Assessing the potential of random forest method for estimating solar radiation using air pollution index

Abstract

Simulations of solar radiation have become increasingly common in recent years because of the rapid global development and deployment of solar energy technologies. The effect of air pollution on solar radiation is well known. However, few studies have attempting to evaluate the potential of the air pollution index in estimating solar radiation. In this study, meteorological data, solar radiation, and air pollution index data from three sites having different air pollution index conditions are used to develop random forest models. We propose different random forest models with and without considering air pollution index data, and then compare their respective performance with that of empirical methodologies. In addition, a variable importance approach based on random forest is applied in order to assess input variables. The results show that the performance of random forest models with air pollution index data is better than that of the empirical methodologies, generating 9.1-17.0% lower values of root-mean-square error in a fitted period and 2.0-17.4% lower values of root-mean-square error in a predicted period. Both the comparative results of different random forest models and variance importance indicate that applying air pollution index data is improves estimation of solar radiation. Also, although the air pollution index values varied largely from season to season, the random forest models appear more robust performances in different seasons than different models. The findings can act as a guide in selecting used variables to estimate daily solar radiation and improve the accuracy of solar radiation estimation.

Daily Global Solar Radiation in China Estimated From High‐Density Meteorological Observations: A Random Forest Model Framework

“…In contrast, WIN and SAT‐mean have the lowest variable importance, with importance values of 1.17% and 1.00%, respectively. In addition to meteorological variables, the DOY (seasonal effects) and latitude (geographical factor) are critical to estimate DGSR, consistent with previous results (Li et al, ; Li et al, ; Sun et al, ).…”

“…According to our results from other models (DT, BP, SVM, and MLR), RF is more suitable for DGSR predictions at a large scale with good performance and shows the importance of all input variables for estimating DGSR at the national scale, which is consistent with previous work (Sun et al, ). Specifically, 10 meteorological variables, as well as the latitudes, longitudes and altitudes of sites, and dummy variables, are evaluated in our work as predictors to employ in RF models, indicating that daily sunshine duration is the most important contributing factor in estimating DGSR, and daily maximum land surface temperature and DOY also play crucial roles in determining DGSR across China.…”

“…This is followed by another four dominant variables-namely, LST-max, DOY, latitude, and LST-mean, with importance values of 17.75%, 8.16%, 5.04%, and 5.02%, respectively. The high importance of SSD to daily solar radiation has also been identified in previous modeling studies Sun et al, 2015Sun et al, , 2016. Note that SSD can also imply and contain the impacts of air pollution on DGSR (Wang et al, 2012(Wang et al, , 2013.…”

“…On the one hand, this has significantly modulated the change in surface solar radiation (Che et al, 2005;Guo et al, 2018;Wang et al, 2012Wang et al, , 2013Wang & Wild, 2016;Zheng et al, 2018;He & Wang, 2020;Yang et al, 2020), while on the other hand, it has led to solar radiation becoming one of the fastest-growing and important sources of clean and renewable energy. Therefore, solar radiation is a topic that has attracted broad and increasing attention in China (Che et al, 2005;Sun et al, 2016;Li et al, 2017;Song et al, 2019;Tang et al, 2016Tang et al, , 2018Liu et al, 2019;He & Wang et al, © 2020 The Authors. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.…”

“…To obtain a longer and more detailed data set of DGSR in China, previous studies have made great efforts to estimate DGSR from surface meteorological observations using various methods, such as the traditional empirical formula and artificial neural network methods (Jiang et al, 2009a;Huang et al, 2011;Qin et al, 2011;Chen et al, 2013;Li et al, 2013;Tang et al, 2010Tang et al, , 2013Tang et al, , 2016Sun et al, 2016;Li et al, 2017). Using these methods, however, the importance of input variables for estimating DGSR could not be identified objectively and automatically.…”