In this project I will collaborate with Global Fishing Watch to detect fishing activity in the ocean using data from satellite Automatic Identification System AIS collected from different vessels around the world. The AIS data contains the latitude, longitude, speed and course of the vessels at different times.

Overfishing and illegal fishing are becoming big problems around the world. For example, there are many records of intensive and often illegal fishing of West Africa’s waters by Asian and European fleets that reduce the regular catch for the local populations, increasing their poverty levels. “Being able to see which vessels are fishing where would be a tremendous help in reducing illegal fishing,” says Josephus Mamie, head of Sierra Leone’s Fisheries Research Unit. [1]

Global Fishing Watch (GFW) is an organization that analyzes data from the Automatic Identification System (AIS), which is collected by satellites and terrestrial receivers, to identify apparent fishing behavior based on the movement of vessels over time.

AIS system was put in place to guarantee the safety of vessels, it provides collision avoidance and allow maritime authorities to track and monitor vessel movements. Each vessel periodically reports information including the vessel’s identity, type, position, course, speed, navigational status and other safety-related information.

Vessels fitted with AIS transceivers can be tracked by AIS base stations located along coast lines or, when out of range of terrestrial networks, through a growing number of satellites that are fitted with special AIS receivers and manage to map the position of the vessels [2].

20M of data points are produced per day this way. In this project we will have data that has been already labeled by different experts and using crowd sourcing methods.

The data set was kindly made available by David Kroodsma and Timothy Hochberg from SkyTruth.

Notes on the data:

The data features include the type of vessel (troller, drifting long lines, fixed gear) and time series of position, speed, and course of each vessel. The intervals of data acquisition is not constant. There may be gaps when the vessels are not on the satellite’s field of “view”. It takes 90 mins for a low orbit satellite to go around the globe, so we can expect gaps of up to 1 hour.

When there is good coverage the time intervals still vary from 2 seconds to 2 minutes, due to differences in satellite communications around the world and signal interference.

[1] http://www.ipsnews.net/2016/09/new-public-website-offers-detailed-view-of-industrial-fishing/ [2] ExactEarth Technologies, https://en.wikipedia.org/wiki/Automatic_identification_system

• data/labeled - This folder has a sample file: "alex_crowd_sourced_Purse_seines.measures.labels.npz", please copy the rest of the data files from GFW training data here.

• images - Images used on the web app

• results:

  • best_reconstructions: Images of the recostructed tracks superimposed to maps.
  • final_models: Best performing models, you can find the pickle files here on the '.pkl' files, as well as a summary of their performance on the '.txt' files.
  • models_performance.png

• src - The scripts used in the analysis

  • grid_search_multicolumns.py

    Grid search for each model specified on the models_dict dictionary. It takes 3 arguments, run it in the console like this:

    python grid_search_multicolumns.py model_number model_type gear_type

    The possible values for the arguments are:

    Argument	values
    model_number	['model_1','model_2','model_3','model_4','model_5','model_6']
    model_type	['RF','GBC']
    gear_type	['longliners','trawlers','purse_seines']

    For example, to run the grid search on a Random Forest (RF), for model_1 and the longliners, you should write:

    python grid_search_multicolumns.py model_1 RF longliners

  • reconstructing_tracks.py For a given mmsi (vessel identification number) this script will generate two images of the tracks, one indicationg the actual fishing sites (in green) and another images with the predicted fishing sites (in blue.) The images then will be kept on ../results/best_reconstructions

  • results_analysis.py Produces a graph of the F1-scores for the different models and stores it on ../results