Estimating Fractional Covers Of Crop And Crop Residue In Cropland Using Multi-spectral Remote Sensing And Spectral Mixture Analysis

Cropland is a mixed vegetation-soil ecosystem that includes vegetation(crop),crop residue,and soil.The accurate estimation of fractional vegetation cover(FVC),crop residue cover(CRC),and bare soil(BS)in agricultural and vegetation-soil ecosystems can help decision making for agricultural management organization.In addition,the dynamic change of CRC,FVC,and BS is critically important for understanding the interrelations of global changes and terrestrial ecosystems.The spectral reflectance of crop residue and soil are very similar in visible and near-infrared bands;the cellulose and lignin absorption in 2100-2300 nm are the most frequently used features to discriminate crop residues from soils.In practices,most satellites high-resolution multispectral remote sensing sensors can only provide the surface reflectance in board bands.On the one hand,the reflectance of the crop residue and soil are similar;on the other,the crop residue spectral reflectance features were covered by moisture absorption.Thus,long-term monitoring of CRC,FVC,and BS in cropland areas using on mid-to high-resolution multispectral remote sensing still remains many challenge.This study focuses on(i)analyzing the spectral features of crops,crop residues,and soils and their response to moisture;(ii)estimating CRC,FVC,and BS based using soil-crop residue and soil-crop-crop residue multispectral remote sensing mixed pixel models;and(iii)evaluating the moisture effect on remote sensing soil-crop residue and soil-crop-crop residue mixed pixel models based CRC,FVC,and BS estimation.Both the soil-crop residue and soil-crop-crop residue mixed pixel scene were simulated in a spectral laboratory and a winter-wheat field;laboratory-and field-based Field Spec 3hyperspectral,Sentinel-2 MSI and MODIS images,and corresponding FVC,CRC,and BS were collected.Firstly,this work explains how soil moisture and crop residue moisture affect the soil,crop residue,soil-crop residue and soil-crop-crop residue mixed spectral and spectral indices.Then,we present a dynamic soil end-member spectrum selection approach for improving the performance of soil and crop residue spectral un-mixing analysis from soil-crop residue mixed spectral.Additionally,we developed a crop residue index which is not sensitive to moisture effect,namely broadband crop residue angle index(BAI).The effectiveness and practicability of BAI and multi-BAI was evaluated using both laboratory-,field-based spectral datasets with Sentinel-2 MSI images.Finally,we applied BAI–NDVI and multi-BAI–NDVI triangle space methods for field FVC,CRC,and BS estimation and mapping using Sentinel-2 MSI and MODIS images.This study presents the following conclusions:(1)The soil,crop residue,and crop moisture content significantly affects the spectral reflectance and optical SIs.Crop canopy spectral reflectance of near-infrared(NIR)and short wave near-infrared(SWIR)region was decreased with the increasing leaf water content.The cellulose and lignin absorption features in 2100-2300 nm were covered by water absorption with the increasing crop residue moisture content.Two water-absorption bands(near 960 and 1100 nm)dominate the soil reflectance in NIR region,and three water absorption bands(near 1450,1900,and 2500 nm)dominate the reflectance in SWIR region.Overall,the stronger the absorption capacity,the faster the soil reflectance decreases and the higher the saturated soil moisture.The moisture effects on vegetation index(VI),crop residue spectral index(CRSI),and water index(WI)for crops were smaller than that of soils and crop residues.CRSI and WI for soil keep increase with the increasing soil moisture content when VI almost unchanged.The response of VI,CRSI,and WI for soils and crop residues on moisture are similar.(2)This study presents a dynamic soil end-member spectrum selection approach for improving the performance of soil and crop residue spectral un-mixing analysis in CRC estimation.The new method used the SM content to obtain an accurate soil end-member spectrum for each pixel,resulting in improved CRC estimation accuracy.This new method(R~2=0.88,RMSE=0.08,MAE=0.06)was more powerful than that of traditional fixed min and fixed mean soil end-member spectrum selection(R~2=0.47,RMSE=0.17,MAE=0.14).Our results demonstrated that,1)when using the fixed min approach,the estimated CRC in low-SM areas was overestimated;2)when using the fixed mean approach,the CRC in high-SM areas was underestimated and the CRC in low-SM areas was overestimated;and 3)error originated primarily from low-SM areas.Because SM plays an important role in soil spectra,the SM distribution information is of significant importance to cropland CRC estimation and mapping.(3)This study propose a new method for estimating fractional cover that uses a broadband spectral angle index(BAI)to estimate CRC.The proposed BAI is the included angle between(i)the line from the reflectance at band a to the reflectance at band b and(ii)the line from the reflectance at band b to the reflectance at band c,where bands a and b represent the VIS or NIR bands,respectively,and band c represents the SWIR1 or SWIR2 of the broadband remote sensing band.The proposed BAI method can mitigate the effects of soil and crop residue moisture content on spectral reflectance.We used two laboratory-based treatments(mixed spectral reflectance of dry,saturated soil and crop residue and the mixed spectral reflectance of winter wheat leaf and dry,saturated soil and crop residue)to analyze the response of BAIs to CRC,SM-CRM,FVC,and vegetation water content.Next,we evaluated the performance of different BAIs and multi-BAIs in determining CRC based on the mixed spectral reflectance of crop residue and soil,as well as the performance of the BAI–NDVI triangular space method and linear spectral un-mixing analysis to estimate CRC,FVC,and BS from mixed spectral reflectance measurements.Our results indicate that the proposed methods reduce the influence of moisture on broadband CRSI,provide accurate estimates of cropland CRC and fractional estimates of CRC,FVC,and BS,and may be applied in croplands where soil and crop residue moisture content varies greatly.(4)This study evaluates the use of broadband remote sensing,the triangular space method,and the random forest(RF)technique,for estimating FVC,CRC,and BS in cropland in which the SM-CRM content changes dramatically.We trained an RF model[designated the Crop Residue Index from Random Forest(CRRF)]that can magnify spectral features of crop residue and soil by using traditional broadband remote sensing data as input,and uses a moisture-resistant hyperspectral SINDRI as the target.The effects of moisture on crop residue and soil were minimized by using the broadband CRRF.Then,the NDVI and CRRF were used to create a two-dimensional scatter map with CRRF as the x-axis and NDVI as the y-axis(abbreviated as the CRRF–NDVI triangular space method).It was hypothesized that the values of vegetation SIs and crop residue SIs are determined by the proportions of crop residue,crop,and soil.Next,we used the CRRF–NDVI triangular space method(CRRF–NDVI)to estimate CRC,FVC,and BS.The proposed method was validated using laboratory-and field-based measurements with the Sentinel-2 MSI broadband remote sensing image.The results of this work indicate that our proposed CRRF–NDVI method(CRC:R~2=0.84,MAE=0.08,RMSE=0.10,n RMSE=10.1%;BS:R~2=0.82,MAE=0.09,RMSE=0.11,n RMSE=13.8%;FVC:R~2=0.93,MAE=0.05,RMSE=0.07,n RMSE=7.4%)can accurately estimate CRC,FVC,and BS,even in areas in which SM-CRM changes dramatically,which is higher than traditional DFI–NDVI(CRC:R~2=0.65,MAE=0.16,RMSE=0.19,n RMSE=21.0%;BS:R~2=0.74,MAE=0.14,RMSE=0.16,n RMSE=21.4%;FVC:R~2=0.92,MAE=0.06,RMSE=0.09,n RMSE=9.1%).We also use CRRF–NDVI method and MODIS image to estimate CRC,FVC,and BS of the North China Plain and lower reaches of the Yangtze River crop production regions in China.Since most broadband remote sensing images can be obtained free of charge,this new method has extensive applicability.