Integrating satellite data and field surveys to predict bird communities in the Peruvian Amazon.

Tropical forests harbour some of the highest biodiversity on Earth but are undergoing rapid loss and degradation. In the Amazon more than one-third of forests have been altered through human activities, with major implications for wildlife communities. While Earth observation satellites effectively monitor forest cover at scale, it remains unclear how well satellite-derived variables capture variation in bird communities. We tested whether Landsat reflectance and vegetation indices can predict bird species occurrence and community composition in the Peruvian Amazon. We analysed 3,129 point counts and mist-net bird surveys conducted over 16 years in the Tambopata Forest, south-eastern Peru. As predictors we compared the effectiveness of remote sensing derived surface reflectance and vegetation indices (e.g., NDVI and tasselled cap), with traditional land-type and forest cover descriptors. Species occurrence probabilities and community composition of 135 frequently recorded bird species were estimated using multi-species occupancy models that account for imperfect detection. Models using Landsat reflectance and vegetation indices outperformed those based on habitat categories in predicting species occupancy (mean AUC = 0.68 vs 0.58). They also achieved high predictive accuracy (AUC > 0.7) for more species (49 compared with 20). However, low detection rates across surveys limited all models' ability to accurately estimate full community composition and to detect change over time. Our results demonstrate that satellite-derived variables can improve predictions of bird occurrence compared with habitat categories, but their effectiveness depends strongly on survey design and species detectability. Integrating remote sensing with well-structured field surveys provides a scalable approach to monitoring biodiversity trends in tropical forests.