AlgAlert: A two-level approach for algae bloom prediction using deep learning
Chlorophyll-a (Chl-a) is essential to detect harmful algae blooms that can damage aquatic ecosystems and cause economic losses. Consequently, governmental agencies and research institutions invest significant effort into monitoring water quality and developing management strategies for aquatic syste...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125002699 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Chlorophyll-a (Chl-a) is essential to detect harmful algae blooms that can damage aquatic ecosystems and cause economic losses. Consequently, governmental agencies and research institutions invest significant effort into monitoring water quality and developing management strategies for aquatic systems. With the increasing availability of real-time water quality, meteorological and tidal sensor data, there is growing potential to harness this information through data-driven approaches such as machine learning to support aquatic systems management. This study presents a comprehensive data-driven framework named AlgaAlert that integrates a regression model to forecast Chl-a concentrations and a classification model to predict the occurrence of blooms in a temperate estuarine system. The framework was developed by benchmarking multiple algorithms and selecting the best-performing regression and classification models for integration. The model evaluation was based on hourly water quality and meteorological data collected from early December 2019 to mid-January 2020 from the Kwilena monitoring site, the South Perth meteorological station, and a tidal gauge on Barrack Street, Perth, Australia. The AlgAlert framework combines K-Nearest-Neighbours Regression (KNN) regression to predict Chl-a levels with a custom classifier to determine bloom or no-bloom conditions based on labelled time-series data. KNN demonstrated superior regression performance, achieving 0.25 MAE, outperforming other models like random forest (RF). Classification results revealed nearly perfect F1-scores, indicating that the model accurately identified bloom events with few missed or false alarms (0.99 for no-bloom and 0.98 for bloom). This demonstrates AlgAlert’s robust predictive capabilities, offering a reliable tool to support timely decision-making in water quality management. |
|---|---|
| ISSN: | 1574-9541 |