QDeepGR4J: Quantile-based ensemble of deep learning and GR4J hybrid rainfall-runoff models for extreme flow prediction with uncertainty quantification

TL;DR

The paper addresses the challenge of predicting extreme streamflows with quantified uncertainty in rainfall–runoff modelling. It extends DeepGR4J by integrating quantile regression into an ensemble framework (QDeepGR4J), calibrates GR4J with differential evolution, and leverages hybrid features to produce multi-quantile, multi-step forecasts (tau in {0.05, 0.50, 0.95}) for 3-day horizons, plus a flood risk indicator derived from a GEV threshold. Key contributions include improved predictive accuracy and uncertainty interval quality, demonstrated across CAMELS-Aus catchments, and a qualitative flood-warning capability that informs early flood risk assessment. The approach offers a practical, computationally efficient alternative to Bayesian methods for operational flood forecasting and water-resource management under climate variability, with generalisation across Australian regions.

Abstract

Conceptual rainfall-runoff models aid hydrologists and climate scientists in modelling streamflow to inform water management practices. Recent advances in deep learning have unravelled the potential for combining hydrological models with deep learning models for better interpretability and improved predictive performance. In our previous work, we introduced DeepGR4J, which enhanced the GR4J conceptual rainfall-runoff model using a deep learning model to serve as a surrogate for the routing component. DeepGR4J had an improved rainfall-runoff prediction accuracy, particularly in arid catchments. Quantile regression models have been extensively used for quantifying uncertainty while aiding extreme value forecasting. In this paper, we extend DeepGR4J using a quantile regression-based ensemble learning framework to quantify uncertainty in streamflow prediction. We also leverage the uncertainty bounds to identify extreme flow events potentially leading to flooding. We further extend the model to multi-step streamflow predictions for uncertainty bounds. We design experiments for a detailed evaluation of the proposed framework using the CAMELS-Aus dataset. The results show that our proposed Quantile DeepGR4J framework improves the predictive accuracy and uncertainty interval quality (interval score) compared to baseline deep learning models. Furthermore, we carry out flood risk evaluation using Quantile DeepGR4J, and the results demonstrate its suitability as an early warning system.

Figures (6)

• Figure 1: GR4J hydrologic model architecture
• Figure 2: DeepGR4J-Extreme framework based on conditional ensembles catering to extreme values of streamflow
• Figure 3: Stations lying in different regions across Australia
• Figure 4: QDeepGR4J-LSTM predictions for two stations located in South Australia
• Figure 5: Comparison of streamflow quantile predictions from quantile-based LSTM ensemble and QDeepGR4J-LSTM ensemble for Pascoe River at Fall Creek station (102101A).