Domain knowledge-driven variational recurrent networks for drought monitoring

• Published in: Remote sensing of environment 2024-09, Vol.311, p.114252, Article 114252
• Main Authors: Zhang, Mengxue, Fernández-Torres, Miguel-Ángel, Camps-Valls, Gustau
• Format: Article
• Language: English
• Subjects: Domain knowledge; Drought monitoring; Hybrid modelling; Long short-term memory (LSTM); Variational encoder-decoder (VED)
• ISSN: 0034-4257
• DOI: 10.1016/j.rse.2024.114252

In the context of climate change, droughts, increasingly frequent and severe, necessitate effective monitoring. Existing methods, such as drought indices and data-driven models, face important limitations. Drought indices are built on prior expert knowledge but lack calibration based on actual drought events, while data-driven models prioritize goodness of fit over real event identification, undermining their credibility and generalization, and also struggling to generalize from regional to large-scale contexts. To address these challenges, here we introduce a hybrid machine learning framework for time series that combines domain knowledge and observational data in a variational recurrent neural network. The network models the joint distribution of total precipitation, air temperature, and real drought events, providing accurate predictions and uncertainty estimates. Extensive experiments focusing on a wide range of European drought events from 2011 to 2018 consistently show that our hybrid model surpasses both drought indices and data-driven models in terms of accuracy in drought detection, underlining its effectiveness, robustness, and stability. Our model achieves the best ROC-AUC (%) results in Afghanistan (79.7 ± 0.5), Italy (84.3 ± 0.6), Russia (89.4 ± 0.2), Europe-0 (84.3 ± 0.1), and Europe-1 (82.8 ± 0.4), effectively capturing the starting and ending times of drought events with lower uncertainty, and also generalizing better for unseen locations. •We propose DK-VRN, a novel hybrid framework for drought monitoring.•DK-VRN takes ECVs as input and integrates multi-scalar SPEI as domain knowledge prior.•DK-VRN is end-to-end supervised by historical drought events.•DK-VRN can detect droughts over diverse timeframes and across continental scales.•Experiments on European events from 2011 to 2018 show the effectiveness of DK-VRN.