Spectral Characteristics Of Saline Soil During Microbial Remediation Processes Based On Partial Least Squares Regression

Soil salinization is mainly affected by natural factors and human factors,not only caused the waste of land resources,but also hindered the growth of crops and agricultural sustainable development.Timely and effective monitoring and control of saline soil is highly significant for the protection of land resources and ecological environment.Hyperspectral technique has the advantages of fast,non-destructive and large amount of information to obtain the information of terrestrial object by fine bands.The hyperspectral characteristics of saline soil are not only the basis for realizing the monitor for large scale of saline soil by remote sensing,but also the important content of realizing the non-destructive and rapid monitor of saline soil by means of networking.The dynamic changes of salinity factor content and spectral characteristics of saline soil is the basic and significant work of monitoring the saline soil and evaluating the effect of remediation on regional scale by remote sensing.To study the change of the salinity factor content and its spectral characteristics,the microbial remediation process of saline soil was concerned,and two types of saline soil were considered:the primary saline soil in Dongying and the secondary saline soil in Jiading,the salinity factor content and the associated spectral reflectance were measured at the same time,the effects of 34 pre-processing methods on spectral characteristics of saline soil during the remediation process were analysed and compared.The three methods including the extremums of correlation coefficients,the different ranges of correlation coefficients and interval partial least-squares regression were used to find the optimal sensitive bands of soil salinity facor content for different spectral data sets and the spectral response characteristics of soils under different transformations were well analysed.On this basis,PLSR was used to build relational models between soil salinity factor and spectral reflectance based on full bands(400-1650nm)and optimal bands during the remediation process,respectively.The results of this study were as follows:(1)The spectral curves of soil samples with different salinity factor contents in saline soil in Dongying and Jiading were similar in the overall morphology.In the range of 400-760 nm,the spectral reflectance of soil showed an obvious upward trend.In the range of 760-1650 nm,the spectral reflectance of soil tended to be gentle.The water absorption at 1411.90 nm can be clearly identified in the original spectral reflectance curve.The continuum removal and first derivative of the original spectral curves amplified and increased the spectral absorption characteristics.The spectral characteristic bands of different degrees of saline soil in Dongying were as follows:442.48nm,472.96nm,485.83nm,558.59nm,907.92nm,1059.47nm,1397.67nm,1411.90nm and 1419nm.The spectral characteristic bands of different degrees of saline soil in Jiading were as follows:432.49nm,485.83nm,647.40nm,689.74nm,739.23nm,844.62nm,1397.67nm,1411.90nm and 1419nm.(2)The spectral reflectance of saline soil has a good indication for the change of dominant salinity factors in saline soil.The soil salinity content of the saline soil in Dongying showed almost the same trend with the average reflectance of full bands during the 30-50 days of remediation.The content of nitrate ion in the saline soil in Jiading was consistent with the average reflectance of optimal bands with 775-899nm and 1025-1149nm during 7-74 days of remediation.With the decrease of salinity during the remediation time,the spectral reflectance of the two types of saline soil decreased.(3)The pre-processing transformations of spectra improved the correlation between the content of soil salinity factor and the spectral reflectance significantly,and the optimal sensitive bands were further focused.In this study,the best spectral transformation was derivative.For the saline soil in Dongying,the smoothed spectral data based on simple mathematical transformation with derivative could improve the correlation between the soil salinity content and the spectral reflectance.For the saline soil in Jiading,the original spectral data based on first derivative could improve the correlation between the content of nitrate ion and the spectral reflectance well.(4)The choice of optimal sensitive bands could reduce the number of independent variables,thus simplified the inversion models of the salinity factor content.This study used three methods to choose the optimal sensitive bands:extremums of correlation coefficients,different ranges of correlation coefficients and interval partial least-squares regression.For the saline soil in Dongying,the optimal sensitive bands of soil salinity content were selected in the first two methods,which mainly gathered in 947.11-949.31nm,1340.27nm,1394.11nm,1457.81-1461.31nm,1537.68-1551.39nm and 1602.32nm.For the saline soil of Jiading,the latter two methods were used to choose the optimal bands of salinity factor content under different spectral transformations.The optimal sensitive bands of soil nitrate ion content were 844.50 nm and 846.18 nm,and the soil salinity content were 1408.35nm,1411.9nm and 1415.45nm.(5)The salinity factor content of two types of saline soil in this study were retrieved by PLSR using the spectral reflectance of the optimal sensitive bands and full bands as the independent variables.R~2 and RMSE were used as the criteria for selecting the best models.After analysis and comparison,for the saline soil in Dongying,the full bands-based models were better than the optimal bands-based models based on PLSR on the whole,the prediction accuracy of SGSD was the best,and the corresponding R_p~2 and RMSEP of the predicted model were 0.673 and 1.256.For the saline soil of Jiading,the best PLSR inversion model for the soil nitrate ion content was established by using the optimal sensitive bands of 775-899nm and1025-1149nm extracted from first derivative of the original spectral reflectance,R_p~2 and RMSEP were 0.962 and 0.057.The best PLSR inversion model for the soil salinity content was established by using the optimal sensitive bands of 778-840nm and 1471-1533nm extracted from logarithm of 1/R of the original spectral reflectance,R_p~2 and RMSEP were 0.962 and0.868.In conclusion,the quantitative study of the microbial remediation process of saline soils is of great significance to the spectral characteristics of saline soils and the monitoring and detection of remediation process.The results of this study can provide the reference of methods and data for the further quantitative and quick analysis of salinity factors content in saline soil.