A Study On Model Of Soil Organic Matter In Arid Areas Of South Xinjiang Based On Hyperspectral Data

Soil organic matter is the general term of carbonaceous organic matters in soil,which can provide nutrients for plant growth.The content of soil organic matters is an important index to measure soil fertility.The spatial distribution of soil organic matters is an important basis of grain yield estimation,soil quality evaluation and modern agricultural production.Although the traditional methods of soil agrochemical analysis can accurately determine the content of soil organic matters,field sampling can easily destroy the soil and it needs to use chemical methods indoors which takes a long time.There are also many other problems,such as high cost,heavy element workload,and impossibility to get spatial distribution data of soil organic matters in a short time and continuous determination of soil organic matter in a large area,all of which make these traditional methods very limited.Conventional optical remote sensing has relatively few bands and low spectral resolution.Although it can cover a wide area,there are great errors in regional monitoring of the content of soil organic matters.Soil hyperspectral technology is a new method to monitor soil organic matters.The corresponding ground remote sensing experiment is an important basis to ensure the detection precision.The spectral measurement close to the ground is an important basis for remote sensing images and data interpretation.Meanwhile,soil hyperspectral technology has the characteristics of pollution-free,damage-free,low cost,large amounts of information and high efficiency,etc.,which is widely applied in the study of soil organic matters.This study takes the farmland soil and desert soil of the arid area in South Xinjiang as the objects of study.The field Spec Pro FR spectrometer of ASD Company is used to measure the organic matter content of the samples and visible near infrared spectrum data in a dark room where light conditions can be controlled.On this basis,Different prediction models were built by adopting various methods of spectral data transformation.After comparing different modeling methods,the results showed that PLSR had the highest modeling precision.When building the spectral quantitative inversion model of the organic matters in desert soil of South Xinjiang,it was found that under the processing modes of SGS+MAN+OSC,the numerical values of R~2 was the highest and RMSE was the lowest in the modeling set,0.81 and 0.98,respectively.The R2,RMSE,RPD of the prediction set of this model were 0.76?0.99 and 2.01 respectively,which indicated that the model had good prediction ability and can be recommended to be a spectral quantitative inversion model for the organic matters of the desert soil in South Xinjiang.When building the spectral quantitative inversion model of soil organic matters of the farmland soil in South Xinjiang,it was found that the precision of the model was the most stable under the processing modes of SGS+MAN+OSC,which can be recommended to be a spectral quantitative inversion model for the organic matters of the farmland soil in South Xinjiang.Finally,soil spectrum can be classified according to the characteristics of soil spectral curve and reflectivity data.And the characteristics of soil reflective spectrum is divided into four categories by adopting Fuzzy K-Means clustering algorithm,and then use the method of PLSR to build the organic matters prediction models for each category and all of them.The results show that compared with global modeling,the precision of spectral classification modeling significantly improves.And the RPDs of the models built for the four categories are all larger than 2.0,which can significantly improve the prediction precision of the models'stability.