Study On Water And Fertilizer Effects And UAV Spectral Monitoring Model For Growth Parameters Of Sugarcane

With population growth,socio-economic development and climate change,sugar safety had become a key issue of concern.Sugarcane was the main raw material for sugar in China,accounting for 90%.Due to the fact that sugarcane was mainly planted on dry slopes,optimizing water and fertilizer regulation was an effective means to improve sugarcane yield and quality.In recent years,the development of modern sugar industry required high yield and sugar content in sugarcane,which had put forward higher requirements for accurate diagnosis and scientific management of sugarcane growth.The use of drone hyperspectral cameras,thermal infrared cameras,and multispectral cameras to obtain remote sensing data for accurate,efficient,and rapid monitoring and analysis of sugarcane growth parameters had become an important foundation for agricultural production management.The growth effects of sugarcane under different water and fertilizer treatment conditions through field experiments were studied and analyzed by the thesis.At the same time,the models of soil moisture content,crop coverage,plant height,stem number,leaf area index(LAI)and chlorophyll content(CC)were conducted by UAV multi-source remote sensing platform.It provided data support for sugarcane yield estimation models,which supporting accurate diagnosis and intelligent management of sugarcane fields.The main research contents and conclusions of the thesis were as follows:(1)Based on the analysis of field experiment data with different water and fertilizer gradients throughout the entire growth cycle of sugarcane for three consecutive years in2018,2019,and 2020,a model for sugarcane yield and sugar content in dry sloping land under different water and fertilizer treatments was established,incorporating parameters such as sugarcane plant height,plant stem,and effective stem,with irrigation amount,fertilization amount,and water fertilizer coupling as variables and yield and sugar content as goals.The model results indicated that when the irrigation and fertilization amounts were1778.40m~3/ha and 879.00 kg/ha,the maximum yield of sugarcane was 105.20t/ha while maintaining a relatively stable sugar content.(2)PROSAIL radiation transfer model for sugarcane was established,and equivalent water thickness was added to simulate the spectral reflectance of sugarcane under different water treatments.It was revealed that the effects of LAI,leaf angle direction(LAD),and CC under different water treatments on the overall reflectance of sugarcane canopy were concentrated in the short wave infrared,near infrared,and short wave infrared bands,as well as visible light bands.The optimal vegetation indices of LAI and CC of sugarcane were(ρ120nm-ρ1900nm)/(ρ120nm+ρ1900nm)and(ρ45nm-ρ310nm)/(ρ45nm+ρ310nm).(3)An empirical model of Crop Water Stress Index(CWSI)for sugarcane was constructed based on drone thermal infrared and sugarcane growth characteristics.The correlation between CWSI and sugarcane soil moisture content was analyzed,and a sugarcane soil moisture monitoring model based on multiple machine learning regression algorithms was established.The research results showed that algorithms such as GBDT,Adaboost,and XGBboost had shown good estimation results for soil moisture content from0-10cm,10-20cm,20-30cm,and 30-40cm from 2018 to 2020.R~2values were 0.869～0.882,0.823～0.864 and 0.750～0.849,RMSE were 0.498～0.667,0.788～1.052 and 0.930～1.464,which were superior to linear regression.(4)Based on drone multispectral and hyperspectral extraction,the sugarcane canopy vegetation index(VI)was extracted,and VIs with strong correlation with sugarcane LAI and CC were selected.An improved VI based on spatiotemporal and multi temporal sequences was proposed,and a monitoring model for sugarcane LAI and CC were established based on multiple machine learning regression algorithms.The research results indicated that the correlation between multispectral VI,LAI,and CC were lower than that of hyperspectral VI,but the operational efficiency were higher than that of hyperspectral VI.The mh CARI,mh MCARI,mh RDVI,and mh NDVI indices constructed by fusing hyperspectral and multispectral data showed significant spectral increases of 0.180～0.398and 0.150～0.250 compared to the spearman coefficients of sugarcane LAI and CC.The R~2estimation of LAI and CC of sugarcane based on GBDT和XGBoost algorithms had increased by 0.176～0.184 and 0.093～0.107,respectively,compared to multispectral estimation.(5)Based on monitoring model and its estimation results of soil moisture and CC,LAI,effective stem,plant stem and plant height,as well as related vegetation indices by the UAV spectral,an empirical statistical yield estimation model,a semi empirical and semi mechanistic yield estimation model,and a new geometric statistical yield estimation model for sugarcane were constructed.The research results indicated that the sugarcane geometric statistical yield estimation model(GSYEMS model)established by integrating the influencing factors of sugarcane yield has the highest estimation accuracy,R~2was 0.8665and RMSE was 13.458kg.The empirical statistical yield estimation model for sugarcane based on the combination of OSAVI at tillering stage,SPVI at elongation stage,and RDVI at maturity stage takes the second place,R~2was 0.8437 and RSME was 20.402kg.The semi empirical and semi mechanistic yield estimation model for sugarcane has low accuracy,R~2was 0.7300 and RSME was 24.832kg.Taking into account factors such as estimation accuracy and modeling efficiency,it was recommended to use the sugarcane geometric statistical estimation model based on drone spectra for large-scale sugarcane production estimation in the region.