Monitoring Of Maize Lodging Disaster Based On Multi-source Satellite Remote Sensing Data

As one of the most crucial grain and fodder in China,maize holds an extremely important position in food security,industrial production,and livestock development.Due to global warming,extreme weather events such as strong winds and heavy rainfall had become increasingly frequent.Combined with poor maize planting and management practices,lodging had become a common natural disaster in maize production.Lodging can result in significant losses in maize yield,reduce grain quality,impede large-scale mechanized harvesting,and have severe impacts on food security,making it one of the major natural disasters in maize production.Fortunately,satellite remote sensing technology has developed to provide efficient and effective monitoring of crop lodging.With advantages such as wide coverage,short cycle,high accuracy,strong reliability,high degree of automation,and high visualization,this technology can provide important monitoring and management support for agricultural production.It also serves as an essential reference for the prevention and mitigation of agricultural disasters.However,large-scale maize lodging monitoring is often faced with cloudy and rainy weather.The research of maize lodging remote sensing monitoring method based on optical and radar image coupling is helpful to improve the data guarantee ability of large-scale monitoring.In this study,the Lishu County,Siping City,Jilin Province is selected as the research area for monitoring maize lodging,and Jilin Province is the region where the technology is applied.The maize population in Lishu County was selected as the research object,and high-resolution satellite imagery from GF-1 and Sentinel-1 were used as the data sources for optical and radar remote sensing to monitor crop lodging,respectively.After constructing the optical and radar remote sensing feature sets of maize lodging,the features were screened,and the most sensitive feature combination for monitoring the severity of maize lodging was obtained.Then,the sensitive feature combinations were classified and validated for accuracy,resulting in the identification of the most sensitive feature combinations and classifiers for monitoring maize lodging using optical and radar remote sensing.Finally,the methods of optical and radar remote sensing were applied in Jilin Province to map the severity of maize lodging.The main research contents and results are as follows:(1)This study used GF-1 WFV remote sensing images as the data source to investigate the method of monitoring maize lodging using optical remote sensing.Spectral features,texture features,and vegetation indices were constructed from GF-1WFV remote sensing images before and after maize lodging.By comparing the feature selection based on Recursive Feature Elimination with Cross-Validation(REFCV)and Mutual Information(MI),the optimal feature combination for monitoring the severity of maize lodging was obtained as ΔRblue,ΔRgreen,ΔRred,ΔNDVI,ΔRVI,ΔEVI,ΔB＿MEA,ΔG＿MEA,and ΔR＿MEA.Random forest classifier was used to extract maize lodging,and the classification accuracy was 87.50% with a Kappa coefficient of0.83.Therefore,with the multiple features extracted from GF-1 WFV remote sensing images,it is possible to monitor the occurrence of maize lodging over a large area.(2)Using Sentinel-1 GRD remote sensing images as data source,a method for monitoring maize lodging based on radar remote sensing was developed.Based on multiple time-phase Sentinel-1 remote sensing images before and after lodging,eight polarimetric feature combinations were selected to form a time-series feature dataset.J-M distance was used to measure the temporal and feature importance of the dataset,and it was found that the sample separation degree of the five Sentinel-1 remote sensing images before and after lodging was the highest.VH time-series polarimetric features were more suitable for monitoring the severity of maize lodging than other temporal features.By comparing the Time-weighted Dynamic Time Warping(Tw DTW)with Random Forest(RF)and Minimum Distance(Min D)in monitoring the severity of maize lodging,it was found that Tw DTW had the highest classification accuracy of 76%and a Kappa coefficient of 0.70,which was higher than RF and Min D.Therefore,longterm features based on Sentinel-1 can effectively monitor the severity of maize lodging at the county scale.(3)The methods of optical and radar remote sensing were applied in Jilin province.The optical remote sensing achieved a regional classification accuracy of 84.38% with a kappa coefficient of 0.63,while the radar remote sensing achieved a regional classification accuracy of 76.67% with a kappa coefficient of 0.60.Both classification accuracies were above 70% and kappa coefficients were above 0.6.The results showed that the methods proposed can effectively monitor the severity of maize lodging in Jilin province.