A new vector-based global river network dataset accounting for variable drainage density

Spatial variability of river network drainage density (D-d) is a key feature of river systems, yet few existing global hydrography datasets have properly accounted for it. Here, we present a new vector-based global hydrography that reasonably estimates the spatial variability of D-d worldwide. It is built by delineating channels from the latest 90-m Multi-Error-Removed Improved Terrain (MERIT) digital elevation model and flow direction/accumulation. A machine learning approach is developed to estimate D-d based on the global watershed-level climatic, topographic, hydrologic, and geologic conditions, where relationships between hydroclimate factors and D-d are trained using the high-quality National Hydrography Dataset Plus (NHDPlusV2) data. By benchmarking our dataset against HydroSHEDS and several regional hydrography datasets, we show the new river flowlines are in much better agreement with Landsat-derived centerlines, and improved D-d patterns of river networks (totaling similar to 75 million kilometers in length) are obtained. Basins and estimates of intermittent stream fraction are also delineated to support water resources management. This new dataset (MERIT Hydro-Vector) should enable full global modeling of river system processes at fine spatial resolutions.

Automated summary

The study introduces a new global hydrography dataset that estimates the spatial variability of river network drainage density worldwide. By utilizing a machine learning approach and incorporating global climatic, topographic, hydrologic, and geologic conditions, high-quality global hydrography data is obtained to support water resources management and enable global modeling of river system processes.