Projecting harmful algal blooms in North Biscayne Bay using machine learning and bias-corrected climate scenarios

Zhengxiao Yan, Nasrin Alamdari,
Projecting harmful algal blooms in North Biscayne Bay using machine learning and bias-corrected climate scenarios,
Journal of Hydrology: Regional Studies,
Volume 62,
2025,
102865,
ISSN 2214-5818,
https://doi.org/10.1016/j.ejrh.2025.102865.
(https://www.sciencedirect.com/science/article/pii/S2214581825006949)
Abstract: Study area
North Biscayne Bay, downstream of urbanized Miami-Dade County, is an ecological hot spot and frequently impacted by harmful algal blooms (HABs).
Study focus
This study combines machine learning models with bias-corrected climate projections to forecast future chlorophyll-a concentrations, a HAB indicator, in North Biscayne Bay from 2021 to 2100. A Support Vector Machine (SVM) model was developed using historical water quality and climate data, then applied to projections from four global climate models under three Shared Socioeconomic Pathways (SSP126, SSP245, SSP585).
New hydrological insights for the region
Results show a steady increase in chlorophyll-a concentrations in all scenarios with SSP126 showing the largest increase at 0.0045 µg/L per year compared to SSP245 at 0.0007 µg/L per year and SSP585 at 0.0005 µg/L per year. SSP126 predicts the most extreme chlorophyll-a events at 36–82 exceedances followed by SSP245 at 32–65 exceedances and SSP585 at 24–54 exceedances. The three most important positive drivers of chlorophyll-a were precipitation, upstream discharge and minimum temperature while wind speed was negative. Data shows chlorophyll-a concentrations were higher during the wet season than the dry season and grew more during the long-term (2071–2100) than the short-term (2021–2050). These projections assume stationarity in key water quality drivers, which may not fully hold under changing land use or management conditions, but they nonetheless provide valuable guidance for anticipating future trends. These findings highlight the value of using climate-adjusted data and interpretable machine learning to forecast future water quality issues in urban coastal areas, supporting predictive tools that can help prioritize wet-season monitoring and early warning strategies in urban estuaries.
Keywords: Harmful algal blooms; Machine learning; SHAP; Seasonal trends; North Biscayne Bay; Climate change