Probabilistic Harmful Algal Bloom Occurrence Dataset
Harmful algal blooms (HABs) represent a major challenge for marine ecosystems, human health, food safety, and shellfish aquaculture. As ocean conditions change, understanding where and when harmful algae are more likely to occur is increasingly important for monitoring, risk assessment and early-warning strategies. However, conventional HAB monitoring programs based on in situ water sampling are often limited by sparse spatial coverage and low temporal frequency, making it difficult to capture the full spatial extent and temporal variability of bloom events. To help address this challenge, the Horizon Europe FOCCUS Project has supported the development of a probabilistic dataset that combines long-term field observations of harmful algae with satellite-derived and ocean reanalysis environmental products and machine learning.
Developed by the Nansen Environmental and Remote Sensing Center (NERSC), the dataset provides weekly maps of HAB occurrence probabilities for key algal taxa along the Norwegian coast (see Figure 1). The product is distributed in NetCDF format on a 4 km grid and includes four harmful algal taxa: Alexandrium spp., the Alexandrium tamarense group, Dinophysis acuta and Azadinium spinosum.
From long-term observations to regional probability maps
The models were developed using weekly algae observations collected between 2006 and 2019 through the Norwegian Food Safety Authority monitoring program at 35 shellfish farming locations. At each site, water samples collected from 0–3 m depth were analyzed to identify the presence and abundance of key HAB taxa. These biological observations were combined with four environmental variables representing the physical and biological conditions influencing algal growth and distribution, including sea surface temperature, mixed layer depth, sea surface salinity, and photosynthetically active radiation. The environmental information was obtained from a combination of satellite products and the Copernicus Marine TOPAZ modelling system.
A supervised machine-learning approach based on support vector machines (SVM) was used to estimate HAB occurrence probabilities under different environmental conditions. The trained models were then applied to satellite observations and ocean model outputs to generate continuous coastal HAB probability maps. The methodology is described in Silva et al. (2024a), the trained SVM models are openly available through the associated data repository (Silva, 2024b), and the probability maps from 2000 to 2024 are public available (Silva, 2026).
Figure 1. Annual mean probability of HAB occurrence in 2024 for four target algae taxa along the Norwegian coast.
Independently validated performance
The models were tested against an independent set of observations collected between 2014 and 2019. The results showed strong agreement between predicted probabilities and observed HAB events, with particularly high performance for Alexandrium spp., the Alexandrium tamarense group and Azadinium spinosum, while Dinophysis acuta also showed good predictive skill. The validation indicates that the dataset can effectively identify periods and areas with an elevated likelihood of HAB occurrence, supporting more targeted monitoring and earlier, better-informed responses to potential bloom events.
Supporting aquaculture, monitoring and coastal management
The dataset can help shellfish farmers, fisheries managers, environmental agencies and researchers:
- plan monitoring activities in space and time;
- target sampling and operational resources;
- investigate environmental drivers of HAB development;
- assess HAB risk in coastal modelling and forecasting systems;
- support future Copernicus coastal services and early-warning applications.
Within FOCCUS, the product contributes directly to coastal ecosystem protection, sustainable aquaculture and the wider blue economy by transforming point-based observations into accessible, regional-scale coastal intelligence.
Limitations and appropriate use
The 4 km resolution means that narrow fjords and island-rich coastal areas may be insufficiently represented or contain gaps. The models also do not include all ecological processes influencing HAB occurrence, such as nutrient availability, prey abundance and grazing pressure.
The dataset is suitable for research, environmental analysis and identifying areas of increased HAB risk. It should not replace official food-safety monitoring or be used to determine whether shellfish are safe for consumption. Decisions related to harvesting and public health should continue to follow the guidance of the Norwegian Food Safety Authority.
By integrating long-term biological observations, Earth Observation, Copernicus Marine model products and machine learning, this dataset supports the FOCCUS vision of more effective and actionable coastal monitoring.
Further information:
Silva, E. (2026). Probability detection of harmful algae blooms [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21389857
Silva, E., 2024b. nansencenter/Probabilistic-Models-for-Harmful-Algae--Application-to-the-Norwegian-Coast: v1.0.1 (Version v1.0.1) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.10958487
Silva, E., Brajard, J., Counillon, F., Pettersson, L.H., Naustvoll, L., 2024a. Probabilistic models for harmful algae: application to the Norwegian coast. Environmental Data Science, 3, e12. https://doi.org/10.1017/eds.2024.11.
Contact: Edson Silva (edson.silva[@]nersc.no)
