Riparian Ecosystem Management Model: Sensitivity to Soil,Vegetation, and Weather Input Parameters1

Abstract: The Riparian Ecosystem Management Model (REMM) was developed by the U.S. Department of Agriculture‐Agriculture Research Service (USDA‐ARS) and its cooperators to design and evaluate the efficiency of riparian buffer ecosystems for nonpoint source pollution reduction. REMM requires numerous inputs to simulate water movement, sediment transport, and nutrient cycling in the buffer system. In order to identify critical model inputs and their uncertainties, a univariate sensitivity analysis was conducted for nine REMM output variables. The magnitude of each input parameter was changed from ?50% to +50% from the baseline data in 12 intervals or, in some cases, the complete range of an input was tested. Baseline model inputs for the sensitivity analysis were taken from Gibbs Farm, Georgia, where REMM was tested using a 5‐year field dataset. Results of the sensitivity analysis indicate that REMM responses were most sensitive to weather inputs, with minimum daily temperature having the greatest impact on the nitrogen‐related outputs. For example, the 100% change (?50% to +50%) in minimum daily temperature input values yielded a 164.4% change in total nitrogen (N), a 109.3% change in total nitrate (NO₃), and a 127.1% change in denitrification. REMM was most sensitive to precipitation with regard to total flow leaving the riparian vegetative buffer zone (199.8%) and sediment yield (138.2%). Deep seepage (12.2%), volumetric water content (24.8%), and pore size index (6.5%) in the buffer soil profile were the most influential inputs for the output water movement. Sediment yield was most sensitive to Manning’s coefficient (46.6%), bare soil percent (40.7%), and soil permeability (6.1%). For vegetation, specific leaf area, growing degree day coefficients, and maximum root depth influenced the nitrogen related outputs. Overall results suggest that because of the high sensitivity to weather parameters, on‐site weather data is needed for model calibration and validation. The model’s relatively low sensitivity to vegetation parameters also appears to support the use of regional vegetation datasets that would simplify model implementation without compromising results.