Prediciton Of Soil Organic Carbon In The Lakeside Area Using Vis-NIR Spectroscopy

China is a great agricultural country.Despite the fact that China only have about9%of the world’s cultivated land,it has to feed 18%of the world’s total population.China produce large amount of crop every year,but it still import huge amout of food.To be more,China’s existing cultivated land is facing many problems such as quality degradation,overload of utilization,improper use of chemical fertilizers,and heavy metal pollution.The China government attaches great importance to the argricultural development and have anouncced a national basic policy that we must make best use of all land and protect it from problems.In land,soil is the core to support arlgiculture,environment and ecology.Monitoring soil properties can help farmers get the state of soil in advance and take necessary measurements in time,which is precision agriculture.Among many soil properties,soil organic carbon（SOC）could significantly affect other ones and plays an important role in global carbon circle and food production.Tradditional methods of measuring SOC is expensive and time consuming.Vis-NIR spectroscopy is a promising method with the advantage of quickness and low cost.In this paper,vis-NIR spectroscopy is applied to estimate the SOC content in lakeside area,namely Honghu City.We reviewed the reasearch of soil spectroscopy in the last 30 years and coucluded that three main problems required further study.Thus,firstly,we try to combine spectral transfomations with sample seletion methods to improve the estimation of SOC.Then,in situ spectroscopy is compared with lab spectroscopy,and their influence on SOC modeling is also compared.Finally,we introduce the Chinese vis-NIR spectroscopy into the lakeside soil set and then build SOC estimation models.In applying Chinese vis-NIR spectroscopy,spiking and extra weighting is deeply study,espeical about the size of spiking samples and the copy order of extra weighting.Accord to our experiments,several conclusions are drawn as follows:（1）Spectral transformations combined with sample selection algorithms could improve the estimation of SOC using vis-NIR spectroscopy.As far as we know,there is no research applying transformations to sample selection methods to improve the accracy of SOC models.In this paper,first spectra is transformed and then is the input of sample selection algorithm.In such a way,representative samples could be selected and a more robust model could be constructed.We build different sizes of calibration sets and results show that the the accuracy of model increases with the size of calibration set and sometimes the model accuracy remains stable because the new selected sample are these ones similar to previous selected samples instead of representative samples.The proportion of positive influence of FD,Lg（1/R）,MSC and SNV on sample selection are 64%,78%,72%and 75%,respectively,and the degree of average improved RPD is 0.08.0.13,0.19,and 0.13,respectively.So in most case,spectral transformation helps sample selection method to select representative samples and in such a way it improves the SOC estimation models.If a transfomrtion has the highest proportion of positive influence,its degree of positive influence is not necessary to be the highest.The reason that transformation could improve SOC estimation is that transformation changes the spectral distance between samples,and the change in distance make selected samples be distributed more evenly,which resulting in a better SOC estimation model.Spectral transformation is originally designed for multivariate regression and is applied to sample selection algorithm in this study.Results show that the degree of transformation’s influence on multivariate regression is that on sample selection algorithm,and there is no correlationship between them.（2）Comparation of characteristics of in situ spectra and lab spectra and their SOC estimation models.The field spectra is collected in farm and the performance of SOC estimation model using the raw ourdoor spectra is poor with R²_cv of 0.04 and RMSE_cvof 18.82 g/kg.The reflectance of the raw is higher than 1 for some wavelengths,so these abonormal wavelengths are removed.and after taking samples to lab we measure the spectra of soil samples.Afer air-dried and ground,the performance of SOC model is better with R²_cv of 0.43and RMSE_cv of 4.89 g/kg.After removing some abnormal smaples,the R²_cv is 0.54 and RMSE_cv is 4.67 g/kg.Thus,for situ spectra,removing abnormal wavelenths and abnormal samples is necessary.For lab spectra before being air-dried and ground,the reflectance is lower than 0.2 and the performance of SOC model is good with R²_cvof 0.51 and RMSE_cv of 4.62 g/kg.After being air dried and ground,the accuracy of SOC model increases greatly with with R²_cvof 0.80 and RMSE_cv of 2.90 g/kg.The reason behind the improvement is that ari-dry and ground eliminates the efffects of soil moisture and particle size.While comparing situ and lab spectroscopy,it is found that field spectra is greatly affected by enviromental factors and the quality of spectra is poor,resulting in poor performance of SOC models.So the SOC models of field spectra should be improved.In addition,the reflectance of situ spectra is lower than that of lab spectra because of soil moisture.The optimal pretreatment for field spectra is SG+SNV with R²_cvof 0.554 and RMSE_cv of 4.622 g/kg,and the optimal pretreatment for lab spectra is SG+MC with R²_cvof 0.831 and RMSE_cvof 2.650 g/kg.For situ spectra,the absoluteΔRMSE_cv is greater than 0.5 g/kg while for lab spectra,the absoluteΔRMSE_cv is lower than 0.5 g/kg.Thus,situ spectra is more sensitive to spectral pretreatment.In addition,spectral prtreatment has the similar function on situ spectra and lab spectra.（3）Application of Chinese soil vis-NIR library to lakeside area to estimate soil organic carbon.When using the whole national soil library to estimate SOC of samples in lakeside area,RMSEP is 5.921g/kg,R_p² is 0.609,and RPD is less than 1,which is a lttile poor.The reason behind that is that the nation al soil library contains diverse samples and is very different from sample in lakeside area.When considering soil type,paddy soil is the most suitable for SOC estimation in lakeside area with RMSEP of4.543g/kg,R_p² of 0.61,and RPD is less than 1.5.We try to mix the national soil library with some samples from lakeside area,namely,spiking.When the number of spiking samples increases from 10 to 50,the R_p² increases from 0.655 to 0.753,the RMSEP decreases from 4.373g/kg to 3.68g/kg,and the RPD increases from 1.438 to 1.709.Thus spiking has significantly improved the accuracy of applying the national spectral library to lakeside area beacause spiking increases the similarity between the national spectral library and the samples in lakeside area and the national spectral library has gradually adapted to the estimation of SOC in the lake area.Based on spiking,we try to copy the lakeside samples in the mixed sample set,namely,extra-weighting（EW）.When there are 10 spiking samples and copying times of the spiking smples is 128,EW increases the RMSEP from 4.251 g/kg to 3.18 g/kg.While using 50 spiking samples,EW increases the RMSEP to 2.907 g/kg and R_p² was 0.802.In the case that spiking samples are 50 and EW is 64,the RMSEP is 2.583 g/kg and R_p² is 0.864,which performs better than the local lakeside model of esimating SOC.When the number of spiking samples is constant and the copy times of EW increases,the model accuracy will increase gradually and the improvement is limited.When copy times of EW the number of spiking samples is constant and the number of spiking samples increases,the model accuracy will also increase gradually.This paper has successfully applied the nation spectral library to the estimation of SOC in lakeside ares and the model accuracy is beyong expected.Our application is a very important reference for the national spectral library and proving methods for the used of national soil library.In this paper,we study the use of vis-NIR spectroscopy to esitmate SOC in lakeside areas.We firstly improve the method of using vis-NIR spectroscopy and improvement is significant.We then compare situ and lab spectra and their influence on SOC model.Finally we apply national soil spectral library to esimate SOC in lakeside area.Our study is applied to Honghu City and more case studies are needed.Our methods could be used in other areas.Ou study of vis-NIR spectrpscopy technology provides a case of rapidlly monitoring soil properties which can be used for agricultural manager to make decisions.