Remote Sensing Inversion Of Salinized Soil Organic Matter Based On UAV Multispectral

Saline soil is an important reserved land resource for arable land in China.Due to the high salt content in the soil,agricultural production and utilization are greatly restricted,which seriously restricts the sustainable development of the regional economy.Soil organic matter(SOM)is an important nutrient source for plant growth.It can adjust the soil structure and improve its physical and chemical properties.It is one of the important indicators for measuring soil quality.Therefore,the content of organic matter in the surface soil of salinized wheat fields can be rapidly monitored in order to provide technical support for the improvement of salinized soil,weight loss increment and promotion of carbon cycle research.In this paper,the saline soil of the core test area of"Bohai granary"in Wudi County,Binzhou City is taken as the study area.The soil organic matter and salinity data are obtained through field surveys,soil samples collection,and laboratory analysis.Combined with the UAV multispectral data obtained from the flight,explore the relationship between soil organic matter content and spectral reflectance data,remote sensing models of soil organic matter content supported by bare soil and vegetation spectral data are respectively established,the models are verified and compared and the optimal model is determined,then the inversion of soil organic matter in the research area is carried out based on the optimal model,and the accuracy analysis and comparison are carried out with interpolation results.It is believed that UAV multispectrum can provide fast and regional-scale accurate estimation of soil organic matter content in salted soil winter wheat at jointing stage.This study mainly completed the following contents and obtained the following results:(1)The results of UAV multispectral acquisition show that the maximum value of soil organic matter in the study area sample is 37.66 g/kg,the minimum value is 14.37 g/kg,the average value is 23.90 g/kg,the standard deviation is 5.21 g/kg,and the standard deviation is5.21 g/kg.Based on the analysis of the spectral reflection curves of the original 4 bands of the multi-spectrum of the UAV,the bands are combined to obtain a total of 11 spectral indices.Using Pearson correlation analysis,B1,B2,B4,NDVI,B1?B2andB2?B3were selected as sensitive indicators for model construction.In order to investigate the situation of wheat soil organic matter at jointing stage,models were established based on bare soil spectra and vegetation spectra of sampling points respectively,and vegetation models were established based on multi-spectral images of unmanned aerial vehicles.The images were filtered by multiple windows in different ways according to the planting characteristics of wheat ridges,and converted into wheat spectral information within a certain range.The median filter of gradient 5×5 reached the highest correlation of 0.66.(2)Compared with the modeling effect of bare soil and vegetation,the modeling effect based on bare soil is not ideal.Vegetation model is better than bare soil model.The support vector machine model established by the 5×5 median filtered spectrum is the best..Among the three models established,support vector machine model has the highest modeling accuracy,R~2 reaches 0.88,RMSE is the smallest,reaching 0.21,RPD is the highest,reaching 5.18,followed by multivariate linear model,R~2 is 0.79,RMSE is 2.35,RPD is 2.61,partial least square model At worst,R~2 is 0.75,RMSE is 2.05,and RPD is 2.57.The difference of the prediction results of the three different models is mainly reflected in the small-scale regional prediction.Multiple linear regression and partial least squares models have defects in dealing with specific values,resulting in too large or too small changes in soil organic matter content in the region.When estimating the content of soil organic matter by image spectrum,the support vector machine model has strong ability to deal with nonlinear mapping.It can reveal the main factors that affect the change of soil organic matter content from spectral data and find the local optimal value,so as to better predict the soil organic matter content at jointing stage of wheat in saline soil area,which is more suitable for the research area and has better prediction effect than other models.(3)Regional differences in soil organic matter content in the study area are obvious,showing an overall trend of more north and less south,which is basically consistent with the growth of wheat.Based on the support vector machine model and kriging interpolation,the total spatial change of soil organic matter can be obtained.The eastern and northern areas of the study area have high soil organic matter content,and the southern area is affected by high soil salt content.sensitive ecological environment and salt crust in soil surface layer destroy soil nutrients and the soil organic matter content is low.For the distribution of soil organic matter content obtained by interpolation and inversion,the inversion map can better reflect the changes in the level of regional organic matter content.The inversion does not depend on the data quality of the sample itself.The relationship can simulate the spatial variability of soil organic matter in the complex landscape of wheat at the jointing stage with high accuracy.The results obtained by the interpolation method are relatively smooth,and the difference between the maximum and minimum values is not obvious.The interpolation value provides a large-scale study.The scope of this study is too small,and there are not a large number of sampling points.The remote sensing inversion of organic matter in the soil spectrum of spectral images can solve a small range of problems and has high accuracy.