Akademik Veri Yönetim Sistemi
7th International Azerbaijan Congress on Life, Engineering, Mathematical, And Applied Sciences, Baku, Azerbaycan, 25 Haziran - 27 Temmuz 2024, cilt.1, sa.1, ss.106-107
In recent years, significant changes in rainfall patterns have been documented to cause socioeconomic and ecological challenges, especially in regions with high vulnerability. Therefore, it becomes imperative to predict future precipitation levels at both short- and long-term scales through the analysis of comprehensive time series data collected from observation stations. In this study, monthly precipitation data from the meteorological station within the borders of Kutahya province between 1960 and 2023 were utilized. Precipitation data from 1960 to 2007 were used as training data, and data from 2008 to 2023 were used as test data. Precipitation at exit time t was estimated using precipitation values at lag times t˗1, t˗2, and t˗3 as input variables. The aim of this study is to compare the performance of pre-processing techniques to develop hybrid models in precipitation forecasting. Gaussian process regression (GPR) algorithm was utilized as machine learning method. Tunable Q-Factor Wavelet Transform (TQWT), Variational mode decomposition (VMD), and Empirical wavelet transform (EWT) were utilized as a pre-processing technique to develop hybrid models. Relative Root Mean Squared Error (RRMSE), Performance Index (PI), Overall Index of Model Performance (OI), and Wilmott’s Refined Index (WI) were used to evaluate model performance. According to performance metrics, the hybrid TQWT-GPR model yielded the best results (RRMSE= 4.015, PI= 0.0201, OI= 0.9932, WI=0.9741). Considering all performance measurements, it was concluded that the parsing success of TQWT exceeded both VMD and EWT. According to the statistical significance level of p > 0.05, the Kruskal-Wallis test at the conclusion of the study established that the measured and estimated data originated from identical distributions. Consequently, the efficacy of the recommended methods for comparisons was validated.
Keywords: Precipitation, GPR, pre-processing technique, Türkiye