Extraction And Analysis Of Soil Salinization Information In Coastal Area Based On Remote Sensing Spectral And Spatial Characteristics

Soil salinization is one of the main causes of agricultural yield decline and ecological environment deterioration.The Yellow River Delta is a typical coastal saline soil distribution area in China.Soil salinization is an important factor restricting the agricultural and economic development in this area.It is of great significance to extract timely and accurate information of littoral saline soil,analyze and master its dynamic changes,and monitor and control the dynamic saline soil.Remote sensing technology provides a simple and efficient method for soil salinization monitoring.At present,it is mainly based on the spectral characteristics of remote sensing image,and the information of saline soil is extracted based on a single pixel.In this study,Kenli District of the Yellow River Delta was taken as the research area.Based on the object-oriented method,the spectral features and texture features with high correlation with soil salt content were selected based on the ground measured data and Landsat-8 and Sentinel-2 satellite remote sensing data.The shape,elevation,distance from the Bohai Sea and other spatial features were combined with different feature combinations.Different classifiers were used to classify the saline soil in the study area.Finally,the feature combination scheme and classifier with the highest accuracy were selected to realize the extraction and dynamic analysis of the saline soil in the study area.The main research contents and conclusions are as follows:（1）The blue light,green light,red light,near infrared and two short wave infrared bands of remote sensing image are significantly correlated with the measured sample salt content.Among the different salt indices and vegetation indices constructed by using sensitive bands,the nine spectral indices SI、S₃、NDSI、NDVI、ENDVI、ERVI、EDVI、EEVI and Int were highly correlated with the salt content of sample points,and were selected as spectral features to participate in the classification of saline soil in the study area.Each texture quantization parameter extracted by using the gray level co-occurrence matrix method under 3×3 window has a significant correlation with the salt content of the sample,and the four texture quantization parameters of entropy,contrast,anisotropy and correlation do not have a strong autocorrelation,so it is preferred to be used as the texture feature to extract the salt soil information in the study area.The correlation coefficients of elevation,distance from Bohai Sea and soil salt content in the study area were-0.458 and-0.522,respectively.The correlation models were y=149.18^-1.921,²=0.6347,y=21.847^-0.642,²=0.634,respectively.Both elevation and distance from Bohai Sea had different effects on soil salinity in the study area,and the classification of saline soil in the study area was optimized.（2）Compared with the single spectral feature,the addition of texture and shape feature can effectively improve the classification accuracy of saline soil information in the study area,while the addition of topography and distance from Bohai Sea feature has little influence on the classification accuracy of saline soil information in the study area.The combination of spectral,texture and shape features is the most accurate method for extracting saline soil information in the study area.In terms of the classification effects of different classifiers,the random forest classification method has the best performance in extracting mild and moderate saline soil information after adding spatial feature information,and the support vector machine classification method has the highest producer accuracy in extracting severe and saline soil information.The classification accuracy of saline soil in the study area is the highest by random forest extraction based on object-oriented spatial feature information of spectrum,texture and shape.（3）Salinite is the most widely distributed in the study area,mainly distributed in the eastern coastal area of the study area.The second is moderately saline soil,mainly distributed in the southwest and central part of the study area,especially along the Yellow River,residential areas and near water bodies.The heavy saline soil is mainly distributed between the middle saline soil and the moderate saline soil in the study area.The area of lightly saline soil is the smallest,and most of them are scattered in the middle and southwest of the study area.From the perspective of time change,the area covered by mild and severe saline soil in the study area showed a decreasing trend in recent years,while the area covered by moderate saline soil showed an increasing trend,and the salinized soil area showed a trend of increasing first and then decreasing.From the perspective of spatial variation,most soil salinization levels in the study area did not change significantly in recent years,and soil salt content in some areas of the northern part of the Yellow River increased,while soil salt content in residential areas,near water bodies,the southwest of the study area and along the Yellow River decreased.This study made full use of the spectral,texture,shape,topography and other spatial characteristic information contained in remote sensing images,and realized the classification of saline soil of different degrees in the study area based on object-oriented machine learning method,providing a more accurate and fast method for the extraction of saline soil information near the sea.