基于多光谱遥感的扎龙湿地湖泡水深反演研究

Inversion of Water Depth of Lakes in Zhalong Wetlands based on Multispectral Remote Sensing

以扎龙湿地龙泡子为研究对象,利用58个实测水深数据和季相最接近的Quick Bird数据,建立湖泊水深的反演模型。探索性地建立了单波段和多波段组合的线性（多元）回归模型、指数模型、二次多项式模型、微分模型和对数模型等;通过对比模型的决定系数R2,比较模型精度;线性模型、对数模型、指数模型和幂指数模型的R2小于0.5,而二次多项式模型和多元线性回归模型的R2大于0.5,精度相对较高;筛选出拟合度较高的模型,用20个实测验证样本,采用相对误差和均方根误差进行模型精度评价;最后,利用精度较高的模型,进行龙泡子水深反演计算。水深反演结果表明,用选出的模型反演得到的龙泡子水深基本一致,为170～200 cm,即使有稀疏的水草覆盖,依然可以表现出水深渐变的趋势。以蓝、绿、红和近红外波段多光谱遥感反射率为自变量,建立的线性湖泡水深反演模型y=123.990-3.332B1＋183.859B2-237.133B3-37.143B4（y为水深;B1、B2、B3和B4分别为蓝、绿、红和近红外波段的水体反射率）,能较好地反演扎龙湿地湖泡的水深。

It will be a large challenge or a great demand to inverse the water depth of lakes due to the bottom composition complex, water impurities and the tiny variety of water depth based on remote sensing technology. Taking lakes in Zhalong Wetlands as the study objects, this paper had an exploration on the water depth inversion. Ecological system of Zhalong Wetlands is the largest natural protection area for the target of protecting the cranes in China. In this paper, the different water depth inversion models are set up with 58 measured data of water depth on October 2-6, 2010 and the Quick Bird images on September 16, 2009. The linear（multiple） model, index model, the quadratic model and logarithm model based on a single band and multi-band reflectivity combination were established respectively. The multiple correlation index of R2 was used to evaluate the precision for all the models. The R2 of the linear model, logarithm model and exponential model, power exponent model was less than 0.5, while those of the quadratic polynomial model and the multivariate linear model were greater than 0.5, which showed that the quadratic polynomial model and the multivariate linear model were superior to the linear model, logarithm model, exponential model and power exponent model. Further, the better models with bigger fitting coefficient were selected to be verified by the observed 20 samples.Relative error and root mean square error were used to evaluate the model precision. Finally, models with high precision were chosen to inverse the water depth. The relationship between water depth from filed measured and multi-spectral Quick Bird images was analyzed, the higher correction coefficients belonged to red and near infrared wave band, meant that the spectrum of red and near infrared wave bands was sensitive with water depth changes. To inverse the water depth of multi-spectral Quick Bird images based on 4 multispectral models and the results indicated that the linear model could reflect water depth gradient well. This paper selected Landsat remote sensing images on September 23, 1988, September 30, 1999, and September 28, 2007 as data to inverse water depth of ten lakes in Zhalong Wetlands. From the inversion results of water depth, the water depth of the selected models in Longpaozi Lake evenly varies from 170-200 cm even covered with sparse plants. The evident water depth gradient could be seen from the inversed water depth map. The water depth gradient can＇t be expressed obviously by the model based on reflectivity of single wave band, but that was much better by the linear model based on reflectivity of the 4 wave bands.