Towards statistical modeling of chlorophyll-a concentrations in Balikpapan Bay, Indonesia: Implications for algal bloom detection

Abstract This study presents a comprehensive statistical analysis of chlorophyll-a dynamics in Balikpapan Bay, Indonesia, combining time series analysis, extreme value modeling, and machine learning techniques to understand phytoplankton variability near Indonesia’s planned new capital city. Analysis of daily chlorophyll-a concentrations (2019–2021) revealed a non-Gaussian distribution (skewness = 2.212, kurtosis = 10.160) ranging from 0.350 to 11.270 mg m −3 , with distinct seasonal patterns showing May maxima (2.110 mg m −3 ) and July minima (1.340 mg m −3 ). Block Maxima extreme value analysis identified 79 extreme events and fitted a Gumbel distribution (location μ = 2.924, scale σ = 1.231, log-likelihood = −141.644), though notably failed to capture two major HAB events (11.270 and 10.430 mg m −3 ). A WeightedEnsemble_L2 model combining ExtraTreesMSE (0.462), CatBoost (0.346), and LightGBMXT (0.192) identified temperature (importance: 0.072, p < 0.001), solar radiation (0.061, p = 0.002), and phosphate (0.047, p < 0.001) as key drivers, achieving moderate performance (RMSE = 0.868 mg m −3 , R 2 = 0.204). The trained model was serialized for potential operational deployment, providing crucial baseline data for harmful algal bloom (HAB) monitoring systems in this rapidly developing coastal region.