基于深度神经网络算法的水体透明度反演方法

水体透明度能够直观反映湖泊水质状态,掌握长时间大尺度湖泊水体透明度是控制和改善湖泊水生态环境的关键。由于滇池的水质原位监测工作起步较晚,导致长时间序列的历史湖泊水体透明度数据的缺失。为此,以滇池为研究区,以深度神经网络算法为理论基础,以原位监测和MODIS遥感影像为数据,对2001年1月1日—2018年12月31日滇池水体透明度进行反演,并利用地理空间分析方法探讨了滇池湖泊水体透明度时空变化特征。研究结果表明:(1)提出的反演模型具有较好的性能(RMSE=0.1359,MAE=0.1134),能够客观反映湖泊水体透明度状况;(2)时间变化特征分析结果表明,滇池水体透明度总体呈现下降趋势,综合变化率为-0.08 m/10 a;(3)空间变化特征分析结果表明,水体透明度较高的区域下降率较大,水体透明度较低的区域变化趋势相对稳定,距离城区及居民区较近的水体透明度相对较低;人类活动将成为影响滇池水体透明度变化的重要因素,同时也是造成滇池水体污染的主要因素。

Secchi depth inversion of Dianchi Lake and its temporal and spatial variation analysis based on deep neural networks

Lake is an important ecological resource, which not only determines the quality of regionally ecological environment, but also guarantees the sustainable development of the city. Lake ecological environment can reflect the environmental situation of the region. Secchi depth (SD) can directly reflect the state of lake water quality, and mastering long-term and large-scale lake SD is the key to control and improve the lake water environment. Dianchi Lake, located in the middle of Yunnan Guizhou Plateau, is the sixth largest freshwater lake in China. It is the main water area for irrigation in Kunming area, and it is also the basic condition for urban development. Therefore, Dianchi Lake is chosen as the research area. Because the in-situ monitoring of lake water quality for this study area was started late, there are few historical SD data. So, it is necessary to simulate and estimate SD by remote sensing images. In this paper, deep neural network algorithm was used to invert the SD of Dianchi Lake from January 1, 2001 to December 31, 2018 based on in-situ data and MODIS remote sensing images, and explored its spatial-temporal variation characteristics by geospatial analysis. The in-situ data was the daily monitoring values of 10 stations (Baiyukou, Caohai Center, Dianchi South, Duanqiao, Guanyinshan East, Guanyinshan West, Guanyinshan Center, Haikou West, Huiwan Center, and Luojiaying) from 2001 to 2010 and the monthly monitoring values of 2 regions (Caohai and Waihai) from 2018, which was provided by Yunnan Academy of Environmental Sciences; MODIS data was provided by NASA LAADS Web, with daily time resolution and 500 m spatial resolution. The methods used in this paper included Grey Relational Analysis (GRA), Long-Short Term Memory (LSTM), and Theil-Sen slope estimation. In the construction of SD inversion estimation model, the main methods include Correlation Analysis (C), Outlier Processing (O), Denoising Processing (D), and estimation model (COD-LSTM). The results showed that (1) the inversion model proposed in this paper had better performance (RMSE=0.1359, MAE=0.1134), which can objectively reflect SD status. (2) The analysis of temporal variation characteristics indicated that the SD of Dianchi Lake presented a downward trend with an average comprehensive change rate of -0.08 m/10 a. (3) The analysis of spatial variation characteristics indicated that the rate of decline in the region with higher SD was larger, and the trend in the region with lower SD was relatively stable. The SD near the urban and residential areas was relatively low. Human activities would become an important factor affecting the SD changes in Dianchi Lake, and it was also the main factor causing lake pollution.