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

Dissolved inorganic carbon (DIC) is defined as a compound group with particle sizes smaller than 0.45 μm, which often constitutes over 70% of the total carbon in lakes. Monitoring DIC concentrations through remote sensing is imperative for dynamically assessing lake carbon stocks and for the conservation and management of lake ecosystems. However, the complex optical properties of lake water, in conjunction with the variability of factors influencing DIC concentrations across different lakes, have impeded the advancement of research in retrieving DIC concentrations from satellite data. Utilizing in-situ DIC data of 135 lakes in China (N = 1386), we developed a two-step Random Forest algorithm designed for large-scale and synchronous remote sensing of DIC concentrations across different lakes. The algorithm achieved a mean absolute percentage error of 15.82%. Subsequently, the novel algorithm was applied to Landsat series imagery, enabling the retrieval of annual mean DIC concentrations at the pixel scale for 24,366 lakes during 1984-2023. Across China, the spatial distribution of lake DIC concentrations generally follows a pattern of "higher values in the northwest and lower values in the southeast". According to the Hu-Huanyong line, the average lake DIC concentration in northwest was 117.67 ± 91.27 mg/L, while in the southeast, it was 29.89 ± 57.15 mg/L. Furthermore, a significant positive correlation was observed between DIC concentration and conductivity (r = 0.72, p < 0.01), indicating that conductivity plays a key role in characterizing the distribution of DIC concentrations. The spatial variation in lake DIC concentration was primarily attributed to precipitation, evaporation, and runoff closely related to salinization, accounting for 53.84% of the variation. Finally, based on the Landsat-derived DIC concentration and water storage, we estimated the total DIC storage of all the 24,366 Chinese lakes in 2015 to be 1.46 Pg C, with 85.89% of this stored in the Qinghai-Tibet Plateau. In sum, this study provides valuable insights into the dynamics and driving factors of DIC concentrations in lakes across China in the context of global change.