Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
81579 | Agricultural and Forest Meteorology | 2015 | 14 Pages |
•Seven existing net radiation estimation empirical models were evaluated using global measurements.•New developed linear model worked better than others in the paper.•Factors impact net radiation estimation were discussed.
All-wave net surface radiation is greatly needed in various scientific research and applications. Satellite data have been used to estimate incident shortwave radiation, but hardly to estimate all-wave net radiation due to the inference of clouds on longwave radiation. A practical solution is to estimate all-wave net radiation empirically from shortwave radiation and other ancillary information. Since existing models were developed using a limited number of ground observations, a comprehensive evaluation of these models using a global network of representative measurements is urgently required. In this study, we developed a new day-time net radiation estimation model and evaluated it against seven commonly used existing models using radiation measurements obtained from 326 sites around the world from 1991 to 2010. MERRA re-analysis products from which the meteorological data were derived and remotely sensed products during the same period were also used. Model evaluations were performed in both global mode (all data were used to fit the models) and conditional mode (the data were divided into four subsets based on the surface albedo and vegetation index, and the models were fitted separately). Besides, the factors (i.e., albedo, air temperature, and NDVI) that may impact the estimation of all-wave net radiation were also extensively explored. Based on these evaluations, the fitting RMSE of the new developed model was approximately 40.0 Wm−2 in the global mode and varied between 18.2 and 54.0 Wm−2 in the conditional mode. We found that it is better to use net shortwave radiation (including surface albedo) than the incident shortwave radiation nearly in all models. Overall, the new model performed better than other existing linear models.