Evaluation of Solar Irradiance Models with a Special Referenceto Globally-Parameterized and Land Cover-Sensitive Solar123

Abstract

Numerical models are often the only means available to generate solar irradiance (SR) information, for historical or future SR specifications as well as due to inadequacies of contemporary SR measurements. This paper evaluates five such models that have been proposed as generic and applied as such. Special emphasis is given to Solar123, an integrative model rooted in Lambert’s Cosine Law and Bouguer’s Law with globally-parameterized atmospheric property functions and with input limited to precipitation, air temperature, geographic location, topography and rudimentary land cover information. The selected SR models in general perform well in reproducing the SR data for the USA, with a root mean square error-to-data mean ratio (RMSE/SRmean ratio) of 9.8–11.4%. A possible exception is the Bristow-Campbell logic as implemented by the Vegetation/Ecosystem Modeling and Analysis Project. Beyond the USA, Solar123 yields an RMSE/SRmean ratio of 8–17% by region (196 stations in total), generally outperforming the other SR models. Compared to time-sequential monthly SR data, projections by Solar123 have an RMSE/SRmean ratio of 8.6–14.1% for six weather stations representative of major climate regimes in Canada, or an RMSE/SRmean ratio of 13–24% for three forest sites in the USA, Germany and Japan. Solar123 projections also compare favorably against the output from the General Circulation Models in terms of ratio change in SR with the doubling of the atmospheric carbon-dioxide concentration: the two fall within ±10% of each other for 85% of a total 264 cases, and within ±20% for all but 3 of the cases. The above statistics suggest that Solar123 represents an improvement over other SR models not only in configuration but also in projection accuracy, and that Solar123 is useful for projecting spatial variation in SR across weather stations around the world and over different land covers, and for projecting temporal variation in SR under the present climate regimes and likely for regimes changed beyond the present fluctuation range. The work further calls into question the common practice of applying SR data irrespective of local land cover.