| Soil is of great ecological and economic values as a kind of natural resources.Soil not only provides nutrients for the growth of crops,provides food security for human beings,but also plays an important role in urban construction,human health and other hot issues.At the same time,soil is one of the most active organic carbon pools on the surface of the earth,and it is an important component of terrestrial ecosystem.Soil organic carbon density(SOCD)is an important indicator of soil carbon storage.How to achieve the rapid and efficient access to soil organic carbon density information is not only related to agricultural development,environmental changes but also matters ecological balance.The convenient and effective acquisition of time and space information is an important direction for future geographical conditions monitoring and land resource investigation.Therefore,visible-near-infrared spectroscopy techniques is becoming a very promising tool to estimate the content of soil organic carbon density for its advantages such as speediness,efficiency,accuracy and environment friendly.It has the broadest application prospect in the quantitative acquisition of soil properties,soil resource investigation,soil digital mapping and precision agriculture development.This paper takes Honghu area of Jianghan plain which is known for its abundant as the research area.A total of 232 soil samples were collected.Then visible-near-infrared spectroscopy data and organic carbon density content were acquired through laboratory analysis.Spatial distribution of soil organic carbon density and the spectral response characteristics are analyzed.And the paper reveals the space correlation of soil spectral.About the SOCD of Honghu area,the content is low.The SOCD mean value is 4.62 kg/m2 and the variation range is between' 1.32 and 8.16 kg/m2.With the increase of SOCD,the spectral reflectivity is decreasing,and they are negatively correlated.After the original spectral data is pretreated,the correlation is enhanced.With the increase of SOCD,the difference of the average spectral curve between irrigated and paddy fields increased,and the effect of land use type on soil reflectance was enhanced.The spectral contribution of the first four principal components is more than 80%,which can represent the spectral information to a certain extent.There was a positive correlation between the principal component variables and SOCD,and the correlation between the first principal component(PCI)was the highest and the correlation with the fourth principal component(PC4)was the lowest.The results of trend analysis show that there are first or second order trends in SOCD and spectral principal components in east-west and south-north directions.The optimal fitting model of the second principal component of soil spectrum and SOCD is the exponential model.The optimal fitting model of the first,third and fourth principal components of soil spectrum is spherical model.In the study area,the ratio of nugget and sill of SOCD and PC1,PC2,PC3 and PC4 are 59.32%,68.45%,26.34%,57.37%and 71.44%respectively.The range of change is 25%-75%.The spatial correlation of variables is at a moderate level.The anisotropy of the principal components of soil spectrum and SOCD is obvious and the isotropic characteristics are not significant in four directions from the semi-variogram curve of 0°,45°,90° and 135 °.The semi-variogram curves have the same nugget and partial sill,have the performance of geometric anisotropy.Moran 's I of SOCD and PC1,PC2,PC3 and PC4 are0.44,0.13,0.30,0.26 and 0.23 respectively.The p values of these five cases are less than 0.05 and the z values are more than 1.96,which indicates that there is significant spatial autocorrelation of SOCD and the principal components of soil spectrum,which is consistent with the results obtained by semi-variogram models.In this study,we compared the accuracy of SOCD spatial prediction results of partial least squares regression,ordinary kriging,co-kiging and regression kriging.The prediction accuracy of the ordinary kriging model is poor,the validation dataset coefficient R2-Pred is only 0.004,the prediction accuracy of the co-kriging model is not satisfactory,R2-Pred is 0.007,the partial least squares regression model and the regression kriging model have higher prediction accuracy,R2-Pred are 0.605 and 0.617 respectively.The modeling accuracy,prediction accuracy and ratio of performance to deviation of the regression kriging model are the highest,which is the optimal model of soil organic carbon density prediction.It is shown that considering the spectral reflectance and the spatial properties of soil properties can improve the prediction accuracy,the spatial structure of the regression model residuals can reflect the spatial variability of soil properties more truly.In the partial least squares regression model,the combination of Savitzky-Golay smoothing,first order derivation and standard normal variate is the best pretreatment method.The ratio of performance to deviation is 1.523.The prediction accuracy is second only to the regression kriging model.The spatial distribution of the predicted results obtained by ordinary kriging,co-kriging and regression kriging show that the overall trend in the study area is consistent,the stripes formed in the northwest-southeast connection are high and the central and southern regions are low distributed. |
PDF Full Text Request