Abstract:
The objectives of this study were to estimate models for organic matter content of Shajiang black soil, which provided a basis and reference for rapid quantitative estimation of soil organic matter. In this study, Shajiang black soil in Shangshui county of Henan province was collected, the organic matter content and hyperspectral reflectance of Shajiang black soil were analyzed and tested simultaneously, and hyperspectral characteristics of Shajiang black soil were studied. Then, the original reflectance was converted to the first derivative spectral with Savitzky-Golay (SG) smoothing. Difference Spectral Indices (DSI), Normalized Spectral Indices (NSI), and Ratio Spectral Indices (RSI) of these two forms spectral were calculated from all available combinations with the reflectance of two random bands between 350 and 2 500 nm, and correlated to soil organic matter content, then the key spectral index and quantitative models for organic matter content of Shajiang black soil were developed. On the other hand, the sensitive bands of soil organic matter were extracted with the Genetic Algorithm (GA) and quantitative models of soil organic matter using Support Vector Machine (SVM) were established. The results showed that the spectral reflectance of Shajiang black soil under different organic matter levels had the same trend, and organic matter levels had a certain influence on reflectance. Organic matter content was higher, the reflectance would be lower, and on the contrary, the reflectance would be higher. The bands with good correlation between organic matter content and DSI, NSI, RSI based on the original reflectance were mainly concentrated near 650, 1 500 and 2 200 nm, DSI composed of reflectance of 995 and 1 911 nm, NSI composed of reflectance of 2 067 and 2 208 nm, RSI composed of reflectance of 1 037 and 1 908 nm had the better fitting degree. The bands with good correlation between organic matter content and DSI, NSI, RSI based on the first derivative spectral after SG smoothing were mainly concentrated in the combined band region of 1 350-2 000 and 600-1 000 nm, DSI composed of the first derivative of 792 and 1 420 nm, NSI composed of the first derivative of 792 and 1 389 nm, RSI composed of the first derivative of 792 and 1 389 nm had the better fitting degree. For all the spectral indices that were calculated, RSI composed of the first derivative of 792 nm and 1 389 nm gave a better prediction performance, the coefficient of determination was 0.81. Testing of the monitoring models within dependent data indicated that the coefficient of determination and root mean square error of validation were 0.91 and 1.56, respectively. In addition, the sensitive band ranges based on the original reflectance which selected by GA were 461-470, 611-620, 661-670, 741-750, 1 461-1 470, 1 891-1 900, 1 901-1 910, 2 011-2 020, 2 071-2 080 and 2 141-2 150 nm, and the sensitive band ranges based on the first derivative with SG smoothing which selected by GA were 521-530, 531-540, 671-680, 761-770, 771-780, 831-840, 1 431-1 440, 1 451-1 460, 1 871-1 880 and 1 881-1 890 nm for the same sample. The bands of the first derivative with SG smoothing mentioned above were used as input to SVR, and the quantitative model of soil organic matter performed the best. The coefficient of determination and root mean square error of modeling and validation was 0.95 and 0.91, 1.01, and 1.69, respectively. The compared quantitative model with Support Vector Machine, RSI composed of the first derivative of 792 and 1 389 nm had a little lower modeling accuracy, but it could meet the need for estimating the organic matter content of Shajiang black soil. It was concluded that both methods based on RSI composed of the first derivative of 792 and 1 389 nm and SVM can estimate organic matter content of Shajiang black soil accurately.