利用高光谱和GF-1模拟多光谱进行土壤有机质预测和制图研究

Predication and Mapping of Soil Organic Matter Content Using Hyperspectra and GF-1 Multi-spectra

利用土壤有机质(SOM)高光谱数据和模拟GF-1多光谱影像的波段响应函数生成的宽波段多光谱模拟数据,对比高光谱预处理和构建土壤植被指数,探索模拟GF-1光谱预测SOM的潜力。研究表明,SOM的一阶微分高光谱和模拟GF-1光谱数据构建的土壤指数与SOM的相关性最好。PLSR建模分析表明采用一阶微分高光谱数据可以很好的对SOM进行预,而且模型稳健(R2=0.962,RPD=4.87);模拟GF-1光谱也可以较好的进行SOM的预测,但是模型的稳定性相对较差R2=0.557,RPD=1.43。同时,SOM制图的空间分布表明,采用一阶微分光谱数据和模拟GF-1数据预测得到的SOM含量与实测的SOM表现出相似的空间分布特征。这为采用多光谱数据进行大尺度、大范围的SOM预测提供了基础。

The present study fo cused on assessing the feasibility of multi-spectral data in monitoring soil organic matter（SOM）. The data source came from hyperspectra measured under laboratory condition, and simulated multi-spectral data from GF-1 remote sensing images. According to the reflectance response function of GF-1, the hyperspectra was resampled for the corresponding bands of multi-spectral sensors. The correlations between hyperspectra and simulated reflectance spectra with different soil vegetation indexes and SOM content were calculated,indicating the highest correlation coefficient for the first derivatives of hyperspectra and simulated GF-1 reflectance.The partial least square regression（PLSR） method was used to establish experiential models of SOM content estimate.The first order derivative hyperspectral data showed good result of predicted SOM with R2= 0.962, and the model was steady with RPD = 4.87. Simulated GF-1 spectral can well predict SOM（R2= 0.557）, but the stability of the model was relatively poor（RPD = 1.43）. Meanwhile, the spatial distribution of SOM showed that similar spatial characteristics between measured SOM and predicated SOM by using the first derivative of the measured hyperspectral data and simulated GF-1 multi-spectral data. Therefore, it had potential ability to evaluate SOM in large-scale with multi-spectral data.