The Cabrera vole (Microtus cabrerae) is a rare Iberian endemism, classified as “Near-threatened” by IUCN, and “Vulnerable” in Portugal and Spain. The species has a restricted range and a fragmented distribution, occurring mostly in patches of tall and dense wet grasslands in a structured meta-population system. Spatial and temporal variation on species’ resource availability poses difficulties when it is necessary to define which specific areas are most important to protect. On this issue, species distribution models (SDMs) are often used to obtain detailed geographical distribution of species, which are then used to define effective conservation and monitoring actions. However, SDMs applications on Cabrera vole, and other rare species, at a local or regional scale are still challenging, likely due to their low detectability, narrow distribution, and short-term occupancy of suitable patches. In addition, most available digital environmental information may not reflect spatial and temporal ecological conditions required for the Cabrera vole occurrence. Nowadays, remote-sensing provides information on landscape structure and associated biophysical products at areas on able time frequency and at an unreleased fine spatial resolution, which might be a solution to increase the accuracy of models, as availability of resources and its variation through time is better described. Our aim was to investigate the usefulness of ESA Sentinel-2 products for the prediction of suitable habitat patches for the Cabrera vole in a Mediterranean agro-silvopastoral system. We aimed to 1) identify which Sentinel-2 derived predictors are best surrogates for occupied habitat patches; and 2) quantify its importance when compared with other classic/static predictors. The study was conducted in the Alentejo region, Southern Portugal, in which herbaceous patches were surveyed in Spring 2017 and Autumn 2018, through presence signs and then classified into presence/absence. Dataset was filtered to retain true absences by excluding patches classified as absences with potential habitat. Thereafter, we calculated 85 predictors from Sentinel-2 images as well as from other sources (Topographical information and Landscape element proximity). Specifically, each satellite image was composed of 10 multispectral bands, combined to describe spectral, biophysical and structural landscape properties for each of the two seasons. To identify predictors to retain, their ecological importance was quantified by utilizing Cabrera vole presence/absence data as response variable through a Random forest model accounting for multi-predictor relationships. A total of 11 uncorrelated predictors were identified as important, namely a distance-based measure, road proximity (~27% importance), while from remote-sensing data were NDI45 “Spring”, SWIR “Autumn”, RAO’s Q “Spring”, NDRE1 “Autumn”, Green “Autumn”, BI2 “Spring”, GLMC_Cor “Spring”, RAO’s Q “Autumn”, Blue “Spring”, GLMC_Con “Autumn”, together contributing with ~73% of importance. Cabrera vole presence is more likely in areas close to roads, and associated to remote-sensing indices translating vegetation with intermediate chlorophyll contents and water retention, and more local scale vegetation heterogeneity. Road verges can act as relatively stable refuges in Mediterranean landscapes, especially when the surrounding matrix becomes environmentally prohibitive, such as when under intensive agriculture or livestock farming practices. Our approach is useful for identifying undiscovered suitable areas, and for planning the placement of mitigation/conservation measures along the road verges as well under other priority areas.
Murcia
