Mapping small mammal optimal habitats using satellite-derived proxy variables and species distribution models

• Published in: PloS one 2023-08, Vol.18 (8), p.e0289209
• Main Authors: Marston, Christopher, Raoul, Francis, Rowland, Clare, Quéré, Jean-Pierre, Feng, Xiaohui, Lin, Renyong, Giraudoux, Patrick
• Format: Article
• Language: English
• Subjects: Abundance; Advance Directives; Agriculture; Altitude; Analysis; Animals; Arvicolinae; Bats; Benchmarking; Biology and Life Sciences; Digital mapping; Disease transmission; Dispersal; Earth Sciences; Ecology and Environmental Sciences; Ellobius tancrei; Engineering and Technology; Genetics; Geographical distribution; Grasslands; Habitats; Humans; Information management; Land cover; Land use; Landscape; Life cycles; Life Sciences; Mammals; Marmota; Modelling; Murinae; Phenology; Physical Sciences; Plant species; Population biology; Population dynamics; Populations and Evolution; Prey; Productivity; Proxy; Remote sensing; Research and Analysis Methods; Seed dispersal; Small mammals; Soil dispersion; Soil structure; Species composition; Variables; Vegetation; Watercourses
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0289209

Small mammal species play an important role influencing vegetation primary productivity and plant species composition, seed dispersal, soil structure, and as predator and/or prey species. Species which experience population dynamics cycles can, at high population phases, heavily impact agricultural sectors and promote rodent-borne disease transmission. To better understand the drivers behind small mammal distributions and abundances, and how these differ for individual species, it is necessary to characterise landscape variables important for the life cycles of the species in question. In this study, a suite of Earth observation derived metrics quantifying landscape characteristics and dynamics, and in-situ small mammal trapline and transect survey data, are used to generate random forest species distribution models for nine small mammal species for study sites in Narati, China and Sary Mogul, Kyrgyzstan. These species distribution models identify the important landscape proxy variables driving species abundance and distributions, in turn identifying the optimal conditions for each species. The observed relationships differed between species, with the number of landscape proxy variables identified as important for each species ranging from 3 for Microtus gregalis at Sary Mogul, to 26 for Ellobius tancrei at Narati. Results indicate that grasslands were predicted to hold higher abundances of Microtus obscurus, E. tancrei and Marmota baibacina, forest areas hold higher abundances of Myodes centralis and Sorex asper, with mixed forest-grassland boundary areas and areas close to watercourses predicted to hold higher abundances of Apodemus uralensis and Sicista tianshanica. Localised variability in vegetation and wetness conditions, as well as presence of certain habitat types, are also shown to influence these small mammal species abundances. Predictive application of the Random Forest (RF) models identified spatial hot-spots of high abundance, with model validation producing R2 values between 0.670 for M. gregalis transect data at Sary Mogul to 0.939 for E. tancrei transect data at Narati. This enhances previous work whereby optimal habitat was defined simply as presence of a given land cover type, and instead defines optimal habitat via a combination of important landscape dynamic variables, moving from a human-defined to species-defined perspective of optimal habitat. The species distribution models demonstrate differing distributions and abundances of host