Drought is a persistent environmental challenge with profound impacts on water resources, agriculture, and ecosystems, particularly in semi-arid regions. This study investigates the spatial and temporal dynamics of drought in the Southern Telangana Zone (STZ) using the Standardized Precipitation Evapotranspiration Index (SPEI) across 12 districts over the past 44 years. Results highlight considerable variability in drought intensity and frequency, with extreme value distribution analysis indicating an increasing risk of severe drought events in the future. To enhance predictive capability, several time series models were developed, including ARIMA, STARMA, and a novel two-stage STARMA-TDNN framework that integrates spatiotemporal linear modeling with nonlinear machine learning. The proposed triangular fuzzy STARMA-TDNN model achieved the highest efficiency, reducing training and testing mean squared error by more than 70 % compared to alternative approaches. While ARIMA captured only temporal linear patterns and STARMA addressed linear spatiotemporal dependencies, the two-stage STARMA-TDNN framework effectively represented both linear and nonlinear spatiotemporal drought dynamics. The Diebold-Mariano test further confirmed the statistical superiority of the two-stage model. These findings underscore the potential of advanced hybrid models for reliable drought forecasting and provide a scientific basis for location-specific drought management strategies in the STZ.