Assessing the performance of multivariate data analysis for predicting solar radiation using alternative meteorological variables

Authors : Mustafa Özbuldu, Yunus Emre Şekerli

Pages : 45-52

Doi:10.51753/flsrt.1590684

View : 23 | Download : 30

Publication Date : 2025-04-30

Article Type : Research Paper

Abstract :This research analyses how well the Partial Least Squares Regression models could predict the monthly average daily global solar radiation for seven stations in the Mediterranean region of Türkiye. Five model scenarios were created with the SARAH-3 satellite dataset from 2005 to 2023 and using ERA5-AG meteorological variables. These included maximum and minimum temperature configurations, dew point temperature, precipitation, wind speed, and vapor pressure. Different models were examined for their prediction success by using different criteria and assessing the models with varying performance evaluation benchmarks. Based on the results, the models were accurate, mainly when all the predictor variables were used. The highest predictive performance was observed at Burdur station with KGE=0.937, NSE=0.901, and RSR=0.322. The greater regional variations showcased the specific meteorological parameters’ relevancy. The results also support the adequacy of the ERA5-AG dataset for climate modelling and resource evaluation purposes. Unlike traditional regression approaches, this study demonstrates the efficiency of PLSR in handling high-dimensional climatic datasets for solar radiation prediction. These findings support the reanalysis of data in renewable energy and agricultural applications, particularly in data-limited regions.
Keywords : ERA5-AG, Mediterranean region, partial least squares regression, solar radiation prediction