Application Research Of Fractional Derivative On The Remote Sensing Monitoring Of Soil Salinization

As a kind of important natural resources,the soil provides food and fiber for human beings,and maintains the ecological system on the earth.Meanwhile,the soil is the basic condition for human society and all social activities,and takes a significant role of a carrier.However,soil salinization is one of the most important ecological environment problems in arid and semi-arid area.Caused by soil salinization,soil hardening,decline of soil fertility,destruction of soil pH balance and some other serious consequences might severely restrict the agricultural development and affect the overall situation of sustainable development strategy of China.With the advantages of large scale,high efficiency and low cost,the remote sensing technology becomes a good complement to conventional methods,and provides a new way for the quantitative monitoring of soil salinization.Based on the previous researches and applications on monitoring the phenomenon of soil salinization by remote sensing technology,the Ebinur Lake basin with severe salinization,situated in the south-west border of Junggar basin in the Xinjiang Uygur Autonomous Region was selected as the study area and typical saline soil was as the research object.Combined with soil sample collection,indoor spectral reflectance measurement,physical and chemical analysis experiments,the physical and chemical characteristics of saline soil in the study area were clarified,and the feasibility of the application of fractional derivative in the monitoring of soil salinization by remote sensing technology was discussed from spectral and spatial perspectives.In the spectral dimension,the spectral reflectance data of the saline soil measured under the controlled light source condition was treated by Grünwald-Letnikov fractional derivative equation?order interval 0.1?.The effects of fractional derivative on spectral reflectance data of saline soil were explored from the points of correlation coefficient,coefficient of variation,spectral information divergence and some other indexes.And then,on the basis of these results,the quantitative estimation model of soil salinity was established by partial least squares regression and ensemble modeling theory.In the spatial dimension,the fractional derivative operator with direction was constructed,and then,multi-spectral image acquired by operational land imager of Landsat 8 satellite was filtered by the fractional derivative operator.Compared with Soble and Laplacian operator based on the first and second derivative,according to the image quality evaluation indexes,such as mean gradient,information entropy,peak signal to noise ratio and structural similarity index,the contributions of fractional derivative operator in image enhancement were studied.After choosing the suitable salt index for spatial visualization of soil salinization in the Ebinur lake natural reserve,the effects of fractional derivative operator on extraction of the salt index were discussed.The main conclusions are as follows:?1?The salt content of topsoil in the Ebinur Lake basin showed strong variability with mean value of 33.838 g/kg,larger than the solonchak grading standard of 20 g/kg.The soil samples in the study area are mainly alkaline and strong alkaline.Positive ion content Na⁺> Ca²⁺>K⁺>Mg²⁺,and negative ion content SO₄^2->Cl^->HCO₃^-.According to the results of correlation analysis,NaCl and CaSO₄ are the major kinds of salt,and Cl^-and SO₄^2-are mostly salinized types.?2?In the study area,on average,the maximum content is the sand,while the minimum is the clay in the soil samples.According to the international standard of soil texture classification,the soil could be divided into silty loam,loam,sandy loam,sand and loamy sand.The fractal dimension had good nonlinear relationships with clay,silt and sand.The estimation model of fractal dimension based on soil mechanical composition by using stepwise multiple regression method is 2.844 0.002 0.006 D clay sandy?28??10?x-x with the coefficient of determination of 0.720.According to the descriptive statistical analysis,the sand content decreased,clay,silt and the fractal dimension increased along with the deepening of the degree of salinization.?3?Spectral curves of soils with different degrees of salinization followed similar shapes,and could be distinguished obviously in the ranges of 400⁵00 nm?550⁷00 nm?1000¹850 nm and 2250²400 nm.Near the water absorption bands of 1400 nm and 1900 nm,the absorption valleys became deeper with the increase of salt content,but it was opposite near the absorption band of 2340 nm.In the range of 400¹875 nm,the correlation coefficient between reflectance and soil salt content passed the significant test at the 0.01 level,while there was no band passed the test in the spectral region of 1876²400 nm.?4?For five mathematical forms of spectral reflectance,the numbers of bands whose correlation coefficient with the soil salt content passed the significant test: 0th derivative>1st derivative>2nd derivative.When the order was extended to fractional,the correlation coefficient curves showed a gradual trend,and some details emerged,with the order increasing,the number of bands followed decreasing firstly,then increasing and finally decreasing trend.Also,the derivative values were gradually close to 0 and the numeric range of data became smaller with the order increment.Combined with coefficient of variation,the dispersion degree of data was generally greater,while the numeric range became smaller.The spectral information divergence displayed a non-linear increasing trend,and the magnitude was larger from 0th to 1st order than from 1st to 2nd order.These results indicated that fractional derivative could detail the variation tendencies of the correlation coefficient,range,coefficient of variation,spectral information divergence,increase the degree of dispersion of the spectral data and enhance the difference between the spectra to a certain extent.?5?105 soil salinity estimation models were built by PLSR method using the 0² order derivative data of five mathematical forms of spectral reflectance,and among these 105 models,the model based on the 1.9 order derivative of root mean square transform of spectral reflectance was optimal with RMSE_C=26.206 g/kg,R²C=0.787,RMSEP=24.955 g/kg,R²P=0.819,RPD=2.352.The weight was given to each sub-model according to the value of RPD,and the simple average method was improved,after combining 21 sub-models with RPD?2,the combination model with RMSE_C=26.291 g/kg,R²C=0.786,RMSEP=24.332 g/kg,R²P=0.828,RPD=2.413 was established,quantitative estimation precision and the prediction ability of soil salinity had been further improved.?6?For true color image and standard false color image,when the order of fractional derivative operator was greater than 0.8,average PSNR of R,G,B channels was less than 20 dB,the image distortion was unacceptable.Compared with the original images,the images filtered by fractional derivative operator lost information,destroyed the structural similarity to some content,however,the PSNR and mean gradient reached balance between 0.6-order and 0.7-order,and image enhancement effect achieved the best.In addition,when the order was greater than 1,fractional derivative operator had dual response to some edges in the image and generated double pixel boundaries.?7?Compared with 12 kinds of commonly used salt indexes extracted from remote sensing image,correlation coefficient between salt index 5SI?28??B ?R?/ G and salt content was 0.3532 and passed the significant test at the 0.01 level,and SI₅ was more suitable for spatial visualization of salinization information in the Ebinur lake natural reserve.After extracting SI₅ from images filtered by 0.1⁰.8-order operators,the correlation coefficients showed first increasing then decreasing trend,reached the maximum value of 0.3583 at 0.4-order,and significantly decreased at 0.8-order,only passed the significant test at the 0.05 level.This indicated that when the fractional derivative operator was used to enhance images,the order in a certain range was conducive to the extraction of soil salinization information,and beyond this range,it had detrimental effects on salinization information extraction.Moreover,with the increment of the order,some details of the salt index spatial distribution emerged,which was closely related to the ability of fractional derivative operator in the image enhancement.In this study,fractional derivative is introduced to the remote sensing data pretreatment,and provides more supplementary methods for extracting the feature information from remote sensing data,and also offers references about the applications on salinization information extraction,salinization monitoring and thematic mapping in large scales by the technologies of airborne and spaceborne remote sensing in order to meet the further requirements for precision agriculture in the future.