This paper presents a multi-stage machine learning framework supported by explainable artificial intelligence (XAI) techniques to predict temperature trends using meteorological data from Zonguldak, Turkey. This hybrid approach aims to provide high prediction accuracy and model transparency by integrating XAI methods such as Principal Component Analysis (PCA) based dimensionality reduction, stable regression modeling, and permutation significance with SHapley Additive exPlanations (SHAP). As a result of the comparison, Linear Regression and Ridge Regression showed the most consistent performance, achieving the lowest error rates and highest coefficient of determination values in both cross-validation and test datasets. Also, methods such as Gradient Boosting and Support Vector Regression provided competitive results. The application of PCA has improved model efficiency by minimizing information loss while reducing the dimensional complexity of the dataset. SHAP and permutation analyses showed that one of the principal components has a significant impact on the model's prediction results. Through these analyses, it's clear which features the model makes decisions based on. The analyses for explainability emphasize the importance of interpretability in modern AI applications. The findings show that the proposed framework successfully balances high accuracy and explainability and is a powerful and scalable tool for environmental decision-making processes by analyzing climate trends. Future work could extend this approach by applying the framework to different climate regions, adding new meteorological variables and integrating more advanced modelling methods.