Estimation Of Diffuse Solar Radiation In China Using Different Methods

Solar radiation is the main energy source for the operation and development of the earth's surface system,which strongly affects regional climate change and surface processes,patterns and functions.The diffuse radiation(Hd),as an important component of global solar radiation reaching the surface,is a key factor affecting the terrestrial ecosystem productivity and global carbon budget.It is of great significance to regional environmental change,ecological processe and atmospheric pollution.Due to the sparse observation stations,high cost of instruments,maintenance cost and technical requirements,and lack of historical data,there is still a lack of high-precision and high-quality diffuse radiation data in China.There is also a lack of comprehensive comparative study of various Hd estimation models using synchronous observation data and unified evaluation criteria.Therefore,this study intends to use the observation data of 17 meteorological stations in different climatic regions of China to establish and evaluate various types of Hd models,in order to obtain the best models for each climatic region,and to provide some technical supports for the development,utilization,planning and management of solar radiation resources in China.The main research contents in this study are as follows:(1)15 sunshine based Hd models were established using meteorological observation data,including 4 mixed variable models established from daily mean temperature,daily maximum temperature,daily minimum temperature,relative humidity and clear sky index.The accuracies of above models was verified and compared,and the applicability of the models in Qinghai-Tibet Plateau,cold-temperate zone,middle-temperate zone,warm-temperate zone,subtropical zone and tropical zone was evaluated.The results showed that the ranges of mean deviation error(MBE),mean absolute error(MAE),mean absolute relative error(MARE),root mean square error(RMSE),relative root mean square error(RRMSE),t-stat,standard deviation(STD)and correlation coefficient(R)for above models were in the ranges of-0.699~0.278MJm-2day-1,1.331~2.508MJm-2day-1,0.208~0.397MJm-2day-1,1.807~6.779MJm-2day-1,24.889%~102.716%,5.511%~14.577%,0.941~1.773 MJm-2day-1 and 0.519~0.792,respectively.The overall accuracy of mixed variable model No.15 is better than that of other 14 models,the averaged MBE,MAE,MARE,RMSE,RRMSE,t-stat,STD and R for all stations were-0.43MJm-2day-1,1.331 MJm-2day-1,0.208MJm-2day-1,1.807MJm-2day-1,24.889%,10.866%,0.941 MJm-2day-1and 0.792,respectively.Comparing the accuracy of Hd models in different climatic regions,it was found that the average accuracy of each model is the highest in subtropical regions and the worst in arid regions.The models No.12 and No.13 performed better in tropical regions;No.15,No.13,No.12 and No.14 performed better in subtropical,temperate,middle temperate and cold temperate regions;No.15 and No.13 performed better in Qinghai-Tibet Plateau than other models.(2)97 multivariate Hd models were established by using meteorological and geographic parameters such as air temperature,relative humidity,sunshine duration,clear sky index and atmospheric mass.According to the number of input variables,these models were divided into five categories,which contained 44,35,5,5 and 8 models,respectively.The accuracy of each model in different climatic regions was analyzed in detail,and the performance of different models at different stations was also compared.The results showed most multivariate models performed well at Zhengzhou,Wuhan,Shanghai,Harbin,Guangzhou,Chengdu and Beijing stations,however,the models did not performed good at Urumqi,Ejina Banner and Sanya stations.No.55,No.59,No.66,No.81 and No.89 were generally performed better than other models;the fourth kind of model with four input parameters generally had higher accuracies.The accuracies of Kd-based and KD-based models were compared and analyzed,it was found that the Kd-based models(No.1-25,45-64,80-82 and 85-97)were generally better than that of KD based models(No.26-44,65-79,83 and 84).Comparing the accuracy of the multivariate model with that of the existing model in literature at Ejina Banner,Lanzhou,Kashi,Urumqi,Golmud,Wuhan,Beijing and Kunming station,it was found that the estimation errors of the No.40,55,39,18 and 54 were lower than those of the existing models,and the Model 55 had the highest accuracy.(3)6 artificial intelligence based Hd models using meteorological observations were developed from Back Propagation Neural Network Model(BP),Adaptive network fuzzy inference system(ANFIS),generalized regression neural network model(GRNN),M5 model tree(M5_Tree),multiple adaptive regression algorithm model(MARS)and particle swarm optimization based limit learning machine(POS_ELM).The optimal combination of input variables for each Hd models were obtained by parameter selection and hyper-parameter optimization.Through comparison and verification,it was found that the accuracies of PSO_ELM and M5 Tree models were higher than other models on the whole;the accuracies of 6 kinds of artificial intelligence models at Beijing and Shenyang stations were obviously better than those of other stations,and the accuracies in tropical and plateau areas were relatively poor.By comparing the accuracy of AI model with that of multivariate model,it was found that the average accuracy of AI model was higher than that of multivariate model in many stations,but in Chengdu,Guangzhou,Shanghai and Wuhan stations,the accuracy of multivariate model was higher than that of AI model.As far as the model accuracy was concerned,M5 Tree had the highest accuracy in Beijing,Ejina Banner,Golmud,Harbin,Kashgar,Kunming,Lhasa,Mohe,Shenyang and Urumqi.