Python app to build jellyfishes risk forecastings.

It works calibrating an ensemble of models using historical data and then using this model to predict the presence of jellyfishes in future dates.

The climatological and biological data, both historical and predictions, is downloaded from MyOcean. The historical jellyfish presence data is provided by the Maritimal Sciences Institute (ICM).

Data processing and modelling is performed using R and the package biomod2. The rest of the system is written in Python.

The detailed process is as follows:

• Download historical data from MyOcean using motu-client, restricting the area to the region containing the beaches of interest and minimal depth.
• Extract the value of each climatological variable at each beach, for all days, using R.
• Combine the values for all climatological variables, beaches and days in one single table.
• Add the historical Pelagia jellyfish presence. In our case this comes from the Institute of Marine Sciences.
• Use this historical table to train an ensemble of models using the R package biomod2.
• Download the forecast for climatological data for the area of interest from MyOcean using motu-client.
• Extract the climatological variables forecasting values for the beaches of interest.
• Apply the trained model to the new data to predict the presence of Pelagia jellyfish in a future day.

This instructions are tested in Ubuntu 10.04.4 LTS.

Install system requirements:

$ sudo apt-get install libgdal-dev libproj-dev

Add to /etc/apt/sources.list:

$ deb http://ftp.cixug.es/CRAN/bin/linux/ubuntu lucid/

Add pubkey for this PPA: $ sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys E084DAB9

Install biomod2 in R: $ R

install.packages('biomod2') install.packages('FNN') install.packages('raster') install.packages('rgdal') install.packages('nlme') install.packages('Matrix') install.packages('mgcv') install.packages('ncdf')

Create a virtualenv, activate it and do:

$ pip install -r requirements.txt

Clone our motu-client-python fork modified to work in python 2.7.x:

git clone [email protected]:Science-o-matic/motu-client-python.git

Copy the settings template file and modify to apply your settings:

$ cp jellyrisk_forecaster/settings.py.tmpl jellyrisk_forecaster/settings.py

To run the scripts, please set the environment variable JELLYRISK_SETTINGS_MODULE pointing to an accessible Python module with all the settings set:

• CARTODB_API_KEY
• CARTODB_DOMAIN
• TEMP_FOLDER: folder to chdir to before executing the R script
• DATA_FOLDER: folder with data files
• Any celery-related config var. `app.config_from_object(JELLYRISK_SETTINGS_MODULE)' will be invoked.

Hint: Use virtualenv's postactivate and postdeactivate hooks.

You can run a prediction with:

$ jellyrisk_forecaster/predict_ahead.py

This will download MyOcean data, run R to generate a forecast for next 2 days, and send it to CartoDB maps so they're updated.