基于GRNN的ALI多光谱遥感数据土壤盐分反演研究

Soil Salt Content Retrieval from ALI Multispectral Image Based on GRNN

受环境变化和人类活动的双重影响,土壤盐渍化已经成为土壤退化的重要形式。及时展开土壤盐渍化研究对改善现状和预防其进一步发展具有重要意义。本文以黄河三角洲一处典型区域为研究对象,在野外光谱测量和实验室理化分析的基础上,采用广义回归神经网络(GRNN)方法建立了土壤盐分反演模型,模型的决定系数为0.855,均方根误差为0.119g.kg-1。将GRNN模型应用到ALI反射率图像上得到土壤盐分反演分布图。结合野外调查结果发现,GRNN方法得到的土壤盐分值的空间分布结果与实际情况一致。

Soil salinization has been a major progress of soil degradation due to environment evolution and anthropological activities. And therefore, it is of great significance to carry out timely detection of soil salinization, aiding the reclamation and prevention of saline soils. In this paper, a typical subregion of Yellow River Delta was studied for soil salt content （SSC） retrieval. For this purpose, generalized regression neural network （GRNN） were performed based on soil samplings and corresponding laboratory physico-chemical analysis. The model shows a good relationship between the measured and predicted SSC with determination coefficient （R2） of 0. 855, and root mean square error （RMSE） of 0. 119. Subsequently, the model was applied to ALI reflectance image to yield the SSC map. A comparison shows that GRNN-derived SSC is more likely to accord with ground observations.