پیش‌بینی دمای نقطه شبنم با استفاده از مدل‌های مبتنی بر درخت و هسته

Background and Objectives
Agriculture in Iran is faced with special problems due to the diversity of climatic conditions, which is why protecting plants against climate change is very important. One way to prevent damage to the agricultural sector is to predict dew point temperature. Dew point temperature is the temperature at which water vapor condenses into liquid water or dew if the air pressure remains constant. A very high dew point temperature can affect the use of air conditioning and also reduce the efficiency of ventilators that use coolants. The factors affecting the formation of dew phenomena in ecosystems can be called radiation exchange between the surface of the earth and the atmosphere, turbulent heat, and vapor pressure. The magnitude of the dew point is usually measured with a moisture meter, although there are empirical equations for the relationship between air temperature and humidity, but dew point forecasting requires common weather parameters such as relative humidity and rainfall that are not measured at most weather stations or have a large error and are estimated using regression methods. Considering the research in the field of dew point estimation, it is possible to understand the importance of data-driven methods. In this research, we used the models supporting vector regression (SVR), tree model (M5P), rules generation model from tree model (M5Rules), Gaussian process model (GPR), linear regression model (LR), random forest model (RF), and random tree model (RT) to estimate dew point temperature in two stations of Gorgan and ShahreKord.
Methodology
In terms of geographic location, the Gorgan basin is limited between 25°54' east longitude and 36°50' north latitude and the ShahreKord basin is limited between 50°49' east longitude and 32°20' north latitude The city of Gorgan is topographically located to the north of the Alborz Mountains and is adjacent to the Khars Sea from the southeast. That's why Jimmy's in good condition. In terms of topography, Shahrekord is located in the eastern part of the Zagros Mountains and on the margin of the Zagros fault. One of the most important heights is the highest mountain in the world. The input parameters were selected from the Iran Meteorological Organization in the time period of 1990 to 2021. The utilized parameters were the daily maximum temperature (Tmax), daily minimum temperature (Tmin), daily average temperature (Tm), sunshine hours (sshn), average wind speed (ffm), average relative humidity (RHm), maximum relative humidity (RHmax), minimum relative humidity (RHmin). To evaluate the accuracy of input parameters and models, DP was extracted from testing data by Weka software and evaluated with regression and tree models. Additionally, eight possible scenarios were defined to estimate the daily DP.
Findings: The obtained results showed that scenarios 1, 2 and 3 had the highest correlation and scenarios 4, 5, 6, 7 and 8 respectively had the lowest correlation with dew point temperature, but by evaluating these scenarios by evaluation criteria, it can be concluded that scenarios 1, 2 and 3 performed poorer than other scenarios. Therefore, with the placement of most parameters (scenarios 6, 7, and 8), the error of models decreases. According to the results obtained for Gorgan models, R ranges from 1 to 952/0. In Gorgan station, the highest RMSE and the lowest RMSE for (RT-3) with 307/2 and (SVR-8) with 222/0, respectively. In addition, the best fit for the (SVR-8) is (RMSE: 222/0, NSE: 999/0, MBE: 092/0, MAE: 147/0, WI: 1, SI: 017/0) and worst fit for (RT-3) is (RMSE: 307/2, NSE: 882/0, MBE: 875/0, MAE: 745/1, WI: 971/0, SI: 179/0). Also, the obtained results of Shahrekord models for R between the ranges from 996/0 to 615/0. In ShahreKord station, the highest RMSE and the lowest RMSE for (RT-2) with 952/4 and (SVR-7) with 55/0, respectively. In addition, the best fit for the (SVR-7) is (RMSE: 55/0, NSE: 989/0, MBE: 374/0, MAE: 346/0, WI: 997/0, SI: -15/0) and worst fit for (RT-2) is (RMSE: 957/4, NSE: 131/0, MBE: 43/1, MAE: 914/3, WI: 754/0, SI: -352/1).

Conclusion
In this research, using different models such as GPR, LR, M5P, M5Rules, RF, RT, SVR meteorological data were fitted and dew point temperature was obtained in Gorgan and Shahrekord stations. Finally, the performance of the models was presented with different scenarios and the selected models were estimated for estimating dew point temperature. The accuracy of estimation of models using single input parameter including Tmin in two stations of Shahrekord and Gorgan had poorer performance than other input parameters. Estimation of dew point temperature in Gorgan and Shahrekord stations with SVR and M5P models with scenario 8 and 7 respectively was the best performance. Comparison of selected models shows that SVR model has more accuracy than other models and M5P, M5Rules, GPR, RF, LR, RT for Gorgan station were ranked from higher accuracy to less for estimating daily dew point temperature and for Shahrekord station M5P, M5Rules, GPR, RF, RT, LR were rated from greater accuracy to less for estimating daily dew point temperatures. The comparison of tree-based and regression data models was estimated that in general, regression models have high accuracy in estimating dew point temperature compared to tree models.