Reliability of Climate Information to Forecast Season-Ahead Flood Quantiles for Indian Catchments

Forecasting floods (peak flows/quantiles) with significant lead time is crucial for effective water resources management. Traditionally, it has been carried out by forcing meteorological drivers onto the hydrological models. However, season-ahead flood forecasting remains challenging due to the limitations of weather forecasting models and the complexities associated with multiple model-chain linkages. Thus, to circumvent this, we applied a climate-informed approach to forecast season-ahead flood quantiles. Briefly, a climate-informed model comprises 1) selection of predictands, 2) identification of suitable large-scale climate predictors that control the predictands, and 3) derivation of a statistical link between predictands and predictors. In our study, we condition the probability distribution parameters of flood samples with large-scale climate predictors, focusing specifically on sea surface temperature (SST) patterns. The rationale behind this approach lies in the established linkage of SST in the Pacific and Indian Oceans to the Indian Monsoon system. To minimise the anthropogenic signals, we restricted our analysis to the gauging stations without significant reservoir influences by filtering the stations with reservoir indices less than 0.1. Both linear and nonlinear relations between the climate predictors and predictands have been applied in this study. Bayesian inference is used to estimate the parametric values of the Climate-Informed model. Furthermore, the selection of the suitable climate predictor and the nature of their relationship to a specific gauge is based on the widely applicable selection criterion (WAIC). WAIC computes log posterior predictive density and adjusts the overfitting using the effective number of parameters; the model with the least WAIC value is preferred. We assessed the skill of the climate-informed model on flood quantile forecasting by performing a leave-one-out cross-validation technique. Various performance metrics, including both deterministic and probabilistic measures, have been used to assess the prediction skill of the model in reference to the stationary model. Overall, our results suggest that for the majority of the gauges, climate indices have the potential to forecast flood-quantiles season ahead. While this initial forecast can inform decision-makers regarding expected flood quantiles, it is recommended that this method be complemented with traditional approaches that account for local catchment behaviour.