Study On Remote Sensing Estimation Of Green Crop And Crop Residue Coverage Over Farmland System

Green crops and crop residues are commonly mixed in conservation tillage farmland,the presence of crop residues can cause changes in the spectral reflectance of the canopy,thus affecting the accuracy of multispectral remote sensing estimates of fractional vegetation coverage(FVC),and similarly,the presence of green crops can cause errors in remote sensing estimates of crop residue cover(CRC).Accurate monitoring of FVC and CRC in farming systems can not only provide important information for fine agricultural management,but is also important for studying the role of surface vegetation in the ecosystem.In addition,green vegetation and crop residues are not only found in agricultural systems,but also in grassland,shrub and forest ecosystems.The search for methods to accurately estimate FVC and CRC has wider application scenarios and is important for terrestrial ecosystem carbon storage,vegetation productivity,soil erosion and grassland fires.In this study,we first analyzed the effect of crop residues on conventional vegetation indices(VIs)using mixed crop-crop residue-soil spectral reflectance simulated by the LESS model,and assessed the error in the estimated FVC caused by it based on the dimidiate pixel model(DPM).It was found that NDVI,RVI,PVI,and TSAVI were susceptible to crop residues,while EVI,NDPI,SAVI,and L-SAVI were relatively less affected by crop residues.SAVI and L-SAVI were relatively unaffected by crop residues,and the DPM method based on EVI and NDPI produced the lowest uncertainty in the estimated FVC(|ΔFVC|<0.1).The presence of green crops can lead to an increase or decrease in crop residue index(CRIs)such as NDI,NDTI,and STI.The hyperspectral crop residue indices CAI and h SINDRI are less influenced by green crops,followed by DFI;the DPM method based on CAI produces the least uncertainty in estimating CRC(|ΔCRC|<0.4).Broadband square index(BSI)was proposed using laboratory spectral measurement data,and six BSI indices were screened using correlation analysis,and these indices have good correlation with FVC.The accuracy of BSI estimation of FVC is significantly improved compared to conventional VIs,with BSI134,BSI345 and BSI235 showing higher estimation performance.In the dark-brown earths background,the accuracy of the BSI134 estimation of FVC was close to that of SAVI and NDP;the accuracy of the BSI134 estimation of FVC was highest better than that of NDPI when albic soil was used as the background;the accuracy of BSI134 estimation was also highest for the black soil background case;the accuracy of BSI345estimation was highest when considering all soil background(R~2=0.8955,RMSE=0.0595,MAE=0.0464),which was better than the selected conventional VIs.Validation of the CA-grass empirical data showed that BSI235 estimated FVC better(R~2=0.6897,RMSE=0.1021,MAE=0.0770)than EVI(R~2=0.6446,RMSE=0.1100,MAE=0.0859).The DH-ag empirical validation results show that the better performers are BSI235(R~2=0.8607,RMSE=0.1109,MAE=0.0732)and BSI134(R~2=0.8533,RMSE=0.1139,MAE=0.073),with slightly lower accuracy than NDPI(R~2=0.9,RMSE=0.0925,MAE=0.0663).The broadband slope index(KI)was proposed using the spectrum measured in laboratory.Compared with traditional CRIs,the accuracy of CRC estimation by KIs is significantly improved.Dark-brown earths as background,KI37 has the highest accuracy in estimating CRC,which was higher than that of CAI.In the albic soil background,KI27 had the highest accuracy of estimation;KI27 had the best estimation performance in the black soil background;considering all soil backgrounds,KI37 had the best estimation of CRC with R~2 of 0.8240,RMSE of 0.0697 and MAE of 0.0552,which was better than DFI(R~2=0.7028,RMSE=0.0906,MAE=0.0713).On this basis,the slope information is used to improve the DFI and obtain the DFIK index,which has the highest estimation accuracy in the field dataset CA-grass validation(R~2=0.8451,RMSE=0.0687,MAE=0.0522).In the DH-ag dataset validation,the accuracy of estimating CRC by DFIK is higher than that of traditional multispectral CRIs(R~2=0.4458,RMSE=0.2266,MAE=0.1915).The simultaneous estimation of FVC and CRC was best based on BSI234-KI37 linear spectral mixture analysis model with dark-brown earths as background(FVC:R~2=0.9175,RMSE=0.0489;CRC:R~2=0.8922,RMSE=0.0579).Albic soil as background,BSI235-KI37had the best simultaneous estimation(FVC:R~2=0.5409,RMSE=0.0900;CRC:R~2=0.6291,RMSE=0.0846);BSI234-KI12 had the best estimation performance when black soil was the background(FVC:R~2=0.9051,RMSE=0.0587;CRC:R~2=0.6984,RMSE=0.0699).For the DH-ag field data,the model of L-SAVI-DFIK has the highest accuracy for simultaneous estimation of FVC and CRC(FVC:R~2=0.8837,RMSE=0.1027;CRC:R~2=0.4612,RMSE=0.2247),which is a large improvement compared to other multispectral indices.The BSI proposed in this study can weaken the impact of crop residues,and the KI proposed can weaken the impact of green crops,improving the estimation accuracy of FVC/CRC.The estimation accuracy of DFIK for field data CRC is also higher than that of traditional multispectral CRIs.For synchronous estimation of FVC and CRC,the estimation accuracy of the BSI-KI linear spectral mixture analysis model is higher than that of the traditional VI-CRI estimation.