Creating long-term weather data from thin air for crop simulation modeling

Simulating crop yield and yield variability requires long-term, high-quality daily weather data, includingsolar radiation, maximum (Tmax) and minimum temperature (Tmin), and precipitation. In many regions,however, daily weather data of sufficient quality and duration are not available. To overcome this limitation,we evaluated a new method to create long-term weather series based on a few years of observed dailytemperature data (hereafter called propagated data). The propagated data are comprised of uncorrectedgridded solar radiation from the Prediction of Worldwide Energy Resource dataset from the NationalAeronautics and Space Administration (NASA?POWER), rainfall from the Tropical Rainfall MeasuringMission (TRMM) dataset, and location-specific calibration of NASA?POWER Tmax and Tmin using a limitedamount of observed daily temperature data. The distributions of simulated yields of maize, rice, or wheatwith propagated data were compared with simulated yields using observed weather data at 18 sites inNorth and South America, Europe, Africa, and Asia. Other sources of weather data typically used in cropmodeling for locations without long-term observed weather data were also included in the comparison:(i) uncorrected NASA?POWER weather data and (ii) generated weather data using the MarkSim weathergenerator. Results indicated good agreement between yields simulated with propagated weather dataand yields simulated using observed weather data. For example, the distribution of simulated yieldsusing propagated data was within 10% of the simulated yields using observed data at 78% of locationsand degree of yield stability (quantified by coefficient of variation) was very similar at 89% of locations. Incontrast, simulated yields based entirely on uncorrected NASA?POWER data or generated weather datausing MarkSim were within 10% of yields simulated using observed data in only 44 and 33% of cases,respectively, and the bias was not consistent across locations and crops. We conclude that, for most locations,3 years of observed daily Tmax and Tmin data would allow creation of a robust weather data set forsimulation of long-term mean yield and yield stability of major cereal crops