三峡库区回水区叶绿素a浓度遥感反演方法研究

发布时间：2019-04-04 09:37

【摘要】：叶绿素a浓度是反映水体富营养化最重要的指标。当前内陆二类水体的叶绿素a浓度遥感反演方法研究主要集中在普通湖泊如太湖、鄱阳湖等区域,对于三峡库区回水区这类河流型湖泊水域的研究较少。本研究以三峡库区回水区高阳湖、汉丰湖为研究区,利用GF-1 WFV遥感数据及实测数据,建立遥感反演模型,实现大面积动态反演三峡库区支流回水区叶绿素a浓度。论文的研究工作及主要结论如下:(1)利用GF-1 WFV遥感数据及实测水体叶绿素a浓度,建立波段比值回归分析模型,反演回水区叶绿素a浓度,波段比值模型公式为=0.002(5.313),x为遥感影像波段组合B4/B3的值,均方根误差RMSE为5.6739,判定系数R2为0.68672。波段比值模型的反演效果并不够理想,其原因为三峡库区回水区是二类水体,其光学性质复杂,应用一类水体线性反演模式拟合较为片面,难以获得最佳拟合效果。(2)相比于波段比值的线性回归方法,BP神经网络模型具有模拟复杂非线性问题的功能。本研究建立模型结构为4-4-1的BP神经网络模型,经相关分析后选取B4-B3、B4-B2、B4/(B2+B3)、B4/B3四个波段组合做为BP神经网络模型的4个输入神经元,设置隐含层为单层,隐含层节点数为4,以实测水体叶绿素a浓度值为输出神经元对模型进行训练。经过训练后的BP神经网络模型R2为0.8389,RMSE为3.8745。将经过水域提取的GF-1 WFV影像用于训练好的BP神经网络模型,反演2016年5月-8月三峡库区回水区叶绿素a浓度分布,反演结果良好。(3)对比两种水体叶绿素a反演模型的精度,BP神经网络模型与波段比值模型的判定系数R2分别为0.8389及0.6867;均方根误差RMSE分别为3.8745及5.6739;平均相对误差e分别为20.6%及55.9%。对比结果证明将GF-1 WFV影像应用BP神经网络模型反演三峡库区回水区叶绿素a浓度较波段比值模型精度有所提高,反演效果优于波段比值模型。
[Abstract]:Chlorophyll a concentration is the most important index to reflect eutrophication. At present, the remote sensing retrieval of chlorophyll a concentration in inland waters mainly focuses on the common lakes, such as Taihu Lake, Poyang Lake, etc., but there is little research on the river-type lake waters such as the backwater area of the three Gorges Reservoir area. Taking Gaoyang Lake and Hanfeng Lake as study area in the backwater area of the three Gorges Reservoir area, using GF-1 WFV remote sensing data and measured data, a remote sensing inversion model was established to realize the large area dynamic inversion of chlorophyll a concentration in the tributary backwater area of the three Gorges Reservoir area. The research work and main conclusions of this paper are as follows: (1) using GF-1 WFV remote sensing data and measured chlorophyll a concentration of water body, the ratio regression analysis model is established, and the chlorophyll a concentration of backwater area is retrieved. The formula of the band ratio model is = 0.002 (5.313), x is the value of the band combination B4/B3 of remote sensing images, the root mean square error RMSE is 5.6739, and the determination coefficient R2 is 0.68672.) The inversion effect of the band ratio model is not ideal. The reason is that the backwater area of the three Gorges Reservoir area is a second type of water body, and its optical properties are complex. The linear inversion model of one type of water body is relatively one-sided. It is difficult to obtain the best fitting effect. (2) compared with the linear regression method of band ratio, the BP neural network model has the function of simulating complex nonlinear problems. In this study, the BP neural network model with the structure of 4 ~ 4 ~ 4 ~ (1) was established. After correlation analysis, B4 / (B2B3), B4 / (B2B _ 3), B4 / (B _ 2B _ 3) were selected as four input neurons of the BP neural network model, and B4 / (B _ 2B _ 3) were selected as the input neurons of the model. The hidden layer was set as the single layer and the number of nodes in the hidden layer was 4. The model was trained with the measured chlorophyll a concentration value as the output neuron. After training, the BP neural network model R2 is 0.8389, and the RMSE is 3.8745. The GF-1 WFV images extracted from the water area are used in the trained BP neural network model to retrieve the chlorophyll a concentration distribution in the backwater of the three Gorges Reservoir area from May to August 2016. The inversion results are good. (3) comparing the accuracy of the two water chlorophyll a inversion models, the determination coefficients of BP neural network model and band ratio model are 0.8389 and 0.6867, respectively; The mean square error (RMSE) was 3.8745 and 5.6739 respectively, and the mean relative error (e) was 20.6% and 55.9% respectively. The results show that the accuracy of the BP neural network model for retrieving chlorophyll a concentration in the backwater area of the three Gorges Reservoir area is higher than that of the band ratio model, and the inversion effect is better than that of the band ratio model.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X87

【相似文献】