Brum-Bastos, V; Long, J; Church, K; Robson, G; de Paula, R; Demsar, U (2020). Multi-source data fusion of optical satellite imagery to characterize habitat selection from wildlife tracking data. ECOLOGICAL INFORMATICS, 60, 101149.

Wildlife tracking data allow monitoring of how organisms respond to spatio-temporal changes in resource availability. Remote sensing data can be used to quantify and qualify these variations to understand how movement is related to these changes. The use of remote sensing data with concurrent high levels of spatial and temporal detail may hold potential to improve our understanding of habitat selection. However, no current orbital sensor produces data with simultaneous high temporal and high spatial resolution, therefore alternative methods are required to generate remote sensing data that matches the high spatial-temporal resolution of modern wildlife tracking data. We present an analytical framework, not yet used in movement ecology, for data fusion of optical remote sensing data from multiple satellites and wildlife tracking data to study the impact of seasonal vegetation patterns on the movement of maned wolves (Chrysocyon brachyurus). We use multi-source data fusion to combine MODIS data with higher spatial resolution data (ASTER, Landsat 4-5-7-8, CBERS 2-2B) and create a synthetic NDVI product with a 15 m spatial detail and daily temporal resolution. We also use the higher spatial resolution data to create a multi-source NDVI product with same level of spatial detail but coarser temporal resolution and data from MODIS to create a single-source NDVI product with high temporal resolution but coarse spatial resolution. We combine the three different spatial-temporal resolution NDVI products with GPS tracking data of maned wolves to create step-selection functions (SSF), which are models used in ecology to investigate and predict habitat selection by animals. The SSF model based on multi-source NDVI had the best performance predicting the probability of use of visited locations given its NDVI value. The SSF based on the raw MODIS NDVI product, one which is commonly employed by ecologists, had the poorest performance for our study species. These findings indicate that, in contrast with current practice in movement ecology, a detailed spatial resolution of contextual environmental variable may be more important than a detailed temporal resolution, when investigating wildlife habitat selection regarding vegetation, although this result will be highly dependent on species. The choice of data set should therefore take into account not only the scale of movement but also the spatial and temporal scales at which dynamic environmental variables are changing.