Machine Learning Approaches Reveal Future Harmful Algae Blooms in Jeju, Korea

Abstract

ABSTRACTCyanobacterial algae blooms have proven to suppress diversity and abundance of other organisms while previous research shows the direct correlation between the growth of cyanobacteria and increasing global temperatures. Freshwater temperatures in Jeju island are most prone to climate change within the Korean peninsula, but research on Harmful Algae Blooms (HABs) in these environments has been scarcely conducted. The purpose of this study is to predict the cell numbers of the four HAB species in Jeju island’s four water supply sources in 2050 and 2100. Using the water quality data across the last 24 years, Scikit-learn GBM was developed to predict cell numbers of HAB based on four variables determined through multiple linear regression: temperature, pH, EC, and DO. Meanwhile, XGBoost was designed to predict four different levels of HAB bloom warnings. Future freshwater temperature was obtained through the linear relationship model between air and freshwater temperature. The performances of the Scikit-learn GBM on the cell numbers of each species were as follows (measured by MAE and R2): Microcystis (132.313; 0.857), Anabaena (36.567; 0.035), Oscillatoria (24.213; 0.672), and Apahnizomenon (65.716; 0.506). This model predicted that Oscillatoria would increase by 31.04% until 2100 and the total cell number of the four algeas would increase 376,414/ml until 2050 and reach 393,873/ml in 2100 (247.088; 0.617). The XGboost model predicted a 17% increase in the ‘Warning’ level of the Algae Alert System until 2100. The increase in HABs will ultimately lead to agricultural setbacks throughout Jeju; algae blooms in dams will produce neurotoxins and hapatotoxins, limiting the usage of agricultural water. Immediate solutions are required to suppress the growth rate of algae cells brought by global climate change in Jeju freshwaters.