Simulating County-Level Crop Yields in the Conterminous United States using the Community Land Model: The effects of optimizing irrigation and fertilization
In this study, we applied version 4.5 of the Community Land Model (CLM) at a 0.125° resolution to provide the first county-scale model validation for simulating crop yields over the Conterminous United States (CONUS). Large bias was found in simulating crop yields against U.S. Department of Agriculture (USDA) survey data, with county-level root-mean-square error (RMSE) of 42% and 38% for simulated US mean corn and soybean yields, respectively. We then synthesized crop yield, irrigation and fertilization data sets from USDA and U.S. Geological Survey (USGS) to constrain model simulations. Compared with fertilization, irrigation has limited effects on crop yields with improvements limited to irrigated regions. In most current-generation Earth system models (ESMs), fertilizers are applied spatially uniformly with fixed amounts and timing without considering crop fertilizer demand. To address this weakness, we propose a prognostic fertilization scheme that dynamically determines the timing and rate of each fertilizer application, with the annual amounts and valid fertilization time windows constrained by surveyed data. The optimized fertilization scheme reduces RMSE to 22% and 21% of the US mean corn and soybean yields, respectively. Compared with the default CLM4.5, our fertilization scheme substantially improves crop yield simulations especially over major crop growing regions. However, to compensate for the widely documented biases in denitrification rates simulated by CLM4.5, the dynamically determined fertilization timing and rates do not match the fertilization practices of farmers exactly. Therefore, caution should be exercised when extending this method beyond the contemporary conditions.