CR 36:219-230 (2008)  -  DOI: https://doi.org/10.3354/cr00750

ABSTRACT: Lack of data on site-specific daily solar radiation (R_g) is a significant impediment for most crop modeling applications. For this reason, 5 methods for estimating R_g were tested: the Ångström-Prescott equation (AP), the Supit - van Kappel formula (SK) and 3 temperature-based methods: Campbell-Donatelli model, Hargreaves equation, and piecewise multiple linear regressions with a breakpoint (PLR). To overcome the lack of long and continuous time series of R_g measurements at multiple sites, satellite-derived R_g from the HelioClim-1 database were tested against observations from 2 stations, and then interpolated for 12 additional stations. When sunshine duration data were available, the AP equation was best, because it (1) produced intercepts and slopes closest to zero and unity, respectively and (2) had the lowest relative RMSE (9 to 18.6%). When cloud cover observations and data on maximum and minimum temperatures were available, the SK equation was equally effective as AP in most assessment metrics. When the only data for a site were daily maximum and minimum temperatures, the PLR approach with a breakpoint, which reflects the value at which the response of daily R_g changes as a function of the extraterrestrial solar radiation R_A and the diurnal temperature range, performed best. The mean relative RMSE of the PLR approach was <3.7% higher than that of AP. The SK equation provided the most suitable simulation of measured R_g for the CERES-Wheat crop model, while among the temperature-based methods PLR produced the smallest yield errors. Future validation efforts should explore the validity of the PLR model in other regions and under regimes with greater availability of R_g data

KEY WORDS: Solar radiation estimation · Crop simulation models · Piecewise multiple linear regression · Greece