Comparison of multivariate statistical analysis and fuzzy recognition algorithm for quantitative mapping soil organic matter content with hyperspectral data

Abstract

The soil organic matter is one of the important criterions of soil fertility. Mapping and dating soil organic matter is of great importance in soil use and evaluation. In this paper we compare two measures of multivariate statistical analysis (MSA) and fuzzy recognition algorithm (FRA) for quantitative mapping soil organic matter content using Hyperspectral remote sensing. This study was tested in Henshan County, northern ShanXi Province of China. On the one hand, the ratio of the reflectivity reciprocal-logarithm's first derivative of 623.6nm against the reflectivity reciprocal-logarithm's first derivative of 564.4nm was chosen as the sensitive retrieval parameter and build up the retrieval models. Then, the best quadratic retrieval model was utilized to map the SOM content by calculating each pixel of Hyperion image, the adjusted R square coefficient is 0.8684. On the other hand, by analyzing the correlation between spectrally reflective data and SOM concentrate, the first derivative of logarithmic reflectance at sensitive bands of 393nm, 444nm, 502nm, 1455nm and 1937nm were confirmed as the retrieval indicators due to the notable correlation coefficients. Finally, the most optimized-retrieval model, utilized to the Hyperspectral data for SOM quantitative mapping, was build up by using the fuzzy recognition theory. The correlation coefficient of the retrieval model is 0.981. It is found that result of fuzzy recognition algorithm is better than that of traditionally statistical analysis, with the mean predicted error of 8.43% as compared to 10.42% for quadratic retrieval model. It is concluded that this fuzzy recognition algorithm for Hyperspectrally quantitative mapping SOM is available and the result map is reliable and significantly correlative with known stabilization processes throughout the study area. Moreover, the fuzzy recognition algorithm developed in this paper could be applied to other domain of quantitative remote sensing.