A two-step Random Forest algorithm for deriving dissolved inorganic carbon in lakes from Landsat satellite data

Area of Science:

• Environmental Science
• Remote Sensing
• Limnology

Background:

• Dissolved inorganic carbon (DIC) is a major component of lake carbon, vital for ecosystem health.
• Accurate monitoring of lake DIC is essential for carbon stock assessment and lake management.
• Previous remote sensing efforts for DIC retrieval were limited by water optical complexity and influencing factors.

Purpose of the Study:

• To develop a robust algorithm for large-scale, synchronous remote sensing of DIC concentrations in lakes.
• To map the spatial distribution and temporal dynamics of lake DIC concentrations across China.
• To identify key environmental drivers influencing DIC concentration variations in lakes.

Main Methods:

• Utilized in-situ DIC data from 135 lakes in China (N=1386) to develop a two-step Random Forest algorithm.
• Applied the validated algorithm to Landsat satellite imagery for pixel-scale DIC retrieval in 24,366 lakes (1984-2023).
• Analyzed spatial patterns, correlations with conductivity, and attribution of variations to climatic and hydrological factors.

Main Results:

• The Random Forest algorithm achieved a mean absolute percentage error of 15.82% for DIC retrieval.
• Lake DIC concentrations exhibited a 'higher in the northwest, lower in the southeast' pattern across China.
• Conductivity showed a significant positive correlation (r=0.72) with DIC concentration; precipitation, evaporation, and runoff explained 53.84% of DIC variation.

Conclusions:

• The developed algorithm provides a novel approach for large-scale remote sensing of lake DIC.
• Spatial variations in lake DIC are significantly influenced by hydrological processes and salinization.
• Total estimated DIC storage in Chinese lakes was 1.46 Pg C in 2015, with most on the Qinghai-Tibet Plateau.