Estimation Of Evapotranspiration In Vineyard In Northwest Subhumid Region Based On UAV Remote Sensing

Grapes are one of the most important fruits cultivated in the Guanzhong region of Shaanxi.Although the irrigation conditions in the Guanzhong plain are relatively favorable,there is still a need to rationally allocate the limited water resources and improve water use efficiency.Accurate estimation of evapotranspiration during the grapevine reproductive period is important to develop a reasonable irrigation system to provide appropriate water supply to ensure high quality and high yield of grapes.Most of the traditional methods for obtaining evapotranspiration are costly and complicated to apply.This paper used the low-altitude UAV thermal infrared and multispectral remote sensing technology to obtain grape temperature information and spectral characteristics,using the evapotranspiration model based on grape canopy temperature and the double crop coefficient method after inversion of grape base crop coefficient K_cbwith spectral feature to estimate the evapotranspiration of vineyard in the northwest semi-humid zone,and verify the feasibility of the UAV remote sensing technology to estimate the evapotranspiration of grapes at farm scale.The main results obtained are as follows:（1）The characteristics of canopy temperature distribution obtained by grapevine UAV thermal infrared remote sensing were revealed.The P-M canopy impedance model and the Seguin-Itier model for estimating daily grape evapotranspiration were compared with no significant differences in the results.It was found that the canopy temperature of grapes was normally distributed by UAV thermal infrared remote sensing,with a wide range of canopy temperature distribution in the middle of the growth stage and a more concentrated distribution in the early and late growth stage.The differences in the canopy temperature of grapes were not significant in the same period within two years.The daily evapotranspiration and its variation trend were estimated by the P-M canopy impedance model,which was consistent with the growth and development of grape.The daily evapotranspiration and daily net radiation values measured by the Bowen-ratio system,as well as the canopy temperature and air temperature from11:00 to 14:00,were used to rate a and b in the Seguin-Itier model.13:00 was found to be the most suitable time to rate a and b.The Seguin-Itier model combined with canopy and air temperature difference was used to estimate the actual daily evapotranspiration of grapes and its variation trend,which had no significant difference from the P-M canopy impedance model.（2）Various forms of inversion models of vegetation index and base crop coefficient K_cbwere established and the model with the highest fitting accuracy was selected.According to the grape spectral data obtained by UAV multi-spectral remote sensing,the inversion models of grape vegetation index and base crop coefficient K_cbat different growth stages were established by using unitary linear regression,polynomial regression and multiple linear regression.Through comparative analysis of the fitting accuracy of various models,it is found that the vegetation index type,growth stage and modeling method selection are three important factors affecting the fitting accuracy of the model.Under the condition of the same modeling methods,most of the fitting accuracy of the 2021 and 2022models is highest in the early growth stage,followed by the later growth stage,and finally in the whole growth stage.Based on the 2-year model fitting accuracy data,it was found that NDVI had a good performance under different conditions,and the highest fitting accuracy of the polynomial regression model in the early growth stage in 2022 was 0.89.When the modeling methods are different,the multivariate linear regression model generally has the best fitting effect,followed by the polynomial regression model,and the last is the unitary linear regression model.However,the unitary linear regression model also has high fitting accuracy,and the three differences are not significant.（3）The accuracy of several grape evapotranspiration models based on UAV remote sensing information was verified and the best estimation model was proposed in the different stage.Compared and verified the measured evapotranspiration value of grape calculated by Bowen-ratio system and water balance method with the predicted evapotranspiration value estimated by UAV thermal infrared and multi-spectral remote sensing technology.It was found that when using Bowen-ratio system as the benchmark,the estimation accuracy of P-M canopy impedance model and Seguin-Itier model based on UAV thermal infrared remote sensing to obtain canopy temperature reached above 0.79 in 2021 and 2022.These two models have good applicability and certain feasibility.In 2022,the accuracy of Seguin-Itier model was the highest（EF=0.87,RMSE=0.68mm/d）.The accuracy of grape evapotranspiration verification of K_cbinversion model based on UAV multi-spectral remote sensing is affected by growth stage,vegetation index type and modeling method,which are also factors affecting the fitting accuracy of K_cbinversion model.Multi-spectral remote sensing of UAV was used to estimate evapotranspiration of vineyards in the subhumid region of northwest China.The accuracy of estimating evapotranspiration in 2022 was the highest with the unary linear regression model at the early growth stage of NDVI（EF=0.87,RMSE=0.63mm/d）.When the water balance method is used as the benchmark,the estimation accuracy of both models decreases,but it does not mean that the model results are unreliable.The analysis shows that the difference in accuracy may be caused by the insufficient accuracy of the data measured by the water balance method,which cannot fully match the time scale of UAV operation,and the limited amount of effective data.Among them,the estimation accuracy of NDVI was the highest in 2021 polynomial regression model（EF=0.81,RMSE=0.74mm/d）.