global biodiversity frameworks, such as the Kunming-Montreal Global Biodiversity Framework (GBF), which call for spatially explicit monitoring of ecosystem condition and change. While field surveys can provide accurate information to inform mitigation strategies for target ecosystems, they are often costly in terms of time and resources. Consequently, the frequency and spatial and temporal extent at which field surveys can be conducted is limited. In contrast, Earth observation (EO) satellites provide freely available, repeated, wall-to-wall information of the Earth’s surface. The combination of EO data with environmental data has been successfully employed to describe land classes, and quantify changes within them, such as levels of deforestation. Although the description of land classes enables some inference of expected ecosystem structure, ecosystems do not always transition categorically, although abrupt land use change can result in clear shifts. More often, ecosystems vary gradually across landscapes, much like the reflectance values recorded by EO satellites. It is hypothesised that the composition of the studied communities will vary in accordance with the structure of the habitat, and that elements of this structure will be described by EO. The objective of this thesis was to investigate the potential of combining EO data with field survey data in multi-species distribution and occupancy models to predict community composition across landscapes. This approach enables the investigation of the influence of environmental factors on biodiversity and the inference of ecosystem condition. The study used four distinct groups of field data: plant communities in the Succulent Karoo of South Africa, insect communities in the vicinity of the Gola Rainforest in Sierra Leone, and bird communities identified in the Peruvian and north Brazilian Amazon. These data sets comprised between 120 and 235 species, which were included in analysis. The study found that for all data sets, models fitted with EO data yielded predictions of community composition that were as or more accurate than those made by models fitted with environmental data. In predicting the validation data, the community mean area under the curve (AUC) values ranged between 0.58 and 0.66. The models generated AUC’s exceeding 0.7 for between 21% and 49% of the individual species modelled. These values represent a moderate predictive performance overall, with the accuracy of models influenced by the strength and extent of ecological gradients, species detectability, and survey design. The findings indicate that there are consistent associations between EO variables and a large number of species from diverse taxonomic groups and geographical locations. It can be reasonably assumed that a significantly higher proportion of taxa could be predicted using similar methodologies provided that sufficient field surveys are conducted to calibrate similar models. This highlights the importance of obtaining high-quality ecological data through field surveys in order to calibrate our interpretation of ecological changes observed through remote sensing. The ability to apply EO data retrospectively and prospectively also offers a valuable opportunity for long-term biodiversity monitoring across changing landscapes.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.