Status and prospects for drought forecasting: opportunities in artificial intelligence and hybrid physical–statistical forecasting

• DOI: 10.1098/rsta.2021.0288
• Publication Date: 2022-10-23T23:05:22Z
• Authors: A. AghaKouchak; B. Pan; O. Mazdiyasni; M. Sadegh; S. Jiwa; W. Zhang; C. A. Love; S. Madadgar; S. M. Papalexiou; S. J. Davis; K. Hsu; S. Sorooshian
• Keywords: Robustness

Despite major improvements in weather and climate modelling substantial increases remotely sensed observations, drought prediction remains a challenge. After review of the existing methods, we discuss research gaps opportunities to improve prediction. We argue that current approaches are top-down, assuming process(es) and/or driver(s) known—i.e. starting with model then imposing it on observed events (reality). With help an experiment, show there develop bottom-up models—i.e. from reality (here, events) searching for model(s) work. Recent advances artificial intelligence machine learning provide significant developing forecasting models. Regardless type (e.g. learning, dynamical simulations, analogue based), need shift our attention robustness theories outputs rather than event-based verification. A focus towards quantifying stability uncertainty models, goodness fit or reproducing past, could be first step this goal. Finally, highlight advantages hybrid statistical models improving This article is part Royal Society Science+ meeting issue ‘Drought risk Anthropocene’.