Effects Of Different Spectral Parameters On Inversion Of Soil Organic Matter Content By Near-earth Hyperspectral Reflectance

Soil organic matter(SOM)is an important part of cultivated soil and an important indicator for evaluating the fertility status of cultivated land.Real-time understanding of the spatial distribution of SOM is the key to manage cultivated land.High-precision digital soil mapping of soil organic matter helps to understand the spatial distribution and variation of its content.A large number of studies have shown that the use of near-earth Visible-NIR spectral data and satellite spectral data can effectively predict SOM content.Therefore,this paper took the western part of Honghu city as the research area,and used the PLS model and the SVM model to explore the influence of spectral data under different band ranges and different spectral resolutions on the accuracy of the model for predicting SOM content.And simulate the satellite spectrum combined with the spectral response functions of different sensors.It was expected to provide a theoretical basis for obtaining high-precision digital soil maps of soil organic matter using indoor measured Visible-NIR spectral data in the future,and also provided technical support for future satellite sensor channel settings.A pilot experiment was conducted to predict soil organic matter content using satellite image data.The conclusions of this study were as follows:(1)Combined with the spectral curve after pretreatment and the correlation analysis results between soil organic matter content and spectral reflectance,it showed that the spectral data in the visible light band had a higher correlation with the soil organic matter content,and the near-infrared long-wave band had a higher correlation.The spectral data in the range had obvious spectral response absorption characteristics.(2)The best preprocessing method for the original spectral data measured by the ground object spectrometer ASD was the combined transformation of Savitzky-Golay convolution smoothing(SG)and SG+Multiple Scattering Correction(MSC).(3)The raw spectral data were divided into five types,which were visible light band(VIS),near-infrared band(NIRS),near-infrared short-wave band(NIRS1),nearinfrared long-wave band(NIRS2)and visible-near-infrared short-wave band(VISNIRS1),according to the band position.The inversion prediction of SOM content was carried out using the spectral data of different wavelength bands.The results showed that the model including the near-infrared long-wave band spectral data had better prediction ability for soil organic matter content,and the model including the visible light band spectral data had relatively high prediction ability.Therefore,it was possible to reduce the number of spectral bands,reduce redundancy and improve computational efficiency by screening out spectral bands with poor correlation and insignificant absorption characteristics.(4)Using the equal-spacing sampling method to resample the near-earth VNIR spectral data with a spectral resolution of 1nm.It was found that when the spectral resolution was appropriately reduced,the number of spectral bands could be effectively reduced.And it sould solve the problems of data redundancy and low computing efficiency in the Visible-NIR spectral data,and did not affect the predictive power of the model.The results showed that appropriately reducing the spectral resolution of the unpreprocessed spectral data could ensure the predictive ability of the model and effectively reduce the number of bands,when the spectral resolution was low,it would affect the predictive ability of the model.The spectral resolution was preprocessing in SG and SG+MSC methods for 1nm spectral data could effectively improve the accuracy of the inversion model.Preprocessing the resampled spectral data did not necessarily improve the model accuracy,and even the model accuracy was lower than that of the original spectral data accuracy,which might be a result of discontinuities in spectral data due to equidistant resampling.(5)Combined with the spectral response functions of different satellite sensors,the near-earth Visible-NIR spectral data could be converted into satellite simulated spectral data,and the simulated spectral data could be used to predict soil organic matter content.The model had certain predictive power.Depending on the sensor settings,the number of bands for the simulated spectral data was different.The hyperspectral sensors had more bands and narrower bands,while the multispectral sensors had fewer bands.The results showed that the accuracy of the SOM content prediction model established by the simulated spectral data of the hyperspectral sensor is higher than that of the multispectral sensor.And the models with more sensor bands generally had higher prediction ability.In addition to the number of sensor bands,the location of the bands also had a certain influence on the prediction ability of the model.The precision of the model established by satellite sensors containing more near-infrared long-wave bands was generally higher.At the same time,because different satellites were equipped with different sensors,the spatial resolution of the sensors was different.And it would also affect the accuracy of the prediction model of soil organic matter content.The higher the accuracy,the better the prediction ability.