Prediction of algal blooms can be improved with the use of satellite data, as mathematician Dr. Joanna Pelc showed in her PhD-research.
A bay, green as a meadow, slow waves covered with slime or knee-deep foam layers on the beach. Algal blooms always seem to pop up overnight. Partly, that’s because predicting them is particularly difficult.
A generic ecological model called Bloom/Gem was developed to calculate algal growth in the Southern North Sea by researchers from TU and Deltares. The model describes amongst others: hydrodynamics in the North Sea, floating sediment and river affluent as well as nutrient cycles and competition between algae species. In all it has some 250 parameters, but even that elaborate set is not considered complete. In an article in the scientific paper Ecological Modelling (2010), the authors state that “real situations are often far more complex than any model can represent.” On the bright side, they also suggest the model’s performance may be enhanced by adding more temporal and spatial data.
This process of feeding observations into a computer model of a real system is called data assimilation. It has applications in geosciences, weather forecasting and hydrology. The process runs in cycles in which forecast and current system state are compared to produce an analysis. This analysis aims to balance the uncertainty in the data and the forecast. In the next step, the model advances in time and its result becomes the forecast in the next analysis cycle.
You could say that data assimilation is a model’s recurring reality check. For simpler systems, such corrections may, and are indeed done, by hand. But as the complexity and the area covered grow, data assimilation needs to be automated.
Modeling algal growth is complicated by the number of parameters involved. The general term chlorophyl covers several different species of algae such as diatoms, flagellates and dinoflagellates, all of which have their own growth cycle. These grow cycles in turn depend on nutrients, oxygen, light availability and grazing (being eaten).
One way to provide estimates for parameters is by using a variational model, says Dr. Joanna Pelc. A variational model is a type of data assimilation that has been applied successfully in meteorology. For ecological models, the application thus far is less fruitful because the complex ecological models often behave non-linearly. In plain English: an eco-system may suddenly overreact.
Pelc’s variational model of choice, called 4D-Var, has the benefit of reducing the model to a much smaller size, making it easier to control.
Leaving all the complex matrix calculus aside, let’s just say that Pelc demonstrated the application of 4D-Var reduction to the aforementioned Bloom/Gem model. Together with the marine biology experts, she brought back the reduced the number of parameters from 250 via 70 to approximately 20.
For an demonstration run of the reduced model, the number of parameters was even further reduced to two: the extinction of visible light in the water and algal growth rate. The runs (of one week and two months respectively) were chosen as to include the algal spring bloom. The results of the demo were encouraging: over a longer period of two years’ time, the prediction of the amount of cholorophyl (algal biomass) improved with 35 percent on average.
A final test of the reduced model against two years of real data from the MERIS*-spectrometer on the Envisat satellite showed that the reduced model (with assimilation of MERIS-data) performed up to 10 percent better than the original Bloom/Gem model. Which is remarkable, says Pelc, because this model has been calibrated many times before.
She therefore concludes that the reduced model in combination with remotely sensed data will improve predictions of algal blooms.
Joanna S. Pelc, Data Assimilation for Marine Ecosystem Models, 16 July 2013, PhD supervisor Prof. Arnold Heemink (EWI)
* MERIS stands for Medium Resolution Imaging Spectrometer
Comments are closed.