2026 CSDMS meeting-032
From CSDMS
(if you haven't already)
Log in (or create account for non-CSDMS members)
Forgot username? Search or email:CSDMSweb@colorado.edu
Browse abstracts
How Well Do Erosion Models Know Iowa? Comparing Semi-Empirical, Physically-Based, and U-Net Deep Learning Approaches Across Physiographic Watersheds
Derrick Platero,
Iowa State University Ames Iowa, United States. derrickp82@gmail.com
Intensive agriculture has driven hillslope erosion rates (5–50 Mg/ha/yr) that exceed soil production thresholds across the low-relief, deglaciated landscapes of the U.S. Midwest, threatening agricultural sustainability and amplifying flood risk. Despite advances in GIS and geomorphic modeling, no validated, large-extent analysis of hillslope topographic change at the decadal scale currently exists, largely because high-precision, repeat elevation data were unavailable until recently. Iowa's statewide LiDAR collections in 2009 and 2020 now provide a unique opportunity to test how well erosion-deposition models predict real landscape change.
This study evaluates three erosion modeling frameworks: GeoWEPP (physically-based, semi-distributed), WaTEM/SEDEM (semi-empirical, distributed), and Landlab (numerical, process-based), across ten HUC-12 watersheds spanning Iowa's major physiographic regions. Model outputs are compared against LiDAR-derived DEMs of Difference (DoD) representing a decade of observed topographic change (2010–2020). Additionally, a U-Net deep learning regression model trained on geospatial terrain derivatives is evaluated as an alternative predictive approach.
Preliminary results indicate that WaTEM/SEDEM systematically underestimates the magnitude of erosion and deposition relative to the DoD, while GeoWEPP and Landlab produce outputs more consistent with observed landscape change. The U-Net model also shows promising predictive performance. These findings represent the first regionally validated comparison of multi-framework erosion models against decade-scale observed topographic change in transport-limited agricultural landscapes, with direct implications for forecasting the consequences of erosion on flooding, soil carbon, and food security under changing climate and land-use conditions.
