Spatially distributed estimates of Manning’s roughness within floodplain areas of the conterminous United States [scripts and datasets]

Reach-Scale Floodplain Manning’s Roughness Dataset for the Conterminous United States Derived from Remote Sensing and Machine Learning
Gabriel Barinas1,2, Stephen Good1,2, Samuel Rivera1,3
1Water Resources Graduate Program, Oregon State University, Corvallis OR, USA
2Department of Biological and Ecological Engineering, Oregon State University, Corvallis OR, USA
3School of Civil and Construction Engineering, Oregon State University, Corvallis OR, USA
Correspondence to: Gabriel Barinas ([e-mail])

Floodplain roughness, quantified through Manning’s coefficient n, is a critical parameter in hydrological models for predicting flood dynamics and managing water resources. Traditional methods to determine n rely on generalized land cover types and often fail to capture the spatial and structural variability of floodplains, resulting in limited understanding of floodplain roughness variation at regional scales. This study integrates high-resolution remotely sensed canopy height and biomass data from NASA’s Global Ecosystem Dynamics Investigation with other spatially distributed data to map Manning’s roughness at reach scales across the conterminous United States. After evaluation of six machine learning models, the best performing approach (Random Forest) was trained on 4,927 roughness estimates from 804 sites and applied to estimate n at 17.8 million reaches within the National Hydrography Database (NHDPlus HR). These n estimates have an R² of 0.51, a root mean squared error of 0.084, and a mean absolute percentage error of 122, capturing spatial variability in floodplain roughness that traditional static methods fail to represent. We find the sparsely vegetated southwest US region exhibits the lowest mean roughness, while the Appalachian region and parts of the southeast US exhibit moderate to high mean values due to denser and more varied floodplain vegetation. Canopy height and biomass were identified as influential non-linear predictors of n, highlighting the importance of vegetation structure on floodplain roughness. This integration of remote sensing data with machine learning models provides spatially distributed estimates of Manning’s n that elucidate patterns in floodplain roughness variability from reach to continental scales. The dataset and companion code are openly available here.