Landsat data reveal lake deoxygenation worldwide

Area of Science:

• Environmental Science
• Ecology
• Remote Sensing

Background:

• Dissolved oxygen (DO) is vital for aquatic life and ecosystem health.
• Long-term global lake DO trends are poorly understood due to data limitations.

Purpose of the Study:

• To develop and apply machine learning models for estimating long-term dissolved oxygen in global lakes.
• To analyze spatial and temporal patterns of DO changes and identify areas under stress.

Main Methods:

• Integration of Landsat data and geographic features to build DO estimation models.
• Utilized machine learning algorithms including Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine Regression (SVR), and Extreme Gradient Boosting (XGBoost).
• Applied the best-performing RF model to analyze DO trends in over 350,000 global lakes from 1984-2021.

Main Results:

• The RF model achieved superior performance (R² = 0.72, RMSE = 1.24 mg/L) compared to other tested models.
• Average epilimnetic DO was 9.72 ± 1.07 mg/L, with higher concentrations in polar and lower in equatorial regions.
• A widespread deoxygenation trend of -0.036 mg/L per decade was observed, with increasing lake area experiencing DO stress.

Conclusions:

• Machine learning, particularly RF, effectively estimates global lake DO using remote sensing data.
• Global lakes are experiencing significant deoxygenation, posing risks to aquatic ecosystems.
• The findings provide crucial data for aquatic ecosystem management and conservation efforts.