Relative Humidity Prediction using XGBoost Machine Learning Model, Case Study: Bajgah Climatological Station, Iran

Subject Areas : Article frome a thesis

Reza Piraei ¹ , Ali Mohammadi ² , Seied Hosein Afzali ³

1 - Department of Civil and Environmental ,Engineering, Shiraz University, Shiraz, Iran.
2 - School of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran
3 - Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran

Received: 2023-12-06 Accepted : 2024-02-14 Published : 2024-02-14

Keywords: Relative humidity, XGBoost Model, Machine Learning, Bajgah,

Abstract :

Relative humidity is one of the most important hydrological parameters that significantly influences evapotranspiration water resource management, plant growth and even concrete settings. Hence, accurate prediction and estimation of relative humidity paramount importance. In this study, since two parameters relative humidity and the minimum and maximum temperatures of preceding days, have the most significant impact on predicting future relative humidity, and given the prevalence of available data for only these two parameters in many parts of the country, various scenarios involving these parameters were studied. The best scenario for predicting relative humidity was obtained using the XGBoost model. To assess the accuracy of the model, the Bajgah region in Fars Province was chosen as a case study, and the accuracy of different scenarios was compared using data from the past 30 years (1993 to 2023). In this regard, missing data were estimated using the KNN Imputer model. The correlation between mean relative humidity of one to ten days before and the target variable (predicted relative humidity on day t) was calculated using Pearson correlation. Based on the results indicating the insignificance of data from the fourth day and earlier, data from one to three days before were utilized. Finally, by comparing the results based on six statistical criteria (RMSE, MAE, MARE, MXARE, NSE, and R2), it was determined the scenario based on relative humidity and the maximum and minimum temperatures of the preceding 3 days provides the best estimation.