基于高光谱的土壤有机质含量估算研究

高光谱遥感技术以其光谱分辨率高、波段连续性强、数据丰富的特点,因而在土壤养分研究中得到广泛应用.通过土壤钉机质的高光谱遥感分析,可以充分了解土壤养分的状况及动态变化,为指导农业生产及保护农业生态环境提供科学依据.本文基于江西省余江县和泰和县采集的34个红壤土样350～2 500 nm波段的光谱曲线,研究了土壤光谱与土壤有机质含量之间的关系.先对土壤反射率光谱进行两种变换:一阶微分(R')、倒数的对数log(1/R),然后在提取特征吸收波段的基础上,运用多元逐步线性回归法和偏最小二乘回归法建立相应的估算模型,并对模型进行检验.结果表明,偏最小二乘回归法优于多元逐步线性回归法,其建立的高光谱估算模型具有快速估算土壤中有机质含量的潜力.

Studies on the estimation of soil organic matter content based on hyper-spectrum

Hyperspectral remote sensing technology can be extensively applied in soil nutrient research due to its three special advantages, high spectral resolution, strong waveband continuity as well as a great deal of spectral information. Based on analyzing the soil organic matter content using hyper-spectral remote sensing technology, soil nutrients status and its dynamic changes can be fully understood, thus providing the scientific basis for guidance of the agricultural production and protection of agricultural ecological environment. The present paper studies the relationship between soil spectrum and soil organic fraction based on spectrum curves (ranging from 350 to 2500 nm) of 34 soil samples, which were collected in Yujiang and Taihe County, Jiangxi Province. First, soil reflection spectrum was mathematically manipulated into first derivative reflectance spectra (FDR) and inverse-log spectra (log(1/R)); second, the relationship between soil spectrum and soil organic fraction was investigated by step-wise multiple linear regression (SMLR) and partial least square regression (PLSR) on the ground of characteristic absorption; third, corresponding estimation model was built and examined. The result conveys that spectral data are compressed by carrying out arithmetic average operation by 10 nm for intervals. The first derivative of the reflectivity is an effective spectrum indicator, in the stepwise multiple linear regression analysis of soil organic matter, for the first derivative transformation, the regression models' precision of establishment and verification increased. The model built by PLSR method based on the characteristic absorption bands precedes that of SMLR. In the PLSR model of soil reflection spectrum and the inverse-log spectra, the test samples' average of relative error is 16% and 17% respectively, the correlation coefficient between retrieval value and measured value is 0.84 and 0.91 respectively, for it's faster to estimate the soil organic fraction.