Spectral transformation of hydroclimate predictors enhances flood and drought forecasting

Hydroclimate extremes such as floods and droughts are associated with increasing socio-economic losses worldwide, reflecting their diverse spatial and temporal characteristics and growing exposure. Reliable forecasting across seasonal to interannual timescales is therefore critical for mitigating their impacts and informing risk management. Although machine learning approaches have demonstrated considerable potential, they often depend on large volumes of high-quality data and on distributional transformations of predictors, while neglecting mismatches in temporal scale and spectral structure between predictors and hydrological responses. These mismatches can mask physically meaningful signals, particularly for extremes influenced by scale-dependent climate variability.

Here we address this limitation by introducing the Wavelet System Prediction (WASP), a frequency-domain method designed to enhance hydroclimate predictors through spectral transformation. WASP employs discrete wavelet transforms to decompose predictors and responses into scale-specific components and systematically adjusts the spectral variance of predictors to align with that of the response under an assumed stationary predictor–response relationship. This approach explicitly accounts for temporal dependence and scale interactions, enabling the extraction and amplification of predictive signals that are weak or hidden in the raw predictor space.

We apply WASP to two contrasting hydroclimate extremes and spatial contexts: seasonal flood forecasting across multiple European catchments and interannual drought forecasting at the continental scale over Australia. In both applications, the proposed method substantially improves forecast skill compared to conventional methods. These results highlight the value of scale-aware, frequency-based transformations for advancing statistical modelling of hydroclimate extremes, contributing to improved hazard assessment and climate risk management.