A Method for Predicting Fishing Activity Based on Geospatial Motion Behaviors - Summarized from an Analyze Technical Report


Illegal fishing is a significant economic and environmental challenge for countries around the world.  Up to 40% of fishing catch in certain parts of the world is unlawful or unregulated, resulting in approximately $10B to $20B in economic losses and significantly depleting international food stocks. 

Using geospatial position information, data scientists at Analyze provided a reliable method for characterizing fishing behaviors among ships on the high seas.  These methods have the potential to significantly improve interdiction of illegal fishing on the high seas.

Using data transmitted from the Automated Identification System, Analyze studied nearly 500,000,000 data points for 110,000 vessels.   They analyzed time-codes, vessel identity and motion data including:  navigational status, rate of turn, speed over ground, lat/long, true heading, true bearing and more.  The hypothesis was that unique motion behaviors could be associated with fishing activity using motion analytics.  For example circular and duplicative motions could indicate fishing behavior. 

Analyze research consisted in identifying and characterizing this unique motion behavior.  To accomplish this, we employed a basic “big data” analytics strategy consisting of data acquisition, data extraction, transformation and loading, data analysis using statistical and machine leaning approaches, predictive analytics and visualization.

Once the data set was identified for a specific geo-fenced area in the, Analyze utilized a number of analytics from the Mercury Motion Analytics Module that would aid in the discovery of motion behaviors including position information, boundaries & geocoding, distance, velocity, acceleration, motion primitives, shape conformance and consistency of motion.  Measures were derived from this data. 

We noticed that frequent and significant changes in the vessel's compass heading (erratic heading) and erratic changes in velocity were strong predictors of fishing activity.  The vessels themselves use a navigational status of 7 to self-report fishing activity but this was under and over reported throughout the data set.  Analyze was able to derive a fishing prediction function using candidate analytics to positively identify fishing behavior on the open seas.

Data Scientists working on this analysis would be willing to discuss the process and methods used in this analysis.   If you happen to be attending Strata 2014 in Santa Claravisit Analyze in booth 928 in the Innovators Pavilion.