Generating Long-Term Location-Specific Weather Data Where None Exist for Yield Gap Assessment.

The models commonly used to simulate potential and water-limited crop yields require long-term (15+ years) observed daily weather data, including incident solar radiation (SR), maximum and minimum temperature (Tmax & Tmin), precipitation, and some measure of humidity. Observed daily weather data are unavailable in many cropping regions but gridded global weather databases with complete terrestrial coverage are available. These are typically derived from global circulation computer models, interpolated weather station data, or remotely sensed surface data from satellites. Studies that have compared simulated yields using gridded versus observed weather data from a location within the same grid have found poor agreement. The present paper evaluates a new method to ‘propagate’ location-specific short-term weather data (3-5 years of Tmax and Tmin) to derive long-term weather data based on (i) calibrated NASA Tmax and Tmin based on correlations between NASA and observed data for the few years of available data and (ii) solar radiation, Tdew, and precipitation, from NASA or TRMM. We evaluated the propagation method using long-term observed weather data from 18 sites located in North and South America, Europe, Africa and Asia. For each location, NASA Tmax and Tmin were calibrated based on all possible subsets of 3, 4, 5 and 10 consecutive years of observed Tmax and Tmin, which resulted into multiple files of propagated weather data and multiple simulated average yields. The distributions of average simulated yields based on propagated weather data were compared against the average simulated yield based on long-term observed weather data. We also evaluated other sources of weather data: (i) non-calibrated NASA weather data and (ii) MarkSim generated weather data. Results indicated that distributions of simulated yield based on propagated weather were consistently within ± 10% of the simulated yield based on observed long-term weather data, with much better agreement than average simulated yields based on non-calibrated NASA and MarkSim weather data. Agreement between simulated yields based on propagated and observed weather data increases slightly with the number of observed years used for the calibration of the propagated weather. We conclude that, wherever 3+ years of data on Tmax & Tmin are available, the propagation technique is a viable and superior alternative to generate long-term weather data for crop simulation compared with gridded or generated weather data.