An artificial neural network to estimate the foliar and ground cover input variables of the Rangeland Hydrology and Erosion Model

Authors: SAEEDIMOGHADDAM, M. - University Of California, Davis; NEARING, G. - Google; Goodrich, David - Dave; HERNANDEZ, M. - University Of Arizona; GUERTIN, D.P. - University Of Arizona; METZ, L. - Natural Resources Conservation Service (NRCS, USDA); WEI, H. - University Of Arizona; PONCE-CAMPOS, G. - University Of Arizona; BURNS, I.S. - University Of Arizona; McCord, Sarah; NEARING, M.A. - Retired ARS Employee; Williams, Christopher - Jason; HOUDESHELL, C.A. - Natural Resources Conservation Service (NRCS, USDA); RAHMAN, M. - University Of California, Davis; Meles, Menberu; BARKER, S. - Retired Non ARS Employee

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/17/2024
Publication Date: 2/13/2024
Citation: Saeedimoghaddam, M., Nearing, G., Goodrich, D.C., Hernandez, M., Guertin, D., Metz, L., Wei, H., Ponce-Campos, G., Burns, I., McCord, S.E., Nearing, M., Williams, C.J., Houdeshell, C., Rahman, M., Meles, M.B., Barker, S. 2024. An artificial neural network to estimate the foliar and ground cover input variables of the Rangeland Hydrology and Erosion Model. Journal of Hydrology. 631. Article 130835. https://doi.org/10.1016/j.jhydrol.2024.130835.
DOI: https://doi.org/10.1016/j.jhydrol.2024.130835

Interpretive Summary: Soil erosion affects landscapes worldwide, threatening food security and ecosystem viability. The evaluation of the risks from erosion in dry areas is challenging because erosion is often an outcome of individual rainstorms and greatly depends on the specific rainfall patterns, local land-use and land cover, soils, and topography. ARS has developed the RHEM (Rangeland Hydrology and Erosion Model) to simulate the risk of erosion on rangelands. However, RHEM requires input parameters that are sometimes difficult or expensive to measure. Specifically, RHEM requires information about vegetation leaf and ground cover fractions that generally must be measured in-situ, which makes it difficult to use models like RHEM to produce erosion or soil risk maps for areas exceeding the size of a hillslope such as a large watershed. We previously developed a Machine Learning (ML) emulator of RHEM that has low computational expense and can, in principle, be run over large areas (e.g., over the continental United States). In this effort, ARS scientists from Tucson, Arizona, Las Cruces, New Mexico, and Davis, California in cooperation with investigators from Google Research, the University of California-Davis, the University of Arizona, and the Natural Resources Conservation Service developed a deep learning model to estimate the RHEM ground cover inputs from remote sensing, and climate time series, and nationally available soils data. This reduces the need for extensive field surveys to produce erosion maps. The emulator predicted runoff well and soil loss and sediment fairly well at 66,643 field locations within the United States. We demonstrate how this approach can be used for mapping by developing runoff, soil loss, and sediment yield maps over 1356 km2 region of interest in Nebraska.

Technical Abstract: Models like the Rangeland Hydrology and Erosion Model (RHEM) are useful for estimating soil erosion, however, they rely on input parameters that are sometimes difficult or expensive to measure. Specifically, RHEM requires information about foliar and ground cover fractions that generally must be measured in situ, which makes it difficult to use models like RHEM to pro-duce erosion or soil risk maps for areas exceeding the size of a hillslope such as a large watershed. We previously developed a deep learning emulator of RHEM that has low computational expense and can, in principle, be run over large areas (e.g., over the continental US). In this paper, we develop a deep learning model to estimate the RHEM ground cover inputs from remote sensing time series, reducing the need for extensive field surveys to produce erosion maps. We achieve a prediction accuracy on hillslope runoff of R2 ˜ 0.9, and on soil loss and sediment yield of R2 ˜ 0.4 at 66,643 field locations within the US. We demonstrate how this approach can be used for mapping by developing runoff, soil loss, and sediment yield maps over 1356 km2 region of interest in Nebraska.