Advancing drought detection and management using ML enhanced impact-based drought indexes

Drought is a slowly developing natural phenomenon that can occur in all climatic zones and propagates through the entire hydrological cycle with long-term socio-economic and environmental impacts. Intensified by anthropogenic climate change, drought has become one of the most significant natural hazards in Europe. Different definitions of drought exist, i.e. meteorological, hydrological, and agricultural droughts, which vary according to the time horizon and the variables considered. Just as there is no single definition of drought, there is no single index that accounts for all types of droughts. Consequently, capturing the evolution of drought dynamics and associated impacts across different temporal and spatial scales remains a critical challenge. In this work, we first analyze different state-of-the-art standardized drought indexes in terms of their ability in detecting drought events at the pan-European scale, using hydro-meteorological variables from the E-HYPE hydrological model and forced with the HydroGFD v2.0 reanalysis dataset over the period 1993-2018. The findings suggest the need of adjusting the formulation of traditional drought indexes to better capture and represent drought-related impacts. Specifically, here we use the FRamework for Index-based Drought Analysis (FRIDA), a Machine Learning approach that allows the design of site-specific indexes to reproduce a surrogate of the drought impacts in the considered area, here represented by the Fraction of Absorbed Photosynthetically Active Radiation Anomaly (FAPAN). FRIDA builds a novel impact-based drought index combining all the relevant available information about the water circulating in the system identified by means of a feature extraction algorithm. Our results reveal a general pattern among different indexes, that Southern England, Northern France, and Northern Italy are the regions with the highest number of drought events, whereas the areas experiencing longest drought durations are instead the Baltic Sea region and Normandy. Clustering the 35,408 European basins according to dominant hydrologic processes reveals that the variables mainly controlling the drought process vary across clusters. Similarly, we obtain diverse correlation between standardized drought indexes and the FAPAN in different clusters. Numerical results also show that, in one of the worst cases (cluster 10), the FRIDA index increases the correlation with FAPAN from 0.16 to 0.69. Lastly, the FRIDA indexes are computed for different climatic projections to investigate future trends in drought impacts.  Results show divergence with respect to the trends of the standardized drought indexes, with correlation values below 0.30. In conclusion, these findings can contribute in advancing drought-related climate services by enabling the analysis of projected drought impacts.