Influence of land cover, native shrub availability, and predator presence on the habitat selection of Meriones hurrianae in the Thar Desert : Data for zero-inflated Poisson and resource selection function

Land use and land cover changes in the Thar Desert have contributed to a marked decline in regional biodiversity. This study examined the habitat selection behavior of the Indian desert gerbil (Meriones hurrianae) using statistical analysis of data obtained from field-based ecological surveys employing a random sampling approach. Field-based ecological surveys were conducted across ~650 km² of the Thar Desert, Jaisalmer district, Rajasthan (26°42′N, 70°31′E), during the monsoon and post-monsoon seasons (August–November) in the year of 2021. 

The study area was systematically classified into distinct land use and land cover categories using a combination of field observations, land ownership and vegetation characteristics. Five land use (LU) categories are protected areas (PA), unprotected rangelands (UPA), mixed land use areas (MLA), formerly cultivated areas (FCA), and cultivated areas (CA). Additionally, eight land cover (LC) categories were defined based on vegetation characteristics: bare land (BL), bare land with vegetation (BLV), grassland (GL), shrubland (SL), mixed vegetation (MV), formerly cultivated areas covered by shrubs (FCV), agricultural land with native vegetation (ALV), and agricultural land (AL).

Landscape features, ecological factors, and human interventions collectively influence rodent distribution and abundance. Therefore, we collected data on Indian desert gerbil burrow density, alongside associated land use and cover classifications, the occurrence of native shrub species, potential predators, and evidence of grazing activity.

Using Factor Analysis of Mixed Data (FAMD) to inform a Zero-Inflated Poisson (ZIP) model, we identified key environmental variables influencing the presence and abundance of gerbil burrows. These included land cover types, predator presence, the availability of native shrubs, and grazing pressure. Additionally, model selection based on the Akaike Information Criterion (AIC) within a Resource Selection Function (RSF) framework indicated that two competing models significantly explained gerbil presence probabilities. To optimize the RSF models, the dataset was randomly split into training (70%) and testing (30%) subsets. Model performance was assessed using measures obtained from confusion matrices, such as sensitivity (true positive rate), specificity (true negative rate), and Receiver Operating Characteristic (ROC) curves. To ensure robustness, the procedure was repeated three times, and mean values of sensitivity, specificity, accuracy, and AUC were reported. Since random splitting can produce different training and testing sets each time, we have also documented the three pairs of datasets generated during the analysis.