CoastObs uses Earth Observation and validated in-situ data to provide harmful algal blooms forecasts in Europe.

Why is it important?

Harmful Algal Blooms (HABs) are considered one of the most dangerous threats to coastal ecosystems worldwide in terms of biodiversity preservation and food security. HABs forecasting models integrating Earth Observation (EO) data and other environmental parameters can contribute to a better management of aquaculture activities to mitigate the HABs impact. Forecasting tools are of great interest to monitoring authorities and aquaculture producers.

What does CoastObs offer?

CoastObs provides a service based on a set of advanced and basic models to forecast Pseudonitzschia spp. blooms in the Rias Baixas (Galicia). Model output is defined as the probability of a toxic bloom causing the closure of a production area (Amnesic Shellfish Poisoning or ASP closure) in a ria between 1 and 5 days after the modelling date. Basic models use as input upwelling indices (freely available from NOAA or IEO) as well as temperature and chlorophyll concentrations from Sentinel-3 images. Advanced models incorporate also nutrient concentrations, i.e. nitrate, nitrite, phosphate and silicate.

How was the data validated?

Forecasting models for Galicia were developed and validated using historical data (including ASP closures) provided by the monitoring centre (INTECMAR) as well as data from MERIS (2002-2012) and Sentinel 3 (2016-2018). Models performance was evaluated as a binary classification (bloom or no bloom). Overall, models show sensitivity (% of bloom correctly classified) values over 70%. Advance models show also good specificity (% of no bloom correctly classified) values over 80%, while basic models show a lower specificity and hence higher false alarm rates (see complete results in D3.8).

Case study example: Pseudo-nitzschia spp. bloom forecasting in Galicia

Figure shows the results of the advanced forecasting model at day +3 for the Ria de Muros between 2004 and 2005. Overall, the model is able to follow the pattern observed in the insitu database with only one false positive.

Figure 1. ASP closure probability observed (0: closed; 1: opened) and predicted using the advanced mode (day +3) for the Ria de Muros between 2004 and 2005.

Required input

Advanced models require nutrient concentrations. Nowadays, these data are not operationally available from the monitoring program or other public services in Galicia. However, they could be available in other areas or be provided by potential users.


  • Models with a higher sensitivity tend to produce more false alarms due to the unequal distribution of both classes in the input dataset (~ 85 % of no bloom).