Neural network modelling of Baltic zooplankton abundances | Emodnet Biology

Dear User, the EMODnet Biology portal will soon be integrated into our new EMODnet Central Portal, which will provide continued, open and free access to our data and information services with unique new features. By providing a Centralised EMODnet service, we aim to help you optimise your research and work.

You can preview our new unified content, document search facilities and unified data catalogue via the EMODnet Product Catalogue
The new and optimised common map viewer will be launched in November 2022.

Follow us on Twitter (@EMODnet and @EUROBIS_VLIZ) for the latest news on our centralisation process and new features.

For any questions or comments, you can contact us via .

Neural network modelling of Baltic zooplankton abundances

Data Series

The input data is a combination of different zooplankton datasets linked with EMODnet Biology, and EMODnet or external environmental layers.

  • Zooplankton observations for 40 different species from the Baltic from
    • the Swedish SHARK database from EMODnet Biology:
    • the Finnish data from the NOAA Copepod database: (download available here)
    • the German and Polish from the HELCOM DOME database:

All these dataset will be integrated in the EMODnet Biology database


Data Product

The data product is a gridded data product for 40 zooplankton species using DIVAnd (Barth et al., 2014; Troupin et al., 2012) and the neural network library Knet (Yuret, 2016). The neural network uses the variables dissolved oxygen, salinity, temperature, chlorophyll concentration bathymetry and the distance from coast as input. Additionally the position (latitude and longitude) and the year are provided to the neural network.

For example, the maps below show the abundance of Acartia (Acartiura) longiremis (see species description in WoRMS) in the Baltic Sea from 2007 to 2013. A subtle decline can be observed in the Kattegat. The color in the small circles represent actual observations, so circles that blend in the surrounding color means that the result of the neural network interpolation was successful.

Maps showing the abundance of Acartia (Acartiura) longiremis  in the Baltic Sea from 2007 to 2013

Map showing the abundance of Acartia longiremis in the Baltic in 2007Map showing the abundance of Acartia longiremis in the Baltic in 2008

Map showing the abundance of Acartia longiremis in the Baltic in 2009Map showing the abundance of Acartia longiremis in the Baltic in 2011

Map showing the abundance of Acartia longiremis in the Baltic in 2012Map showing the abundance of Acartia longiremis in the Baltic in 2013

Abundance values in the NetCDF files are expressed in number per m² and transformed by the function log(x/a + 1) where a is 1 m⁻².

The covers the area from 9°E to 30.8°E and 53°N to 66.1°N at a resolution of a tenth of a degree. Gridded data product for the years 2007, 2008, 2010, 2011, 2012 and 2013 have been made. No observations were available for the year 2009. The fields represent the yearly average abundance.

For every species the correlation length and signal to noise ratio is estimated using the spatial variability of the observations.

The interpolation error is computed using the so-called "clever poor man's error" (Beckers et al., 2014).

The full list of the species is:

Acartia (Acanthacartia) bifilosa

Acartia (Acanthacartia) tonsa

Acartia (Acartiura) clausi

Acartia (Acartiura) longiremis

Amphibalanus improvisus



Bosmina (Eubosmina) coregoni


Calanus finmarchicus


Centropages hamatus

Cercopagis (Cercopagis) pengoi




Daphnia cristata



Evadne nordmanni

Fritillaria borealis



Keratella cochlearis

Keratella cruciformis

Keratella eichwaldi

Keratella quadrata

Limnocalanus macrurus macrurus





Pleopis polyphemoides

Podon intermedius

Podon leuckartii





Temora longicornis

More Information


A. Barth, J.-M. Beckers, C. Troupin, A. Alvera-Azcárate, and L. Vandenbulcke. divand-1.0: n-dimensional variational data analysis for ocean observations. Geoscientific Model Development, 7(1):225–241, 2014.

C. Troupin, A. Barth, D. Sirjacobs, M. Ouberdous, J.-M. Brankart, P. Brasseur, M. Rixen, A. Alvera-Azcárate, M. Belounis, A. Capet, F. Lenartz, M.-E. Toussaint, and J.-M. Beckers. Generation of analysis and consistent error fields using the Data Interpolating Variational Analysis (DIVA). Ocean Modelling, 52–53:90–101, 2012.

Yuret, D. Knet: beginning deep learning with 100 lines of julia. In Machine Learning Systems Workshop at NIPS 2016.

J.-M. Beckers, A. Barth, C. Troupin, and A. Alvera-Azcárate. Approximate and efficient methods to assess error fields in spatial gridding with data interpolating variational analysis (DIVA). Journal of Atmospheric and Oceanic Technology, 31:515–530, 2014.

Code and methodology

Citation and download link

Barth, A.; Herman, P.M.J.; (2018): Neural network modelling of Baltic zooplankton abundances. Marine Data Archive.


Biological data preprocessing by Peter M.J. Herman, Neural network modelling by Alexander Barth



Website hosted and developed by VLIZ