Study On Rapid Determination Method Of Organic Matter In Black Soil Region Based On Spectrum And Its Application

The black soil regions with high organic matter content and fertility are an important grain productions base in China. However, due to irrational farming and blind reclamation, the quality of black soil is deterioration and the soil erosion is serious. So how to adjust and judge the reasonableness of reclamations has become an urgent problem nowadays. The organic matter as one important part of soil plays a key role in forming the soil structure and improving the quality of soil. Thus it is a basic condition to master the organic matter dynamic characteristics in soil for agricultural production management, execution of precision and sustainable agricultural development. However, conventional soil property analysis is relatively slow, expensive and laborious. Although using Spectrometer to build soil hyperspectral reflectance models can fast get a large amount of the organic matter content, the Spectrometer is expensive、complex and undefined standard, which is restricted to use in some fields. Therefore, this paper based on the present existing methods and problems in practical application studies the quantitative relationship between the color and black soil organic matter by the digital camera as an alternative tool for rapid and accurate estimation for black soil organic matter content.This research took the black soil region of Jilin province as the study object and discussed the spectra properties and the relationship between organic matter content and the soil surface RGB intensity values in black soil region through ASD portable spectrometer, digital cameras, remote sensing image of MODIS and other methods respectively at first. Secondly, it defined the sensitive band range and quantitative relationship with RGB for organic matter. Next, it built the soil organic matter estimation model. Finally, the organic matter mapping of the study area was achieved by combining the remote sensing of MODIS with Kriging interpolation method. The main research results and contents were as follows:By analyzing the correlation between organic matter and soil spectral reflectance, the organic matter showed a high correlation in the visible spectrum. The maximum correlation was located at 699nm and the correlation coefficient was -0.7022. After the spectral data were smoothed by the nine point weighted mean, the maximum correlation coefficient was -0.702 as before. The spectral curves were smoothed and reduced spectra data noise while retaining the original spectral information by the nine point weighted mean. The original spectral data were transformed through the reciprocal, logarithmic and differential mathematical means and increased the correlation between organic matter and reflectance. The reciprocal differential transformation was the most obvious to increase the correlation between organic matter and reflectance, and the maximum correlation coefficient was up to -0.8346. The correlation curve of the original spectrum was similar with the reciprocal and logarithmic transformation, and the correlation cure of differential transformation was fluctuating greatly that was quite different from the original correlation curve. The similar correlation cure of original spectrum, reciprocal and logarithmic used the overlay analysis to determine the wavelength range of 400-1400nm that was sensitive to the change of organic matter content as the variable scope for the modeling of reciprocal and logarithmic, and the 710-990nm was selected as the modeling variable range for reciprocal differential spectrum data. By means of stepwise regression analysis, the best estimation model of soil organic matter content was the reciprocal regression model, with the calibration R2 to 0.8179, RMSE as 0.3232, and the prediction R2 to 0.7534, RMSE as 0.2584, which could completely express the regional variation characteristics of organic matter.After soil surface images were acquired by the digital camera, the different AOI (Region of interest) of images were selected and discussed. In order to ensure that the sampling areas not only represent the true color values but also eliminate the influence of edge effect, the 800×800Pixels were determined as the optimum sampling areas, and extracted the average as actual RGB values of AOI. The RGB values of soil samples showed a high correlation with the organic matter. The largest correlation with organic matter was R, and the correlation coefficient was up to-0.7334. The soil surface RGB gray value would increase and decrease due to the heterogeneity and non-soil material, which had a great influence on the extraction of true RGB values. It would be a good way to eliminate the influence of subjective reasons or non-soil substances on the image RGB values through gray histogram statistics to obtain maximum, and the correlation with organic matter significantly improved when used the maximum as the image RGB values. The maximum correlation coefficient of R increased to -0.7432 by gray histogram processing. The RGB color model was transformed into La*b*, Lc*h* and Lu*v* of CIE, and the univariate model of Lu*v* showed the best prediction results, with the calibration R2 to 0.6930, RMSE as 0.4196. In the multivariate model, RGB color model was the best for overall result, with the calibration R2 to 0.7339, RMSE as 0.3907, and the prediction R2 to 0.7405, RMSE as 0.265. By comparing with the spectrometer at same band, the prediction accuracy of RGB model was better than the spectrometer due to the influence of the size of sampling area and non-materials. The RGB values acquired by digital camera can express the variable characteristics of organic matter and the digital camera as an alternative tool can be used to fast, accurate and non-destructive predictions of SOM contents in black zone.The organic matter made correlation analysis with soil particles of lmm,0.5mm, 0.25mm,0.15mm and 0.088mm respectively. The R value of 0.088mm among all soil particles has largest correlation with the organic matter, and the correlation is up to-0.7682. When the soil particle reduced to 0.25mm, the correlation with organic matter was going up. When the soil particle was 0.15mm, the correlation increased significantly. The different soil particles improve the correlation with organic matter through the mathematical transformation, and the reciprocal transformation has highest correlation with organic matter. The correlation of R value of 0.08mm through reciprocal transformation is up to 0.8128. In multivariate regression model, the soil particle of 0.088mm has the best prediction, with the calibration R2 to 0.7546. RMSE as 0.3751, and the prediction R2 to 0.815, RMSE as 0.2238, which was well to reflect the variable characteristic for organic matter. In the comparison of different particle correlation and predication performance, when the soil particles reduced to 0.25mm, the more the larger particles hold proportion, the worse the noise of RGB was reduced. Thus, the different proportion of particle would exist in uncertain factor for 0.25mm so that the prediction result was unstable for overall. When the size of particle was 0.15mm, all of different particles proportion could be very good to reduce the impact of noise on the prediction results, so the correlation with organic matter was remarkable increased. The correlation analysis of mixed samples showed the coexistence of big and small particle could be very good to fill the gap between the particles and reduce the impact of micro-topography, wihcih increased the correlation with organic matter. The particle proportion of 0.15mm would directly affect the prediction accuracy for all particles. The mixed sample of 0.25mm could, taking into account time costs as well as predictive performance for model, be used as the best choice for rapid prediction of black soil organic matter.The accuracy improvement for geostatistical interpolation would be based on a large number of samples, thus the limited samples resulted from a worst effect for Kringing interpolation. The higher organic matter content is, the lower the remote sensing records reflectance. Thus, according to the correlation between the reflectance and organic matter, the spatial mapping of organic matter could implement used a small amount of samples.Due to the presence of mixed pixels and other environmental factors, so that the recorded reflectance of remote sensing can’t really correspond to the actual surface reflectance, thus the accuracy of mapping would be affected when using MODIS to map organic matter in study area. Therefore the model established by the reflectance was R2=0.349, RMSE=0.425, which was low prediction accuracy and couldn’t reflect the variable characteristics for organic matter in black soil zone. In this paper, taking advantage of the geostatistics to express autocorrelation between samples, the difference between the geostastics and remote sensing retrieval for organic matter could eliminate the region that the variability was large, reducing the impact of mixed pixels, and achieve the effective choice for building samples. Meanwhile, the predicted mapping for organic matter implement by connecting the soil moisture and reflectance, which could fully consider the actual situation in the field and the samples characteristics of autocorrelation and independence. Thus the methods effectively improve the predictive accuracy, and carry out the study area mapping for organic matter.