Global spatiotemporal variation analysis and AI prediction of terrestrial high-temperature droughts

In recent years, the frequency and intensity of compound high-temperature drought events have significantly increased on a global scale, posing severe challenges to agricultural production and ecological environments. To elucidate the spatiotemporal variation patterns of such extreme events and enhance prediction accuracy, this study systematically analyzed the distribution patterns and evolutionary trends of terrestrial high-temperature drought events based on the Standardized Precipitation Index (SPI) and Standardized Temperature Index (STI), utilizing global multi-source observational data from 1980 to 2022. The results indicate that regions such as Brazil, West Africa, the Arabian Desert, South Asia, and Mexico exhibit particularly prominent high-temperature trends, while precipitation significantly decreases in parts of South America, South Asia, Libya, western United States, eastern Canada, and southwestern China. Additionally, the recurrence intervals of high-temperature droughts in Venezuela, Brazil, northern Russia, Iran, and southwestern China have markedly shortened. To further improve prediction accuracy, this study employed wavelet transforms in combination with three deep learning methods—Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory Network (LSTM)—to develop multi-scale predictive models for SPI and STI. The results demonstrate that all three models achieved coefficients of determination (R2) exceeding 0.98 for SPI and STI predictions, with mean absolute errors (MAE) below 0.036 and 0.07, root mean squared errors (RMSE) below 0.09 and 0.05, respectively, indicating high reliability in extreme event prediction. Forecasts for 2019–2026 suggest that the frequency and intensity of compound high-temperature drought events will generally continue to rise, providing critical references for subsequent climate risk assessments and agricultural disaster prevention and control.