Historical Retrospect of Aquatic Microbial Communities\nBased on Water–Carbonate Equilibrium and pH Values in Dianchi\nLake Using the Random Forest Algorithm

Aquatic\nmicrobial communities are crucial for the dynamics of algal\nblooms. Due to the limited development history and high cost of high-throughput\nsequencing, microbial information is commonly unavailable in past\naquatic quality surveys. This study comprehensively analyzed 19 years\nof historical data from alkalitrophic Dianchi Lake to explore the\ninterrelationships among pH, carbonate equilibrium (CO<sub>3</sub><sup>2–</sup> and HCO<sub>3</sub><sup>–</sup>), and\nphysicochemical/microbial variables, and developed predictive models\nof carbonate equilibrium and pH using the Random Forest (RF) algorithm.\nMicrobial taxa, particularly Proteobacteria, Cyanobacteria, and Bacteroidetes,\nwere more predictive of pH and carbonate equilibrium than physicochemical\nvariables. The Mean Absolute Percentage Error (MAPE) was 1.5%, 26.9%,\nand 30.1% for pH, CO<sub>3</sub><sup>2–</sup>, and HCO<sub>3</sub><sup>–</sup>, respectively. Different microorganisms\nplayed predominant roles across varying pH ranges, with photoautotrophs\n(e.g., Cyanobacteria) determining carbonate equilibrium at pH <\n9.0 and heterotrophs (e.g., Proteobacteria and Bacteroidetes) at pH\n> 9.0. Furthermore, a backtracking algorithm was used to reconstruct\nthe historical microbial community structure based on pH values, with\nthe backtracked cyanobacterial abundance matching well with historical\nchlorophyll-a data from 2000 to 2018. Our results highlight the strong\ncorrelations between microbial community structure, aquatic pH, and\ncarbonate equilibrium and provide a powerful backtracking algorithm\nto obtain historical microbial information for lake management.