Global reconstruction of gridded aridity index and its spatial and temporal characterization from 2003 to 2022

Aridity index (AI) is an effective estimator of drought status, and spatiotemporally continuous long-term AI dataset is critical for drought assessment and applications. Due to the spatial heterogeneity of global climate and topography, there exist significant uncertainties of AI estimates in areas with sparse ground observations, and high-resolution global AI estimation remains a challenge. In this study, we propose an LSTM-based approach to model the nonlinear intra-annual relationship between satellite-derived data and AI and enhance model performance through ensemble learning by leveraging MODIS data at different observation times. A long-term annually gridded global AI dataset is generated at a resolution of 0.05° × 0.05° from 2003 to 2022. Validation against the Global Surface Summary of the Day database yields biases, root mean squared errors and coefficients from −0.04 to 0.02, 0.19 to 0.86, and 0.62 to 0.83 across different continents. Comparisons with AI estimates based on Climatic Research Unit or ERA5-Land datasets further demonstrate the high accuracy of our AI estimates. Preliminary analysis reveals a global wetting trend over the past two decades. This dataset offers valuable support for research on dryland ecosystems, agriculture, and climate change, offering critical insights to address global environmental and sustainability challenges.