Hyperspectral Estimation For Soil Organic Carbon Stock And Its Influencing Factors In Winter Wheat Field Of Loess Plateau

Increasing and promoting soil carbon sequestration was identified as one of the long-term effective approaches to mitigating the rise of CO2 concentration and national strategy plan for committing to peak carbon dioxide emissions before 2030 and achieving carbon neutrality before 2060.Therefore,the accurate estimation of farmland soil organic carbon stock(SSOC)is the foundation of researching the soil carbon circulation,revealing the potential of soil carbon sequestration and assessing the soil quality.However,the accurately estimation of SSOC has the practical problem of many research methods with inconsistent results,and there is an urgent need for the new technology and method.Visible and infrared spectroscopy(VNIRS)technology possesses the advantages and features of rapidity,non-destruction and high efficiency and has potential practical value in estimating SSOC.Hence,combing with the methods of Pearson correlation analysis,spectra preprocessing techniques and multivariate statistical algorithm,we chose the soil of winter wheat field in south Shanxi of Loess Plateau as research object and carried out the study for estimating SSOC and soil properties associated with SSOC.The main conclusions in this study were as follows:(1)The reflectance curves of SSOC and soil properties associated with SSOC in four levels were differed in wavebands of 400-2450 nm,and the difference mainly came from the values of spectral reflectance and there was no obvious difference with the shape and trend of all spectral curves.The relationship results between SSOC and soil properties with preprocessed spectra showed that,the correlation between SSOC and spectra was the highest(r=0.59),which mainly showed indirect response mechanism.While,for all soil properties associated with SSOC,soil organic carbon(SOC)had the highest correlation with spectra(r=0.57),which showed direct response mechanism.The number and locations of the sensitive wavebands differed for SSOC and soil properties associated with SSOC,but there were similar spectral characteristics.The sensitive wavebands of SSOC,soil physical properties,chemical properties and enzymatic activity properties were located in the regions of 400-419,550-641,875-975,1002-1094,2121-2256 nm and positions near 1418 and 1873 nm,regions of 400-403,497-602,974-1191,1791-1957,2207-2215,2421-2450 nm;regions of 400-438,525-645,823-1197,1231-1315,1700-1993,2166-2343;and regions of 400-557,1250-1422,1759-1953,2218-2389 nm,respectively.(2)Based on the NC preprocessed spectra and important bands,the successive projections algorithmmultiple linear regression(SPA-MLR)models for all soil properties factors were established.Also,their partial least squares regression(PLSR)and support vector regression(SVR)models were built.Considering the performance of calibration and validation models,among all models for physical properties,the soil bulk density(BD)predictive model based on PLSR method performed best(Rc~2=0.823,RMSEc=0.066,RPIQc=2.278 for calibrated model and Rv~2=0.640,RMSEv=0.043,RPIQv=1.715 for validated model,respectively).Among all chemical properties models,the SPA-MLR model for soil organic carbon(SOC)achieved the best estimation(Rc~2=0.650,RMSEc=0.829,RPIQc=2.124 for calibrated model and Rv~2=0.704,RMSEv=0.740,RPIQv=2.414 for validated model,respectively).In addition,the best models of soil protease activity(SPA)and soil sucrase activity(SSC)were simultaneously obtained by PLSR with the SPA model performed better(Rc~2=0.951,RMSEc=0.214,RPIQc=5.644 for calibrated model and Rv~2=0.305,RMSEv=0.474,RPIQv=1.037 for validated model,respectively)than SSC.(3)The soil pedotransfer functions(PTF)quantitative estimation model between SSOC and important soil properties,the spectral estimation model based on VNIRS technology and the indirect spectral estimation model based on VNIRS and PTF coupling technology were built and the results showed that,the performance of PTF model behaved best with Rc~2=0.987,RMSEc=0.041,RPIQc=10.496 and Rv~2=0.842,RMSEv=0.132,RPIQv=3.663.For all VNIRS spectral estimation models,the model based on noise correction spectra and SVR modeling approach had the optimal predictive ability with Rc~2=0.934,RMSEc=0.094,RPIQc=4.669 and Rv~2=0.787,RMSEv=0.149,RPIQv=2.851.For all indirect spectral models,the SSOC model based on the optimal PLSR model of important soil properties and PTF model obtained the best performance of Rc2=0.947,RMSEc=0.083,RPIQc=5.598 and Rv~2=0.613,RMSEv=0.190,RPIQv=2.013,respectively.Comprehensively considering the applicability,efficiency,mechanism explanation and performance improvement ability of the estimation model,we came to the conclusion that the SSOC model coupled VNIRS and PTF technology had the great practical advantages,which will further broaden the application potential of VNIRS in quantitative monitoring SSOC,especially in large scale region.(4)Comparing with raw spectra(R),first derivative(FD),multiplicative scatter correction(MSC)and noise correction(NC)preprocessing methods could increase the relationship between spectra with SSOC and soil properties associated with SSOC and caused the differences to the number and distribution of characteristic bands for SSOC and its soil properties factors,and improve their model performance.Among them,the NC preprocessing method was regarded as the optimal transformation.SPA-MLR,PLSR and SVR multivariate statistical methods produced various effects on models performance of SSOC and its soil properties factors,and PLSR was considered as the best multivariate method for physical properties and enzymatic activity properties,while SVR was the optimal modeling approach to SSOC and its chemical properties factors.In conclusion,for soil in wheat field in south Shanxi of Loess Plateau,the hyperspectral response mechanism of SSOC and majority of soil properties factors was revealed,the accurate assessment of SSOC and majority of soil properties factors was also realized,and the optimal combined methods system for monitoring SSOC and its soil influencing factors by VNIRS was finally established in this study.It will provide technical references for the dynamic estimation and influencing mechanism of SSOC and establish the theoretical foundation for serving local farmers to formulate reasonable farmland carbon sequestration management measures in the winter wheat field in south Shanxi of Loess Plateau.