Monitoring Maize Water Stress Based On UAV Remote Sensing Data

Maize is one of the most important food crops in China and plays an indispensable role in the national economy.However,under the condition of limited agricultural water resources,there is an urgent need for the field irrigation management with higher accurac y and higher effectiveness.It is necessary to achieve the maximum food output with the minimum water input.Accurate and effective estimation of crop water stress distribution and its variability is one of the important ways to achieve this goal.Studies were conducted in maize fields with different water applied levels at the late vegetative,reproductive,and maturation stages in 2013,2015,2018,and 2019 to focus on the accurate and effective estimation of crop water stress distribution and its variability.Multiple sources remote sensing data,including ground RGB and thermal images,and UAV RGB,multispectral and thermal images,maize physiological and biochemical parameters including leaf area index,plant height,chlorophyll content,stomatal conductance,leaf water potential,and sap flow,and meteorological parameters including air temperature/humidity,net solar radiation,and wind speed were obtained.High spatio-temporal maize water stress maps with fine water stress estimation accuracy were obtained by the combination of all the above data,providing technical support for agricultural irrigation management with higher accuracy and higher effectiveness.The main research contents and conclusions of the thesis are as follows:(1)The establishment and analysis of CWSI(Crop Water Stress Index)models for maize.The establishment method of non-water-stressed baseline of the CWSI empirical model was determined based on the analysis of the daily variation of air-canopy temperature difference with vapor pressure deficit.The applicability of different potential canopy resistance calculating methods were comprehensively compared for the establishment of the CWSI theoretical model.Finally,the performances of CWSI empirical and theoretical models for mai ze water stress estimation were analyzed using leaf water potential,sap flow,and stomatal conductance as crop water stress references.The results show that under clear sky condition,air-canopy temperature difference with vapor pressure deficit presents a linear relationship during 11:00 to 15:00 h.The non-water-stressed baseline of the CWSI empirical model could be established based on this linear relationship using simple linear regression.Changes were observed annually and at different study sites for the non-water-stressed baseline of the CWSI empirical model.The main reasons for these changes could be attributed to the differences in climatic conditions,measurement methods of air temperature and humidity,and canopy temperature collection equipment and methods.Different calculation methods of potential canopy resistance would not affect the changing trend of the CWSI theoretical model for different water applied treatments in maize growing season.The CWSI theoretical model constructed by using the inverse method to calculate the potential canopy resistance has a more stable numerical distribution and is more suitable for monitoring maize water status.Compared with the CWSI theoretical model,the CWSI empirical model has a more stable correlation with traditional crop water stress indices,and is more suitable for monitoring maize water stress.(2)Estimating maize water stress based on UAV thermal remote sensing.Due to the limited load capacity of micro UAV remote sensing platform,the integrated thermal camera is usually light-weight uncooled thermal cameras which have problems of unstable temperature collection and difficult extraction of crop canopy temperature.In order to solve the above mentioned problems,the temperature calibration process of the UAV thermal remote sensing image was determined,the extraction method of maize canopy temperature was proposed,and a variety of canopy temperature based crop water stress indices were established and compared with maize stomatal conductance.Finally,the optimal technical process for obtaining the temporal and spatial distribution of maize water stress based on UAV thermal infrared remote sensing was determined.The results show that acquisition height of the UAV thermal infrared remote sensing images would affect the temperature acquisition accuracy.When performing temperature calibration,it is necessary to ensure that images used for correction is consistent with the final thermal infrared data acquisition height.Compared to the reference temperature,higher temperature would be obtained and was needed to be calibrated by the calibration model.Maize plant suffering from water stress has obvious symptoms of leaf rolling and leaf area reduction,compared to maize plant which does not suffer from water stress.When maize plant was not suffering water stress,more pixels with lower temperature would appear and thus higher canopy temperature extraction error would be obtained.With the symptoms of leaf rolling and leaf area reduction,more pixels with higher temperature would appear and thus higher canopy temperature extraction error would be obtained.Therefore,the extracted maize canopy temperature would first be underestimated and then overestimated compared to the reference value.When stomatal conductance was used as maize water stress reference,CWSI_Ehas the best maize water stress estimation performance with R² of 0.63 for 2018 and 0.42 for2019 with stomatal conductance.The corresponding RMSE is 0.12 and 0.16 mol·m^-2·s^-1,respectively.(3)Estimating maize water stress from UAV multispectral images.In order to further study the feasibility of obtaining maize water stress distribution and its variability based on vegetation index derived from UAV multispectral remote sensing images,and to explore whether the nonlinear machine learning regression algorithm will enhance the water stress estimation performance of vegetation index,the selection of vegetation indices and analysis of their crop water stress estimation performances,and the establishment of water stress estimation model based on non-linear machine learning algorithms were conducted.Finally,the optimal technical process for obtaining the temporal and spatial distribution of maize water stress based on UAV multispectral remote sensing was determined.The results show that UAV multispectral vegetation indices responded well to different water applied treatment and changes in water application during maize growth stages in 2018 and 2019.Among them,NDVI and TCARI are respectively the optimal vegetation index used to describe the change of maize canopy structure and chlorophyll content.At the same time,leaf area index and chlorophyll content had positive correlations with stomatal conductance(p<0.001).Compared with non-linear regression algorithms of random forest and BP neural network,multiple linear regression has the best stomatal conductance estimation performance with R² of 0.64 for 2018 and 0.58 for 2019,and with RMSE of 0.13 for 2018 and 0.12 for 2019,showing that the linear regression algorithm had better performance and could meet the demand when estimating maize water stress water stress based on the UAV multispectral vegetation indices.(4)Research on maize water stress estimation based on UAV RGB remote sensing.In order to explore whether the vegetation index distribution of maize canopy in the RGB image could effectively estimate crop water stress,and the feasibility of obtaining maize water stress distribution and its variability based on the UAV RGB remote sensing image,the construction of a new vegetation index and analysis of its water stress response,and applicability analysis of the new index at the UAV remote sensing scale were carried out.Finally,the optimal technical process for obtaining the temporal and spatial distribution of maize water stress based on UAV RGB remote sensing was determined.The results show that the mean value of Ex G Gaussian distribution of the maize canopy in the RGB image could be used as a new crop water stress index(named MGDEXG)to estimate maize water stress,which could distinguish different gradients of water applied treatment and could respond well to different water applied amount at different maize growth stages.MGDEXG had significant correlations with traditional crop water stress indices.Compared with the widely used CWSI_E,MGDEXG has two main advantages:1)MGDEXG requires only RGB imagery,making it easier to calculate,and it is cheaper,making it easier to popularize;2)MGDEXG is resistant to the micro-meteorological conditions within the field.The resolution reduction of RGB images based on the bilinear interpolation algorithm would not have a significant impact on the crop water stress estimation performance of MGDEXG,which lays the foundation for the subsequent rapid extraction of MGDEXG.MGDEXG could effectively estimate the maize water stress at the two UAV RGB resolution scales of 2.7 mm(10 m)and 14 mm(50 m)with the respective R² of 0.78 and 0.62and RMSE of 0.09 and 0.11 mol·m^-2·s^-1 with stomatal conductance.MGDEXG had significant negative correlations with CWSI_E at different segmentation scales of UAV RGB orthophoto images of 2 m×2 m,4 m×4 m,6 m×6 m,8 m×8 m,10 m×10 m and 12 m×12 m with corresponding R² of 0.74,0.75,0.75,0.75 and 0.76,indicating that different segmentation scales would not affect the water stress estimation performance of MGDEXG.(5)Maize water stress distribution and its variability could be obtai n effectively based on UAV RGB,multispectral,and thermal remote sensing.By analyzing maize water stress maps derived by UAV multi-source remote sensing images,it could be found that heterogeneous of water stress could observed in both full irrigated areas and areas suffering water stress due to the heterogeneity of soil and crop,indicating that more accurate irrigation management is of great significance.In addition,compared with UAV multi-spectral and thermal infrared remote sensing data,maize water stress index extracted from UAV RGB remote sensing data had a better correlation with stomatal conductance,and it is more suitable as a low-cost way to estimate crop water stress distribution and its variability.