Conservation interventions for threatened species must be based on accurate assessments of the effects of anthropogenic pressures on habitat suitability. We used multiscale multivariable species‐distribution modeling to evaluate habitat suitability for an Asian elephant (Elephas maximus) population in Shangyong Reserve, Yunnan Province, southwestern China. We investigated the scales at which measurements of environmental variables best reflected elephant habitat selection, and examined whether these responses changed over 2 decades (2000–2010 and 2011–2020) in response to 20 environmental variables, including 14 variables reflecting landscape fragmentation, the extent of buildings, and transport infrastructure. Elephant presence was sensitive to the scale of each variable, and the effects differed among variables within and between decades. More than half of the variables influenced elephant presence at coarse scales of 8 or 16 km, including 12 variables reflecting anthropogenic pressures in 2000–2010 and 10 in 2011–2020. Overall, multivariate models with variables at their optimal scales had higher discrimination than models at uniformly fine scales of 1 km or 2 km. The extent of suitable habitat for elephants declined by 24% over 2 decades. Less than half of elephant habitat was located within Shangyong Reserve (49% in 2000–2010, 40% in 2011–2020), indicating the importance of managing suitable habitat beyond reserve boundaries. Roads and buildings reduced the probability of elephant presence, with effects that extended beyond their immediate footprint. We advocate that infrastructure be planned with buffers, ≥8 km wide, between roads or buildings and core elephant habitat. Multiscale multivariable species‐distribution modeling should be employed to ensure that all suitable habitat for the remaining fragmented elephant populations in Yunnan is identified, mapped, and protected.