A simplified drought indicator based on high-resolution GRACE terrestrial water storage anomalies

Abstract

Hydrological drought indices based on meteorological data do not fully reflect impacts on hydrological systems, and the coarse spatial resolution of GRACE data limits its usefulness for local-scale drought assessment. To address this, we developed fine-scale drought indices based on Gravity Recovery and Climate Experiment (GRACE)-derived terrestrial water storage anomalies (TWSA) using a statistical downscaling approach. This was achieved by employing a Random Forest machine learning algorithm to integrate key water budget terms (i.e., precipitation, evapotranspiration, runoff and deep drainage) into the original GRACE grids to achieve a drought index at 5 km spatial resolution. The resulting downscaled GRACE drought index (dGdi) is effective for localized drought predictions, providing a comprehensive picture of hydrological and climatic conditions over major river basins in Australia. Application of this downscaled drought index over the Canning Basin, Western Australia, reveals long-term drought evolutions indicating that the region is at a risk of a permanent shift in ecosystem composition (e.g., dominance of drought-tolerant invasive species), land degradation and aquifer depletion. Overall, we found that global climate indices have weak influences on Australia's drought progression. The Back Propagation Neural Network confirmed these indices contribute to drought occurrence in the Canning (r = 0.37) and Central Eromanga (r = 0.36) Basins. The dGdi developed in this study supports local-scale drought assessment by capturing changes in key biophysical indicators and effectively highlighting intensifying drought patterns. Given its reliance on widely available water budget variables and its adaptability to diverse hydrological settings, the dGdi can be extended to other regions beyond Australia for enhanced drought monitoring and water resource management.