RiverMamba: A State Space Model for Global River Discharge and Flood Forecasting

TL;DR

RiverMamba introduces a global, high-resolution (0.05°) deep learning framework for medium-range river discharge and flood forecasting up to 7 days ahead, integrating sparse gauged data with reanalysis and high-resolution meteorological forecasts. It deploys a novel Mamba-based state-space architecture with hindcast and forecast blocks, space-filling serialization, and location-aware normalization to capture large-scale spatio-temporal river network dynamics. Trained on long-term reanalysis and fine-tuned with GRDC observations, RiverMamba consistently outperforms both AI- and physics-based baselines across reanalysis and observational data, including rare extreme floods. The approach offers a scalable, operationally relevant pathway to dense global river-discharge maps and timely flood risk assessment, though it acknowledges the need for uncertainty estimation and improved handling of human water management in real-time contexts.

Abstract

Recent deep learning approaches for river discharge forecasting have improved the accuracy and efficiency in flood forecasting, enabling more reliable early warning systems for risk management. Nevertheless, existing deep learning approaches in hydrology remain largely confined to local-scale applications and do not leverage the inherent spatial connections of bodies of water. Thus, there is a strong need for new deep learning methodologies that are capable of modeling spatio-temporal relations to improve river discharge and flood forecasting for scientific and operational applications. To address this, we present RiverMamba, a novel deep learning model that is pretrained with long-term reanalysis data and that can forecast global river discharge and floods on a $0.05^\circ$ grid up to $7$ days lead time, which is of high relevance in early warning. To achieve this, RiverMamba leverages efficient Mamba blocks that enable the model to capture spatio-temporal relations in very large river networks and enhance its forecast capability for longer lead times. The forecast blocks integrate ECMWF HRES meteorological forecasts, while accounting for their inaccuracies through spatio-temporal modeling. Our analysis demonstrates that RiverMamba provides reliable predictions of river discharge across various flood return periods, including extreme floods, and lead times, surpassing both AI- and physics-based models. The source code and datasets are publicly available at the project page https://hakamshams.github.io/RiverMamba.

Figures (60)

• Figure 1: Example of a 5-day forecast of river discharge and flood events. In early June 2024, a significant flood affected Southern Germany. While the top row shows the floods obtained from the GloFAS reanalysis, the bottom row shows the river discharge forecast by our approach. The severity of floods is categorized by the statistical flood return period, i.e., occurring every 10 years.
• Figure 2: An overview of the proposed RiverMamba model for river discharge forecasting. The model forecasts at time $t$, high-resolution river discharge maps $\textbf{X}_{dis24}^{t+1:t+L}$ from initial conditions ($\textbf{X}_{ERA\textit{5}}^{t-T:t-1}$, $\textbf{X}_{GloFAS}^{t-T:t-1}$, $\textbf{X}_{CPC}^{t-T-1:t-2}$), static river attributes ($\textbf{X}_{static}$), and meteorological forecasts ($\textbf{X}_{HRES}^{t+1:t+L}$).
• Figure 3: Illustration of the spatial scans in RiverMamba. Larger images are in the supp. material.
• Figure 4: The structure of the hindcast block and forecast block. Both use a bidirectional Mamba block and the forecast block has the same structure as the hindcast block, but it additionally incorporates meteorological forecasts (HRES) by concatenation. The forecast block also includes LOAN layers although it is not shown in Fig. \ref{['fig:Model']}
• Figure 5: Results on GloFAS reanalysis across lead times and flood return periods.