Estimating The Soil Organic Carbon Content Based On Multiangular Reflectance Factor

Soil Organic Carbon is an important index reflecting soil quality and fertility.Conventional laboratory chemical analysis to determine Soil Organic Carbon Content(SOCC),despite its high accuracy,has a long analysis cycle,high cost,and some pollution to the environment,so it is difficult to popularize and use on a large scale.Remote sensing technology can accurately perceive the change of soil composition content,and provide accurate,rapid and non-destructive method to obtain SOCC information timely and effectively.Most of the previous studies on remote sensing estimation SOCC were based on the zenith direction or spectral data from a single observation Angle.However,in actual measurement,due to the influence of remote sensing detection geometry,the reflection information input to the sensor is obtained from different detection angles when the remote sensing platform obtains the reflection information.This indicates that remote sensing multiangular detection not only obtains the spectral dimension information of soil,but also obtains the angular dimension information.Therefore,it is necessary to explore the possibility of implementing SOCC estimation model based on multi-angular reflection measurement data,make full use of spectral dimension and angular dimension information,and effectively improve the accuracy of SOCC estimation.In this study,164 soil surface samples collected in Northeast China were measured by multi-angular reflection spectrum in laboratory and field,and their SOCC was determined and analyzed.After different pretreatments for the original reflection spectrum R(SG smoothing,continuum removal: CR,multiple scattering correction :MSC,first derivative :FD and second derivative: SD),Partial least squares(PLSR),support vector(SVR)and random forest(RFR)regression models were established for multi-angle data and single-angle data based on six spectra(R,SG,MSC,CR,FD and SD).The ability of several regression models to estimate SOCC from different detection angles was evaluated.The results show that the SOCC estimation model established based on multi-angular data of six spectra(R,SG,MSC,CR,FD and SD)can achieve stable estimation of SOCC at different detection angles,and the accuracy of the estimation model established based on RFR is the highest.RMSE were0.24%,0.24%,0.23%,0.27%,0.29% and 0.40%,respectively.This is because the machine learning model(RFR),as a data-driven model,provides more abundant information about soil optical properties from more observation perspectives,so that more information describing soil optical properties in a more comprehensive way can be input into the model,which can improve the accuracy of SOCC estimation model.Therefore,this study believes that the SOCC estimation model established on the basis of multi-angular spectral data can realize the high-precision estimation of SOCC at different observation angles.In contrast,the SOCC estimation model based on the single-angular data of six spectra cannot perform better estimation in other angles,especially the SOCC estimation model based on the forward Angle data has a low estimation accuracy in the backward Angle data.In addition,the Spiking method and EW method were used to verify the portability of the SOCC estimation model combining multi-angle datasets in other datasets(LUCAS dataset601 samples and GSSL dataset 484 samples).For the estimation accuracy of different data sets,the modeling regression method of the best estimation model is inconsistent.For the estimation of LUCAS data set SOCC,the multi-angular data based on six kinds of spectra,R,SG,MSC,CR,FD and SD,showed the highest accuracy of the multi-angular data estimation model established by using RFR method.The verified RMSE were 0.89%,0.87%,0.91%,0.89%,0.92% and 0.98%,respectively.For the GSSL data set SOCC estimation,the multiangular data estimation model based on four kinds of spectrum,R,SG,MSC and FD,the model established by SVR regression method has the highest estimation accuracy,and the verified RMSE are 0.62%,0.63%,0.69% and 0.68%,respectively.For the multi-angular data estimation model based on CR and SD spectra,the model established by RFR regression method has the highest estimation accuracy,and the verified RMSE are 0.62% and 0.72%,respectively.For the three regression methods,the SOCC estimation model established by combining multi-angular data can achieve high-precision estimation of SOCC of other datasets,indicating that the SOCC estimation model established with multi-angular data is transferable in heterogeneous areas.In conclusion,the combination of multi-angular data with machine learning method(RFR)can significantly improve the accuracy of SOCC estimation due to the information of spectral dimension and angle dimension.The multi-angular data estimation model can also be applied to estimate SOCC in other regions.This provides a basis for nondestructive estimation of SOCC based on multi-angular reflection characteristics of soil,and also provides a new possibility for accurate estimation of SOCC in heterogeneous soil environments.