Beyond Static Priors: Unlocking High-Resolution Dynamic River Networks via a SWOT-Driven Machine Learning Pipeline

The advent of the Surface Water and Ocean Topography (SWOT) mission has ushered in a new era of global hydrology, providing unprecedented high-resolution 2D observations of water surface elevation and extent. However, standard processing chains for hydrological applications rely heavily on static a priori databases such as the SWOT River Database (SWORD). This reliance introduces significant biases, as static centerlines fail to capture the morphological dynamism of rivers, and divergent flows such as bifurcations, braided reaches, multi-threaded systems, and artificial canals. This leads to reduced accuracy in hydrological and hydraulic modelling, flood forecasting, and water resources management.

In this study, we present a proof-of-concept workflow that uses the SWOT mission’s Pixel Cloud (PIXC) dataset to generate a high-resolution (~20 m), vector-based river network with flow direction. Moving beyond the constraints of the static SWORD, our approach utilizes SWOT-derived surface-water features as inputs for Random Forest and XGBoost classifiers. The resulting classification undergoes a semi-automated post-processing chain including rasterization, skeletonization, cleaning, and vectorization to reconstruct the network topology, with flow direction inferred directly from SWOT water-surface elevation measurements.

We evaluated this methodology in the Indus River Basin (Pakistan), a system distinguished by its dense network of artificial channels and high flood frequency. Results show that the method identifies missing river segments, divergent channels, and small-scale artificial waterways not represented in existing global river datasets, and extends significantly beyond the SWORD database. This work highlights the potential of SWOT Pixel Cloud data to move beyond static river representations and support dynamic river network generation for hydrological applications. Future efforts will focus on full automation and global scalability, as well as integration with operational hydrological and flood forecasting systems. The proposed framework provides a scalable pathway toward next-generation river network products that better exploit SWOT’s unique observational capabilities.