Estimation And Prediction Of Reference Crop Evapotranspiration In Ili River Valley

Reference crop evapotranspiration（ET₀）is one of the important factors for estimation of crops water requirement.Thus,accurate estimation of ET₀ can achieve efficient water-saving irrigation,also provide support for water resources planning and optimal utilization.FAO Penman-Monteith（P-M）equation is recommended as the standard model to predict ET₀.However,its application is restricted by lack of complete meteorological data in many regions.Therefore,it is imperative to find a model that can accurately estimates ET₀ with incomplete meteorological data.Based on the daily meteorological data of eight meteorological stations in Ili River Valley from 1980～2019,the P-M equation is used to calculate the daily ET₀.The spatial-temporal distribution characteristics of the ET₀ and meteorological factors are analyzed by using the linear regression,Mann-Kendall mutation test,cumulative anomaly and ANUSPLIN interpolation methods.Also,correlation analysis and path analysis are used to explore the impact of meteorological factors on ET₀.Using two commonly used empirical formulas（Pristley-Taylor,Hargreaves-Samani）and three machine learning models（Multiple Adaptive Regression Spline（MARS）,Generalized Regression Neural Network（GRNN）,Gradient Boosting Tree（GBDT））to calculate ET₀.The results are evaluated by mathematical statistics methods.Finally,the R/S analysis is used to predict the future changes of ET₀ in the Ili River Valley.Conclusion have been drawn as follow:（1）From 1980 to 2019,T_max and T_min show a significant upward trend,while the RH,n and u₂show a significant downward trend.The interannual variation trend of ET₀ is consistent with T_max and T_min.（2）T_max and T_min show an upward trend in four seasons,while the RH and u₂ show a downward trend.The n shows a downward trend in all seasons except spring.The ET₀shows an upward trend in spring with a maximal tendency rate,also in summer,which are different in autumn,the tendency rate of ET₀ in spring is 7.606 mm·d^-1·10a^-1,the tendency rate of ET₀ in autumn is-0.232mm·d^-1·10a^-1.However,the change in winter is more complex,with a trend of"rising-falling-rising".（3）Except for RH,the meteorological factors present a single peak.The high value of temperature appears in July,and the low value appears in January.The maximum of n appears in July,followed by June and August,the minimum appears in December.The u₂reaches the maximum in April,the minimum appears in December and January.The distribution of RH is different from others.The annual variation of the ET₀ is basically the same as T_max and T_min.（4）T_max and T_min are higher in the north and lower in the south.The RH increases from the northwest to the south.On the contrary,the high value of n appears in the northwest and the low value appears in the southern mountainous area,while u₂ in the mountainous area is higher than that in the valley.The difference of the spatial distribution of the ET₀ is obvious.The ET₀ in the whole region ranges from 406.22mm/d to954.65mm/d.Generally,the ET₀ in the northern is larger than that in the southern,in valley is larger than that in mountain areas.There is a relationship between the spatial distribution of the ET₀ and T_max,T_min,n,u₂ and RH.（5）Correlation coefficient and path analysis are used to analyze the sensitivity of ET₀.On the whole,the sensitivity is ranked:RH,T,n and u₂.RH is negatively correlated with ET_0,and the other three factors are positively.The sensitivity ranking of four meteorological factors in different stations is slightly different.（6）Among the two empirical equations,the Priestley-Taylor method has the minimum error,the results are statistically consistent with P-M to some degree.Eight input combinations were used to establish MARS,GRNN and GBDT models respectively.The results show that,the precision of the model increases with the added of input factor.When the number of factors is the same,the precision is related to the combination.The accuracy of MARS,GRNN and GBDT are better than empirical equations,MARS have the best performance and the highest accuracy compared with GRNN and GBDT.（7）R/S analysis is used to predict the future trend of ET₀in Ili River Valley.The Hurst index of Horgos,Huocheng,Chabuchar,Nileke,Yining County,Xinyuan,Zhaosu,Tekes and the whole Ili River Valley is 0.67,0.92,0.75,0.97,0.97,0.85,0.71,0.91 and0.95 respectively.ET₀ will continue to decrease in Huocheng and Nileke,and will continue to increase in the others.Besides,the duration is 8a,17a,11a,18a,18a,11a,11a,12a and 17a respectively.