From grid to field: Assessing quality of gridded weather data for agricultural applications

High quality measured weather data (MWD) are not available in many agricultural regions across the globe. As a result, many studies that dealt with global climate change, land use, and food security scenarios and emerging agricultural decision support tools have relied on gridded weather data (GWD) to estimate crop phenology and crop yields. An issue is the agreement of GWD with MWD and the degree to which this agreement may influence the utility of GWD for agricultural research. The objectives of this study were: (i) to compare the agreement of two widely used gridded weather databases (GWDs) (Daymet and PRISM) and MWD, (ii) to evaluate their robustness at simulating maize growth and development, and (iii) to examine how GWD compare relative to weather data interpolated from existing meteorological stations for which MWD are available. The U.S. Corn Belt, a region that accounts for 43 and 34% of respective global maize and soybean production, was used as a case of study because of its dense weather station network and high-quality MWD. Historical daily MWD were retrieved from 45 locations across the region, resulting in ca. 1300 site-years. To test the accuracy of GWDs, separate simulations of maize yield and development were performed, separately for the two GWDs and MWD, using a well-validated maize crop model. For both GWDs, small biases were observed for temperature and growing degree-days in relation with MWD. However, accuracy was much lower for relative humidity, precipitation, reference evapotranspiration, and degree of seasonal water deficit. There was close agreement in duration of vegetative and reproductive phases between GWD and MWD, with root mean square error (RMSE) ranging from 3 to 7 days for the different crop phases and GWDs. However, robustness of GWDs to reproduce maize yields simulated using MWD was lower as indicated by the RMSE (18 and 24% of average yield for Daymet and PRISM, respectively). There was also a high proportion of site-years (20 and 32% for Daymet and PRISM, respectively) exhibiting a yield deviation >15% in relation to the yield simulated using MWD. Data interpolation using a dense weather station network resulted in lower RMSE% for simulated phenology and yields relative to GWDs. Findings from this study indicate that GWD cannot replace MWD as a basis for field-scale agricultural applications. While GWD appear to be robust for applications that only require temperature for prediction of crop stages, GWD should not be used for applications that depend on accurate estimation of crop water balance, crop growth, and yield. We propose that the evaluation performed in this study should be taken as a routinary activity for any research or agricultural decision tool that relies on GWD. Published by Elsevier B.V. This is an open access article under the CC BY license.