Applying machine learning to biodiversity management: Lessons from two California biodiversity hotspots

Highlights

• Used high resolution AVIRIS-NG hyperspectral imagery and machine learning to distinguish invasive and native plant species.
• Achieved 87% overall accuracy in mapping invasive iceplant at Dangermond Preserve.
• Found that May imagery provided the strongest spectral separation between vegetation types due to seasonal differences.
• Demonstrated how remote sensing and artificial intelligence can support conservation planning, monitoring, and resource allocation.

Abstract

Mapping invasive species and native vegetation is crucial for effective land management and conservation, especially in pervasively invaded landscapes. It is often difficult to differentiate between species using traditional remote sensing methods. However, advances in imaging spectroscopy and machine learning techniques offer a solution by leveraging the often-distinct spectral signatures of different plant species. This study explores the potential of high-resolution, multi-temporal hyperspectral data from the Airborne Visible-Infrared Imaging Spectrometer - Next Generation (AVIRIS-NG) to map vegetation on two California preserves: The Nature Conservancy's Jack and Laura Dangermond Preserve and the University of California Natural Reserve System (UCNRS) Sedgwick Reserve. We employed machine learning algorithms to address two distinct case studies: mapping invasive iceplant (Carpobrotus edulis) at Dangermond Preserve and identifying native perennial grass patches (Stipa spp.) at Sedgwick Reserve. By analyzing AVIRIS-NG data from February to May 2022, we found that May imagery provided the highest spectral separability for both species, likely due to phenological differences. Using a support vector machine classifier, we achieved high accuracy in mapping iceplant (87% overall accuracy). We achieved an accuracy of 59.4% for mapping native grasses in annual grass regions. Our results show that integrating multi-temporal hyperspectral remote sensing with machine learning can provide land managers with the high-resolution data needed to prioritize resource allocation, monitor management efficacy, and enable informed decision-making in a changing landscape. This collaborative, technology-driven approach provides a valuable model for biodiversity conservation.

Citation

Miner, Kimberley, Latha Baskaran, Frank W. Davis, Kelly Easterday, Mark Reynolds, Sara Tjossem, Kristen Zumdahl, et al. 2026. “ Applying Machine Learning to Biodiversity Management: Lessons from Two California Biodiversity Hotspots.” Ecosphere 17(7): e70709. https://doi.org/10.1002/ecs2.70709