Uncertainties in Driving Agricultural Models with Downscaled and Bias-Corrected Climate Information

Abstract:

The agricultural sector represents an important coupled human-environment system that is sensitive to extreme weather and climate events. Efforts to understand the challenges to agriculture posed by a changing climate often rely on downscaled and bias-corrected climate information. As such, it is important to quantify the uncertainty introduced by the use of these products. Many previous studies have performed uncertainty analysis focused on climatic variables, but fewer have investigated how these uncertainties propagate into human systems by including an econometric model. In this work, we use a multi-model ensemble of downscaled and bias-corrected climate model outputs as well as the corresponding parent model outputs to investigate uncertainty in corn yield projections in the United States. By using both ensembles to drive a pre-validated agricultural model, we evaluate in a probabilistic manner how well each ensemble can reproduce historically observed yields. We find that for many important corn-growing counties in the United States, the downscaled and bias-corrected models systematically underestimate the largest historically observed yield shocks. We project yields through the year 2100 and find that these biases, which arise from the downscaling and bias-correction process, are only amplified in future projections. Our results highlight important limitations and trade-offs regarding the use of downscaled and bias-corrected climate information, and in particular their ability to properly sample extreme events when simulating coupled human-environment systems.