Akademik Veri Yönetim Sistemi
JOURNAL OF WATER AND CLIMATE CHANGE, 2025 (SCI-Expanded, Scopus)
This study investigates three approaches for modeling intensity-duration-frequency (IDF) relationships of precipitation in the G & ouml;ksu River Basin. The first approach applies statistical frequency analysis at eight stations for 14 precipitation durations using five probability distributions (Gumbel, Log-Normal, GEV, Pearson III, and Log-Pearson III) with three parameter estimation methods (PWM, MLE, MOM). The best distributions were selected via Kolmogorov-Smirnov and Chi-Square tests. The second approach employs machine learning (ML) models, including standalone and hybrid models enhanced with the Discrete Cosine Transform (DCT), trained on data from six stations. The third method uses 11 empirical IDF equations optimized through the Honey Formation Optimization (HFO) algorithm. The study's novelty lies in combining DCT-enhanced ML and HFO for IDF prediction, specifically applied to the G & ouml;ksu Basin. Model performance at the Mut and Silifke stations was evaluated using MAE, RMSE, R-2, Performance Index, Pearson Correlation, and Mean of Min/Max error. Results revealed that the DCT-based Least Squares Boosting (LSBoost-DCT) model achieved the best accuracy, with MAE (15.83, 30.49) and RMSE (27.47, 48.19) values, reducing errors by up to 42% and 37%. Overall, integrating DCT with ML substantially improved IDF prediction accuracy, providing a robust framework for future hydrological research.