Accurate measurement of lake surface area is essential for understanding eco-hydrological processes in arid regions, yet long-term records are often limited by cloud contamination, seasonal ice cover, and data gaps. In this study, we developed an optimized extraction framework that integrates seasonal index selection, adaptive thresholding, maximum connectivity analysis, and mutual information – based gap filling to construct a continuous monthly lake area series for Bahannao Lake from 1984 to 2024. This method effectively addressed common challenges in remote sensing water extraction and provided reliable long-term lake dynamics in a data-scarce desert region. Based on the reconstructed time series, we examined the multi-factor drivers of lake evolution using an XGBoost model combined with climatic and energy-balance variables. Results reveal pronounced interannual and seasonal variability: precipitation dominates lake expansion in spring and summer, while shortwave radiation is the main driver of evaporation in autumn and winter, even under cold conditions. Long-term trends indicate a shift in controlling mechanisms – from humidity and precipitation decline (1984–1999), to increased radiation and humidity variability (2000–2014), and finally to intensified sensible heat flux and potential evapotranspiration (2015–2024).Our findings highlight the nonlinear and evolving interactions between hydro-climatic factors regulating arid-region lakes. The proposed framework provides a robust approach for generating long-term lake records, advancing understanding of eco-hydrological responses to climate change, and offering scientific support for water resources management and adaptation in arid regions.