Prediction of daily evaporation values using support vector regression coupled with firefly algorithm in the coastal areas of Iran

In the hydrologic cycle, evaporation is the primary step that causes water loss. Evaporation takes into account various parts of the water balance under completely different climates, and its correct prediction is very important for water resources management. The importance of evaporation and its impact on surface water balance is highlighted through its relation to climate change and global warming. The latest outputs of meteorological models suggest that global warming has caused an increase in evaporation from the land surface and surface water bodies, which is anticipated to have a serious impact over time on water resources management and the global population. In arid and semi-arid regions, accurate prediction of evaporation is very important for decision-makers due to water scarcity. Estimating daily evaporation with the highest accuracy and in the fastest possible time is essential to determine the water needs of different products, design irrigation programs, and manage water resources in different areas, especially when there is insufficient meteorological information. Evaporation has complex and non-linear behavior. Also, the evaporation parameter is not measured in some meteorological stations. Furthermore, meteorological stations are not correctly distributed in many developing countries including Iran. Since coastal areas have more evaporation than others, in many cases the amount of evaporation is higher than the global average. Despite the high importance of evaporation in coastal areas, very few studies have predicted this parameter in Iran. Moreover, accurate prediction of water loss in these areas leads to a better understanding of the hydrological cycle and is essential for optimal water management and agriculture. Thus, the purpose of this research is to predict daily evaporation values in four coastal stations of Abadan, Ramsar, Bandar Abbas, and Bandar Anzali.

The main meteorological parameters including average relative humidity, minimum relative humidity, maximum relative humidity average temperature, minimum temperature, maximum temperature, sunshine hours, and wind speed, under separate scenarios, as input for support of vector regression (SVR) and SVR with firefly algorithm (SVR-FFA) for estimating evaporation values were used on a daily scale. Statistical parameters in the time period of 1990-2021 were utilized as input to the mentioned models. In order to evaluate the performance of the implemented models, various statistical parameters were used, including correlation coefficient (R), root mean squared error (RMSE), Nash-Sutcliffe coefficient (NS), and Willmott's Index of Agreement (WI). To better estimate the daily evaporation values, eight different scenarios were used as the combinations of input parameters.

Based on the obtained results for all studied stations, the SVR-FFA-8 showed the least error with RMSE = 2.843 (mm day-1) for Abadan station, RMSE = 1.13 (mm day-1) for Ramsar station, RMSE = 1.985 (mm day-1) for Bandar Abbas station and RMSE = 1.225 (mm day-1) for Bandar Anzali station. For the indices of correlation coefficient, Nash-Sutcliffe coefficient, and Wilmott’s index of agreement, the SVR-FFA-8 model also indicated in the highest values between observed and predicted amounts. Also, the indices of correlation coefficient, Nash-Sutcliffe coefficient, and Wilmott’s index of agreement illustrated the highest accuracy in Abadan station for all combinations compared to other stations, which shows the high correlation of observed and predicted values in this station. After SVR-FFA-8, SVR-FFA-7 model in Abadan and Bandar Anzali stations and the SVR-FFA-6 in Ramsar and Bandar Abbas stations showed acceptable performance. Thus, the RMSE for Abadan and Bandar Anzali stations is 2.995 (mm day-1) and 1.272 (mm day-1), respectively, and for Ramsar and Bandar Abbas, 1.176 (mm day-1) and was obtained 1.993 (mm day-1). Comparing the results of SVR combinations also revealed that for Abadan, Ramsar, and Bandar Anzali stations, SVR-8 and for Bandar Abbas station, SVR-6 showed the highest accuracy among all SVR combinations in all four studied stations. Also, Ramsar station presented the lowest RMSE compared to other stations. After the SVR-8 model for Abadan, Ramsar, and Bandar Anzali stations, the SVR-7 and SVR-6 models for the Bandar Abbas station showed a weaker performance due to having less input parameters. The comparison between the input parameters also concluded that the sunny hours is the most important parameter in predicting the daily evaporation values in all four stations, thus increasing the accuracy of the models.