Efficiency of Machine Learning Techniques for Predicting Vapor Pressure Deficit in Arid and Semi-Arid Regions (Case Study: South Khorasan Province)

Climate change, as one of the global challenges of the present century, has profound impacts on water resources and agriculture. Increase in temperature and decrease in rainfall in arid and semi-arid regions have made the optimal water resource management a top priority. In countries facing climate change and drought, accurate estimation of evapotranspiration plays a vital role in water resource management and ensuring food security. One of the key factors affecting evapotranspiration is the vapor pressure deficit (VPD), which significantly impacts the curacy of related calculations. This study focuses on predicting the vapor pressure deficit using advanced machine learning techniques. The methods employed include Linear Regression ),Generalized Additive Model (GAM), Random Subspace (RSS), Random Forest (RF), and M5 Pruned model (M5P). In this study, monthly average data, including temperature, humidity, precipitation, and vapor pressure deficit, were extracted from the Japanese 55-year Reanalysis (JRA-55) database for the period from 1958 to 2023. The analysis on the vapor pressure it ta in Birjand, Sarayan, Qaen, and Tabas showed that the annual average VPD increased by 6 Pa, 10 Pa, 4 Pa, and 5 Pa, respectively. In the next step, the extracted data for temperature, cipitation, and humidity were used as input variables, and VPD was used as the target variable in machine learning algorithms. Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), Pearson correlation coefficient (CC), and Kling-Gupta efficiency (KGE). Results showed that the GAM model outperformed other models in all regions. The evaluation indices for each region were as follows: Birjand [RMSE=0.308, MAE=0.247, KGE=0.914, and CC=0.920], Sarayan [RMSE=0.401, MAE=0.303, KGE=0.937, and CC=0.951], Qaen [RMSE=0.072, MAE=0.055, KGE=0.987, and CC=0.997] and Tabas [RMSE=0.230, MAE=0.184, KGE=0.920, and CC=0.942]. Predictions showed that, over the next 10 rs, the annual average VPD in the studied regions will significantly increase. This increment is as follows: Birjand 9 Pa, Sarayan 10 Pa, Qaen 7 Pa, and Tabas 5 Pa. This increase signifies serious challenges for water resources and an increase in water consumption. Eventually, this study suggests the GAM model as an effective tool for future research, especially for use in the development of smart irrigation systems, which play a crucial role in sustainable water resource management.