A Regression Metamodel To Replace SWAT In Crop Yield Prediction For Big Creek Watershed

• Main Author: Johnson, Andrew
• Format: Others
• Published: OpenSIUC 2013
• Online Access: https://opensiuc.lib.siu.edu/theses/1238; https://opensiuc.lib.siu.edu/cgi/viewcontent.cgi?article=2249&context=theses

Climate change will have a significant impact on the productivity of agricultural lands and ecosystem services in the coming decades. Variability in temperature and precipitation will alter many central U.S. watersheds. Simulation models such as the Soil and Water Assessment Tool (SWAT) offer the ability to model changes in watersheds by varying inputs. Unfortunately, SWAT requires a large number of input parameters and computation time to process the output data. Regression metamodels offer an alternative that seeks to replace the simulation model with a regression equation. This research created a linear regression metamodel to approximate SWAT in crop yield prediction. Results show that regression models can account for 45-84 percent of variance in yields for corn, soybean, alfalfa, switchgrass, and cotton in Big Creek Watershed. The coefficient of variation for each of these models ranged from 13 to 41 percent. These metamodels were able to reduce simulation time from hours to minutes. The tradeoff for utilizing metamodels is computation time versus accuracy. The results of this research indicate that the considerable reduction in computation time coupled with a moderate degree of accuracy in predicting crop yields necessitates the use of metamodels over SWAT. Regression coefficients for each metamodel can reveal how various weather and farm management techniques impact crop yields. These metamodels will be utilized by the Agent Based Model to determine how farmers will respond to future economic policies and crop prices based on a series of climate scenarios.