Inversion Of Soil Organic Matter In Cotton Fields Based On In Situ Spectra In The Field

Soil organic matter（SOC）is a key indicator of soil quality and is related to important functions such as nutrient cycling and storage,uptake and retention of pollutants in soil ecosystems,plays an important role in many physicochemical properties of soils,and has a significant impact on soil spectral reflectance.Therefore,in the context of rapid population growth and global climate change,maintaining and improving soil quality is essential to secure global food production.Spatially distributed high-resolution and soil analysis data can support the development of new agricultural management strategies and practices,but data access can be limited by the costs associated with traditional laboratory analysis.Soil spectroscopy in visible-near infrared（VIS-NIR）spectroscopy can replace traditional soil organic matter determination with its advantages of efficiency and economy.In situ visible-nigh infrared spectroscopy（VIS-NIR）in the field can effectively improve the detection efficiency of soil properties,but the model predictions are reduced due to the influence of moisture factors in the in situ soil.The effective elimination of the influence of moisture factor is the challenge of using in situ field spectroscopy to predict soil properties with high accuracy,and it is also a breakthrough for soil spectroscopy technology to shift from indoor to field.The effective solution of this problem can eliminate the process of soil sample collection and indoor pre-processing,and realize the in situ field spectroscopy determination of soil properties.In this study,a total of 116 surface soil samples from 0-20 cm were collected by grid sampling method in the cotton field of the 12th regiment of Alar Reclamation District in the southern region of Xinjiang,and the in situ spectral data of 231 sample points in the field were collected using an SR-3500portable feature spectrometer.The soil samples were air-dried,ground and sieved,and then the indoor spectra and soil organic matter contents were determined.After the spectral pre-treatment,four regression models were combined to establish the organic matter regression models of soil field in-situ spectra and indoor spectra respectively,and the accuracy of indoor and field in-situ spectral models were compared and analyzed.Three water removal algorithms were used to improve the model accuracy of predicting soil organic matter using field in situ spectra.Finally,the optimal model for organic matter prediction after water removal was screened to map the distribution of organic carbon content in the soil surface layer.The results of the study showed following.（1）The higher the soil organic matter content,the lower the soil spectral reflectance,and the characteristic bands affected by soil organic matter were concentrated near 1400 and 1900 nm.The reflectance of soil field in situ spectra and the reflectance after the treatment of continuum removal were lower than that of soil indoor spectra,especially at the 1400 and 1900 nm bands,the soil spectral reflectance was lower and the trough was deeper.（2）The indoor spectra and in situ field spectra were modeled by combining twelve spectral preprocessing methods such as continuous statistical removal and first-order differentiation with four regression models,including partial least squares,principal component regression,support vector machine,and random forest,respectively.The accuracy of the indoor spectral organic matter models were all greater than that of the in situ spectral organic matter models in the field.Among the different spectral pre-processing,the accuracy of the organic matter model after first-order differentiation,continuous statistical removal,baseline correction and logarithmic transformation was improved compared with the original spectrum,and the accuracy of the organic matter model after second-order differentiation and inverse treatment was decreased.The model accuracy of indoor spectra after first-order differential transformation was the best,with R² of 0.90,RPD of 2.92,RMSE of 1.10 g Kg^-1and MAPE of 0.10,which could accurately predict the content of soil organic matter;while the model accuracy of field in situ spectra after continuous statistical removal was the best,with R²of 0.73,RPD of1.89,RMSE of 1.70 g Kg^-1 and MAPE of 0.17,which could The organic matter content of the soil was estimated approximately.（3）After modeling predictions by three water removal algorithms,EPO,DS,and PDS,combined with the RF model,the models were all able to improve the accuracy of predicting soil organic matter in in situ spectral data with accuracy（RPD>1.5）.Among them,the prediction accuracy of field in situ spectra processed by EPO algorithm is the highest,with R² of 0.83,RPD of 2.04,RMSE of 1.58 g Kg^-1and MAPE of 0.14,which is closer to the model accuracy of indoor spectra with R² of 0.90,RPD of 2.92,RMSE of 1.10 g Kg^-1 and MAPE of 0.10.It can realize the improvement of the model accuracy of direct prediction of soil organic matter by in situ spectra in the field.（4）The inversion results of soil organic matter in the Twelve Regiments of Alar in South Xinjiang were compared by two different interpolation methods,ordinary kriging and inverse distance interpolation.The inverse distance interpolation method with the highest mapping accuracy after removing the influence of moisture factor to invert the organic matter,with R²of 0.85,RPD of 2.52,RMSE of 1.19 g Kg^-1 and MAPE of 0.10,can accurately reflect the distribution of soil organic matter content in the study area.