Inversion of lake transparency using remote sensing and deep hybrid recurrent models

Area of Science:

• Environmental Science
• Remote Sensing
• Water Quality Monitoring

Background:

• Lake ecological research faces challenges in accurately measuring water transparency due to complex optical properties and limited sampling.
• Existing simplified models struggle with large-scale application and capturing intricate optical characteristics of lake water bodies.

Purpose of the Study:

• To develop an accurate and efficient lake water transparency inversion model (WTIM) using advanced deep learning techniques.
• To enable rapid, large-scale, and automated remote sensing-based inversion of lake transparency.

Main Methods:

• Integration of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) network structures.
• Utilized Landsat-8 remote sensing data, field measurements, and simulated data to train and validate the WTIM model.

Main Results:

• The WTIM model achieved high accuracy in inverting lake water transparency (R²=0.78, MAE=0.64, RMSE=0.84).
• Analysis revealed a decreasing trend in Chinese lake transparency from 2014-2021.
• Increasing transparency in the Qinghai-Tibet Plateau contrasts with decreasing transparency in eastern and northeastern plains.

Conclusions:

• The WTIM model offers a robust solution for large-scale lake transparency monitoring.
• Global warming impacts Qinghai-Tibet Plateau lake transparency, while industrial and agricultural activities affect plains lakes.
• The findings provide valuable insights for lake transparency inversion and environmental management.