Estimation Of Reference Crop Evapotranspiration In China Based On CLDAS Reanalysis Data

Reference crop evapotranspiration（ET₀）is a key factor for calculating crop evapotranspiration in practical work.It has important reference value for the planning and design of farmland water conservancy projects,and is of great significance for predicting the water demand of cultivated land and optimizing water resources management.The mainstream calculation method of the ET₀ standard today is the Penman-Monteith（FAO-56 PM）method recommended by the Food and Agriculture Organization of the United Nations（FAO）,which combines energy balance and aerodynamics It can be widely used in different climatic conditions and in different countries or regions,but requires many meteorological factors.Due to the uneven distribution of meteorological stations in my country,it is relatively difficult to obtain accurate and complete meteorological factors in some areas.The ET₀ data set based on the reanalysis product can make up for the discontinuity in time and the insufficiency of space in the surface meteorological platform data.However,rigorous evaluation must be performed to verify the application value of reanalysis products;and when meteorological data are insufficient,there is still confusion about using reanalysis data or machine learning to calculate ET₀.Therefore,this paper evaluates the ability of the second-generation CLDAS data set officially released by the China Meteorological Administration to estimate ET₀ for the first time（using local weather station data as the reference data set）,taking five meteorological factors and ET₀ as the research objects,and exploring different regions in China.Correlation of meteorological data under CLDAS reanalysis data with data from local meteorological stations.The details are as follows:（1）Predict and evaluate five meteorological factors(relative humidity RH,wind speed U₂,minimum temperature T_min,maximum temperature T_max,and solar radiation R_s)based on the downloaded CLDAS grid data and data from 689 local meteorological stations,In addition,the FAO Panmen-Monteith formula was used to calculate the daily ET₀ from 2017 to 2020,and four statistical indicators were used to evaluate the applicability of the method for estimating ET₀ in different years,regions and seasons.The effect of combinations on fit estimation performance.（2）The replacement of local meteorological stations with CLDAS under different data missing conditions was evaluated,and the Penman-Monteith（FAO-56 PM）formula was used to estimate ET₀ in various regions of China.The data were obtained from 43 stations in 7 different climate zones.For the meteorological data from 2017 to 2020,8 parameter combinations are set to calculate,and the ET₀ results calculated by the two machine learning models（XGB,GPR）are compared and analyzed,and it is more suitable to estimate China in the absence of different meteorological data.The regional ET₀ method provides a whole new option for ET₀ estimation!The main conclusions reached are as follows:（1）Except for the Qinghai-Tibet Plateau（QTP）region,the temperature data of CLDAS has high accuracy in all regions,while the global solar radiation data is of average accuracy,and the relative humidity and wind speed data are of poor quality.Except for QTP,the overall accuracy of ET₀ is acceptable,but there are also large errors in less than 15%（103）of the stations.In terms of seasons,the error is largest in summer and smallest in winter.There are also interannual differences in the ET₀ of this dataset.Overall,the CLDAS dataset is expected to have good applicability in the grassland areas of Inner Mongolia,northeastern Taiwan,semi-northern temperate,humid sub-humid warm temperate,and subtropical regions,but there are certain risks in other regions.Furthermore,among all seasons,summer and spring have the smallest deviations,followed by autumn and winter.From 2017 to 2020,2019 and 2020 have the smallest deviations.Coastal regions and borders of different climate zones also underperformed and showed overestimation.（2）In the absence of one kind of meteorological data:When the local data U₂ or RH is missing,it is recommended to use the CLDAS method,that is,replace the data of the local meteorological station with the CLDAS grid data U_2CLD or RH_CLD to calculate,or the method of machine learning to predict ET₀;when the local data R_s is missing,that is,under combination 3,the performance of CLDAS method prediction drops significantly,and it is recommended to use machine learning to predict ET₀.Therefore,in the CLDAS grid data,the solar radiation R_{s CLD} has a great influence on the error of predicting ET₀;In the absence of two kinds of meteorological data:The performance of the three methods predicting ET₀ has decreased to varying degrees.And we suggest that in the absence of meteorological factor data RH and U₂,that is,under combination 6,the corresponding CLDAS grid data can be used instead of local data or machine learning methods to predict ET₀.In addition,combination 6 is also a highly cost-effective combination,only it requires relatively little local measured data to have good prediction performance;and when the lack of data involves R_s,it is recommended to use machine learning methods to predict ET₀.Therefore,it is concluded that the grid data R_{s CLD} affects the estimation accuracy of ET₀ to a greater extent than other meteorological factor data;In the absence of multiple meteorological data:under combination 7,when local data R_s,U₂ and RH are missing,it is recommended to use the machine learning method GPR to predict ET₀.In addition,when all data are missing,it is recommended to replace all grid data to calculate ET₀ and the prediction performance is good.In addition,this paper also analyzes the prediction performance of the CLDAS method in seven regions under different combinations.Generally speaking,there is a better performance in the seven regions.This paper concludes that the performance in regions1 to 4 is relatively good.Well,performance may fluctuate in zones 5-7.To sum up,when CLDAS grid data is used to estimate ET₀,T_{max CLD},T_{min CLD},RH_CLD,and U_2CLDhave higher accuracy,and the calculation results have better accuracy and are recommended to be used,while grid data R_sCLD will greatly reduce the estimation accuracy of ET₀.