Forecasting flood magnitudes (peak flows or quantiles) with significant lead time is quintessential for flood control measures and water resource management. Traditional approaches rely on meteorological information to derive flood quantiles. However, these approaches are often computationally intensive and limited in their ability to forecast floods at seasonal lead times. To address these limitations, our study adopts a climate-informed flood forecasting approach, i.e., incorporating large-scale climate indices, to forecast season-ahead flood quantiles in Indian catchments. The rationale behind this approach lies in the established relationship between large-scale climate indices and the Indian Monsoon system. Briefly, climate-informed flood frequency analysis (CI-FFA) comprises (1) selection of predictands (i.e., seasonal flood quantiles); (2) identification of suitable climate predictors that influence the predictands; and (3) developing a statistical relationship between predictands and predictors. To reduce the anthropogenic signals, we focus on catchments with minimal reservoir influence. We implement CI-FFA models of varying complexity, incorporating different numbers of climate covariates. Bayesian statistical inference is employed for parameter estimation. The selection of the climate predictor and the nature of their relationship to the predictand in a specific catchment are based on the widely applicable information criterion. Model performance is evaluated using leave-one-out cross-validation, employing both deterministic and probabilistic skill metrics. Results show that the CI-FFA outperforms the traditional stationary model (T-FFA) in forecasting seasonal flood quantiles. Although models with more covariates and higher complexity tend to produce wider uncertainty bands, the overall uncertainty for higher return periods remains comparable to that of T-FFA. Importantly, the amplification factor, defined as the ratio between CI-FFA and T-FFA flood quantile estimates, often exceeds 1.5, indicating the amplified risk in the CI-FFA model in comparison to the T-FFA model. These findings highlight the potential of large-scale climate indices to enhance seasonal flood quantile forecasting at many gauges across the region.
Ganapathy et al. (Thu,) studied this question.