Calculation Of Reference Crop Evapotranspiration Based On Machine Learning Algorithm In Shaanxi Province

Reference evapotranspiration?ET₀?is a major research area of hydrology and water resources management,especially in agricultural irrigation.The accurate calculation of ET₀can provide theoretical basis for rational planning of irrigation areas,formulation of irrigation plans,and optimal allocation of water resources.As the ET₀ standard calculation formula recommended by FAO,the FAO P-M formula is the closest to the actual value.However,this formula requires a large amount of meteorological data and is difficult to apply in areas lacking meteorological data.Therefore,obtaining ET₀ calculation model with low input meteorological data,high calculation accuracy,and wide application range becomes crucial.In this paper,57a?1960?2016?daily meteorological data from 30 meteorological stations in Shaanxi Province are used to calculate the standard ET₀ using the FAO P-M formula.Analyze the spatiotemporal distribution characteristics,sensitivity,perform periodic analysis and prediction of ET₀.In addition,the applicability and portability of ANFIS,RF,and SVM three machine learning algorithms in ET₀ calculation in Shaanxi Province are analyzed,and compared with other ET₀ simplification methods.The main conclusions are as follows:?1?The multi-year average ET₀ of Shaanxi Province is 978.77 mm.The annual ET₀ in Shaanxi and Guanzhong areas is decreasing,and an upward trend in northern and southern Shaanxi.Except for Guanzhong,the change trend in other areas is not significant.Except for the significant upward trend in summer,ET₀ showed no significant upward trend in other seasons.The ET₀ in Shaanxi Province tended to weaken from north to south,and there is a big difference between the north and the south.The ET₀ is high in the north and low in the south in spring and summer,and becomes high in the south and low in the north in autumn and winter.?2?The R/S analysis of ET₀ in Shaanxi Province for many years showed that the Hurst indices of Shaanxi,Northern Shaanxi,Guanzhong,and Southern Shaanxi were 0.775,0.707,0.806,and 0.845,respectively.ET₀ would continue to increase in northern Shaanxi,and other regions would continue to decrease with a duration of 9 years.The Hurst index of ET₀in spring,summer,autumn and winter was 0.699,0.855,0.526,and 0.560.ET₀ would continue to increase in spring and autumn,and would continue to decrease in summer and winter.The duration of ET₀ changed in spring,summer,autumn and winter was 9 a,5 a,10 a,9 a.The wavelet analysis of ET₀ in Shaanxi Province for many years showed that the ET₀cycle variation characteristics in different regions and seasons were similar.The cycle change is most obvious in the 24?32 a scale and the main cycle scale is about 28a.The cycle length is about 18a and experienced about 3 cycles in 57a.?3?The conclusions obtained from the sensitivity analysis of ET₀ using the three methods of correlation coefficient method,partial derivative sensitivity analysis,and path analysis were consistent.Except that the relative humidity was negatively correlated with ET₀,all other meteorological factors were positively correlated with ET₀.The magnitude of sensitivity is:temperature>sunshine hours>relative humidity,wind speed.The rankings of relative humidity and wind speed are slightly different at different sites.Temperature contributes the most to ET₀ contribution rate,and temperature is the main driving force for ET₀ change.?4?In this paper,four kinds of meteorological factors were used to construct 15 kinds of meteorological factor input combinations,so as to construct ET₀ calculation models such as ANFIS11?ANFIS15,RF11?RF15,SVM11?SVM15.ET₀ calculated by FAO P-M was used as an evaluation standard to compare the model simulation results.The temperature and sunshine hours increased the accuracy of the model significantly more than the relative humidity and wind speed.As the meteorological factors increase,the accuracy of the model would increase.The ANFIS15,RF15,and SVM15 models constructed using four meteorological factors had the highest accuracy.Although the meteorological data used were the same as those of FAO P-M,the calculation results still had certain differences.SVM was the best simulation among three machine learning algorithms.When using the same meteorological factor to calculate ET₀,the SVM model was better than HS and Iramk.Makkink,PT and other four traditional ET₀ calculation methods,SVM could be applied to ET₀ calculation in areas lacking meteorological data in Shaanxi Province,instead of other ET₀ simplification methods,to improve the calculation accuracy.These three machine learning algorithms have good portability in the calculation of ET₀.In Shaanxi Province,trained ANFIS,RF,SVM and other models can be applied to areas with similar meteorological conditions.The simulation effect is different,so it is necessary to choose a suitable machine learning algorithm to establish the ET₀ calculation model according to the use situation in different regions.