Simulating landscape catena effects in no-till dryland agroecosystems using GPFARM

This study evaluated the site-specific applicability and efficacy of the GPFARM decision support system (DSS) based on underlying simulation model performance for dry mass grain yield, crop residue, total soil profile water content, and total soil profile residual NO3-N across a landscape catena for dryland no-till experimental locations in eastern Colorado. Relative error of simulated mean, normalized objective function (root mean square error divided by the observed mean), and index of agreement evaluation statistics were calculated to compare modeled results to observed data. A one-way, fixed-effect ANOVA was also performed to determine differences among experimental locations and summit, sideslope, and toeslope landscape positions. GPFARM simulations matched observed data trends, with the model correctly distinguishing variations between the summit and toeslope landscape positions. In addition, experimental observations and GPFARM simulations both indicated that the toeslope landscape position was the most productive for grain yield and also exhibited higher amounts of crop residue, total soil profile water content, and total soil residual NO3-N. The GPFARM crop model performed adequately but was inconsistent in simulating winter wheat, corn, and sorghum dry mass grain yield. GPFARM performance in simulating crop residue was poorer than for crop grain yield. GPFARM predicted mean total soil profile water content was generally within +/- 20% of the observed mean across locations and landscape positions, with the model somewhat biased towards overpredicting total soil profile water content at the summit and sideslope landscape positions. Total soil profile residual NO3-N was underpredicted by GPFARM across all locations and landscape positions by an average of 30%. Although GPFARM appears to have reasonably simulated long-term output responses across a landscape catena for the eastern Colorado experimental locations (especially given the simplifying assumptions in many of the GPFARM simulation components and the inherent variability present at the experimental plot level), different interpretations of GPFARM performance can be made depending on the evaluation statistic of interest. Furthermore, the model cannot fully account for water and chemical movement across the landscape catena: simulation results suggest that addition of a spatially-distributed routing component should offer improvements in GPFARM prediction accuracy across a catena where surface runoff or lateral subsurface flow is occurring. Published by Elsevier Ltd.