基于小波理论的干旱区内陆湖泊叶绿素a的TM影像遥感反演

叶绿素a(Chl-a)是衡量湖泊富营养化的重要指标,利用遥感技术动态监测面积较大的湖区水体中Chl-a浓度对了解湖区水质具有重要意义。以内蒙古乌梁素海为例,提出利用TM影像中的水体实测光谱进行小波去噪和光谱信号重构,并结合水质采样实测数据进行神经网络拟合,建立光谱反射率比值与Chl-a浓度的反演模型的方法。结果显示:小波理论和神经网络相结合的模型可以适用于估算乌梁素海Chl-a浓度,去噪后Chl-a浓度与光谱信号的相关系数(-0.575)较去噪前(-0.417)明显增强,去噪后的采样点光谱信号与Chl-a浓度之间表现出比原始信号更强的负相关性,证明了去噪后的观测值可进一步减弱随机误差的干扰和去除噪声,使观测数据更加逼近Chl-a浓度的真实情况,图像去噪重构结果显示重构后的光谱范围较之前有所缩窄,部分信号点得到了增强,但基本剖面结构并没有产生较大变化,反演模型的平均相对误差为0.142,与其他研究相比差别不大。反演得出的乌梁素海Chl-a浓度分布反映了污染源的分布,同时说明了乌梁素海Chl-a浓度在时空分布上呈现一定的差异,表现为丰水期呈现浅水区Chl-a浓度值高于湖心区,来水区高于其他湖区的分布趋势,枯水期乌梁素海中部呈现由西向东Chl-a浓度逐步降低的分布规律,西部呈均一化分布。反演模型基本可以满足实际预测的需要。但模型在具体应用中在影像数据采集、数据量及算法方面还有很大的改进空间,该方法的提出为干旱区大型内陆水体富营养化的实时定量遥感监测提供了新的解决方案。

A wavelet theory based remote sensing inversion of chlorophyll a concentrations for inland lakes in arid areas using TM image data

Chlorophyll a (Chl-a) concentration is an important indicator for measuring eutrophication and lake water quality. Therefore, a fast and sensitive remote sensing method for Chl-a concentrations is urgently needed, as this will enable real-time spatio-temporal monitoring of Chl-a distribution in large inland lakes, which will enhance water quality management and protection. Using Wuliangsuhai Lake (Inner Mongolia) as an example, this study established an effective remote sensing inversion method for Chl-a concentrations, based on Landsat Thematic Mapper (TM) image data. Chl-a concentration data from January 2010 to November 2014 was collected by the Environmental Monitoring Station of Bayannur city. TM images were acquired by the Information Center of the Chinese Academy of Sciences. After pre-treatment, the Wuliangsuhai TM images were de-noised and reconstructed based on a wavelet analysis. A neural network method was subsequently used to construct a model that relates the TM spectral reflectance ratios and Chl-a concentrations. The results indicated that the proposed method of combining wavelet analysis with a neural network model is suitable for inversely remote sensing Chl-a concentrations. The correlation coefficient between the wavelet de-noised spectral signal and the Chl-a concentration (-0.575) was higher than when the original spectral signal was used (-0.417). Furthermore, the negative correlation between the de\oised spectral signal and water sample Chl-a concentrations was stronger than the original one. This demonstrated that the de\oised monitoring values could further reduce the interference of random errors and noise. Furthermore, the remotely sensed Chl-a values could approach the sampled Chl-a concentrations. In addition, the de\oised reconstruction of the TM images had a narrower reconstructed spectral than before, and part of the signals were enhanced. Nonetheless, the basic cross\sectional structure of the images did not change notably. The mean relative error (MRE) of the proposed method was 0.142, and differed little from other models. In addition, the distribution of Chl-a concentration based on the TM inversion method was consistent with the distribution of the Wuliangsuhai Lake pollution sources. The spatio\temporal distribution of Chl-a concentrations showed some variability. In the wet season, the Chl-a concentrations in shallow water areas were higher than those in the central area, whereas the Chl-a concentrations in the inlet area were higher than those in other areas. In the dry season, the Chl-a concentration decreased gradually from west to east in the middle of the lake, and showed a homogeneous pattern in the west of the lake. Overall, the precision of the TM remote sensing inversion method achieved a satisfactory prediction accuracy. However, given the lack of sufficient Chl-a monitoring sites and monitoring data, some factors that influenced the spectral reflectance ratio of TM image could not be removed or controlled for. Some improvements on TM image data acquisition, such as algorithm optimization and model verification, should therefore be a priority for the future. Alternatively, high\esolution remote sensing image data could be used to acquire the spectral reflectance ratio of lake water, instead of TM images. In conclusion, this study could be used to improve lake water quality monitoring technologies, as well as contribute to real\time water quality monitoring. The proposed method for Wuliangsuhai Lake could be applied in other areas as well, and for other water pollutants.