Estimation Standard And Seeded Pan Evaporation Using Modelling Approach

Abstract

Abstract. Evaporation is an important meteorological variable that has also a great impact on water management. In this study, FAO-56 Penman-Monteith equation (FAO56-PM), multiple stepwise regression (MLR) and Kohonen self-organizing map (K-SOM) techniques were used for the estimation of daily pan evaporation (Ep) in three treatments, where C was the standard class A pan with top water, S was A pan with sediment covered bottom, and SM was class A pan containing submerged macrophytes (Myriophyllum sipctatum., Potamogeton perfoliatus, and Najas marina), in an six-season experiment. The modelling approach included six measured meteorological variables; daily mean air temperatures (Ta), maximum and minimum air temperatures, global radiation (Rs), relative humidity (RH), and wind speed (u) in the 2015–2020 growing seasons (from June to September), at Keszthely, Hungary. Average Ep varied from 0.6 to 6.9 mm d−1 for C, 0.7 to 7.9 mm d−1 for S, whereas from 0.9 to 8.2 mm d−1 for SM during the growing seasons studied. Correlation analysis and K-SOM visual representation revealed that Ta and Rs had stronger positive correlation, while RH had a negative correlation with the Ep of C, S and SM. Performances of the different models were compared using statistical indices, which included the root mean square error (RMSE), mean absolute error (MAE), scatter index (SI) and Nash-Sutcliffe efficiency (NSE). The results showed that the MLR method provided close compliance with the observed pan evaporation values, but the K-SOM method gave better estimates than the other methods. Overall, K-SOM has high accuracy and huge potential for Ep estimation for water bodies where freshwater submerged macrophytes are present.

Machine Learning for Pan Evaporation Modeling in Different Agroclimatic Zones of the Slovak Republic (Macro-Regions)

“…ML models based on multiple regression, such as DR in this study, have the best accuracy of results, which is in line with the hypotheses of the previous studies [11,40,42,45]. However, the results do not fit with the theory according to Zounemat-Kermani (2021) [40], where the kriging model, as well as the support vector regression (SVR), radial basis function neural network (RBFNN), and = Levenberg-Marquardt (MLP-ML) models, performed better compared to the RSM (the modified response surface method) and M5Tree (M5 model tree).…”

“…The different studies used different numbers of meteorological stations for their calculations, from one [44,45] or two [9,16,40] up to eight [11,42]. However, in this case study, 35 meteorological stations were used to improve the calculation accuracy, which provides new insight and is an advantage of this study.…”