Assessment of Data-Driven Models for EstimatingRelative Humidity, case study of Bajgah, Iran

Meteorological variables, particularly relative humidity, play a pivotal role in influencingvarious aspects of human life and natural environments, spanning agriculture, waterresource management, and renewable energy systems. Acknowledging the complexity ofdeveloping equations for relative humidity prediction based on meteorological variables,which requires knowledge of the underlying relations between them, this study underscoresthe advantage of machine learning models that do not rely on explicit equations.Considering limited studies in this domain, this study utilized six machine learning modelsfor predicting daily mean relative humidity, employing historical mean relative humidity,minimum temperature, and maximum temperature from one to three preceding days(resulted from Pearson correlation analysis) as input variables. The models, includingArtificial Neural Network, Decision Tree Regressor, Random Forest Regressor, AdaBoostRegressor, Gradient Boosting Regressor, and XGBoost Regressor, are evaluated for theirperformance using a comprehensive ۳۰-year dataset from Bajgah meteorological station inIran. The assessment involved the application of six distinct statistical criteria, i.e. RMSE,MAE, MARE, MXARE, R۲, and NSE, which highlighted the commendable performanceof all models utilized in this study. Based on a ranking method, the Random ForestRegressor, XGBoost Regressor, and Gradient Boosting Regressor models jointly achievedthe top rank, with Random Forest Regressor excelling in training data, XGBoost Regressorin test data, and Gradient Boosting Regressor demonstrating robust performance acrossboth datasets. Following these top-performing models, the Artificial Neural Networkmodel secured a subsequent ranking. In contrast, the Decision Tree Regressor andAdaBoost Regressor models exhibited lower performance compared to their alternatives.The study highlights the potential of machine learning models in predicting relativehumidity and encourages future exploration of alternative models.

کلیدواژه ها:

Hydrology ، Machine learning ، Meteorological variables ، Relative humidity ، XGBoost