Harmful algal bloom prediction modeling in the face of a changing climate

Author

Jordan E. Spears, University of KentuckyFollow

Date Available

8-18-2025

Year of Publication

2025

Document Type

Master's Thesis

Degree Name

Master of Science in Civil Engineering (MSCE)

College

Engineering

Department/School/Program

Civil Engineering

Faculty

Dr. James Fox

Abstract

Harmful algal blooms (HABs) in rivers pose significant threats to drinking water quality and public health, yet the predictive understanding of their proliferation remains limited, particularly regarding the influence of flow conditions in large rivers. This thesis addresses the critical knowledge gap of the relationship between seasonal flow variability and HAB proliferation in the Ohio River, with the Markland Pool serving as the primary study site. This study quantified the impact of early summer high flow events (>7500 m³ s^-1) and late summer to early fall low flow conditions (< 1500 m³ s^-1) by integrating observational sensor discharge data over 40 years (1983-2022) with advanced machine learning models, notably artificial neural network with a long short-term memory (ANN-LSTM) component.

The HAB prediction models contained herein achieved over 85% percent accuracy in identifying bloom occurrences revealing that flow conditions during specified periods can serve as reliable predictors. The HAB prediction model predicted five algal bloom occurrences in the years 1997, 1998, 2008, 2015, and 2019, of which 2008, 2015, and 2019 showed algal bloom occurrences in existing data. Results demonstrate that these early summer high flow events are strongly correlated with increased turbidity (> 100 FNU turbidity reading at sensor 176 km upstream) and sediment flux, which may also facilitate nutrient loading and algal seeding, while low flow periods promote clear water (< 2 FNU turbidity reading at sensor 176 km upstream) conducive to bloom proliferation. Further expansion of the model indicated that water velocity is relatively low and residence time is high, and these conditions may be additional critical factors influencing HAB development. The periods of low flow in the late summer to early fall corresponded to very low velocities of 0.62 m s-¹ and residence times of 2.8 days in the 153 km Markland pool, while peak water discharge events in the early summer period nearly double the low flow velocities with velocities of 1.0 m s-¹ and residence times of 1.8 days.

Results from ANN-LSTM modeled discharge showed accurate prediction of overall discharge trends (NSE=0.73; KGE=0.77; pBIAS=6.8) and bias correction to improve the model’s ability to capture flow events associated with HAB proliferation yielded similar or improved results for capturing overall trends (NSE=0.73; KGE=0.83; pBIAS=-1.0). The ANN LSTM modeled discharge was used to model discharge in the integration of 15 realizations of global climate models (GCMs) from the coupled model intercomparison project (i.e., CMIP3 and CMIP5). GCM modeled discharge provided valuable predictive capacity, allowing proactive assessment of HAB risks under future climate scenarios. GCM modeled discharge for the forecast period (2048-2065) showed an overall decrease in discharge from the hindcast period (1983-2000), most notably in May and June, where discharge decreased by 10% (>400 m³ s^-1) and 25% (>600 m³ s^-1), respectively. These changes are seen in the HAB prediction model, where the number of high flow events in the early summer show a 40% decrease, with 3.1 years out of 18 years in the hindcast period and 1.9 years out of 18 years in the forecast period. These decreases in discharge ultimately led to less HAB predictions in the forecast period, falling from an average across the 15 GCMs of 2.1 years out of 18 years in the hindcast period to 1.4 years out of 18 years in the forecast period. Despite these decreases, the late summer to early fall low flow events increase in the forecast period, from 14.9 years out of 18 years in the hindcast period to 16.4 years out of 18 years in the forecast. This result underscores the early summertime high flow event as the driver for HABs in the Ohio River.

These findings advance the predictive capacity for HAB events, enabling proactive water quality management. This study underscores the importance of hydrodynamic monitoring in large river systems for HAB risk mitigation, providing a framework applicable to other regulated rivers facing similar threats.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2025.426

Funding Information

This study was supported by the National Science Foundation's SUCCESS Project Fellowship as well as the United States Army Corps of Engineers in 2023.

This study was supported by the University of Kentucky College of Engineering's Lauderdale Fellowship in 2023 and 2024.

Recommended Citation

Spears, Jordan E., "Harmful algal bloom prediction modeling in the face of a changing climate" (2025). Theses and Dissertations--Civil Engineering. 157.
https://uknowledge.uky.edu/ce_etds/157