Comparative Parameter Estimation in Weather Modeling Using Kink and OLS Regression Methods

Abstract

Temperature fluctuations are a key indicator of changing weather patterns and have a direct impact on environmental planning and public awareness. This study utilizes historical weather data collected from Sulaymani, Iraq, covering the period from 2005 to 2024, to analyze how humidity, vapor, and wind direction influence temperature trends. Additionally, the analysis is based on a comparison between the traditional multiple linear regression model and an enhanced kink regression model, designed to account for a structural shift in the effect of humidity at a specific threshold. The study initially applies OLS-based multiple regression to model temperature as the dependent variable. However, due to observable changes in the relationship between humidity and temperature, a kink model is introduced to capture this shift more effectively. Moreover, this kink approach, which utilizes piecewise linear regression, revealed improved model performance. Specifically, the mean square error (MSE) decreased, and both the R² and adjusted R² increased, indicating a better fit and more reliable predictions. These findings validate the use of threshold-based models in environmental studies and highlight the added value of kink regression in uncovering hidden structural relationships within climatic data.