Evaluating the Reliability of Reanalysis as a Substitute for Observational Data in Large-scale Agricultural Assessments

Future projections of food security require historical agricultural assessments to validate, improve, and understand the limitations of yield estimates. Poor observational climate networks often force historical assessments to rely on reanalysis data- climate model output nudged by observations- for inputs to crop models. However, agricultural yields are sensitive to changes in precipitation, and since reanalysis products generally use little or no observational precipitation in the data assimilation process, its use may compromise the validation exercise. Previous studies do not systematically assess whether reanalysis data is sufficient or data measurements are required. We test the reliability of reanalysis data for agricultural analyses with simulations of maize yields in the U.S., where observational data are extensive. We drive the widely used Decision Support System for Agrotechnology Transfer (DSSAT) crop model with climate inputs from a combination of data sources: bias- and unbias-corrected reanalyses, and observation-based precipitation and solar radiation. We find that driving DSSAT with reanalysis precipitation produces unreliable yield estimates, but driving it with reanalysis bias-corrected with monthly observations is more robust. Bias corrections do require observational data, but gathering reliable monthly data may be easier than gathering daily data. The approach is therefore promising for data-poor regions where observational precipitation is less available and existing data is unreliable. The priority for climate monitoring networks may not be in daily records but instead in lower-cost observational systems that estimate data over coarser temporal resolutions.