Mapping fishing effort beyond AIS

Estimating global fishing effort where vessels stop broadcasting, by combining AIS with Sentinel-1 satellite radar using a random forest.

What this does

Public monitoring of industrial fishing leans heavily on AIS, the transceiver system vessels use to broadcast their position. Coverage is uneven: not every vessel carries it, not every vessel keeps it switched on, and satellite reception varies by region. Sentinel-1 radar sees vessels regardless of what they broadcast, but a radar detection tells you only that something was there, not how long it fished.

A random forest is trained on cells where both signals exist, then used to predict fishing hours in cells where only radar detections are available.

• Panel A: AIS fishing effort, 2017 to 2020
• Panel B: Sentinel-1 SAR detections of fishing vessels, same period
• Panel C: reported hours plus model estimates for radar-detected cells

Predictors are SAR detections per satellite overpass, coordinates, distance to shore, distance to port and bathymetry, at 0.1 degree resolution. R-squared is 0.77 on cells where both signals report.

On normalising by overpass frequency

An earlier version of the model used raw SAR detection counts and scored higher (R-squared 0.82). That version was partly exploiting uneven satellite coverage: cells overflown more often accumulate more detections regardless of how much fishing happens there, so the model was learning the orbit as much as the fishery. Normalising by overpass frequency removes the artefact at the cost of some apparent accuracy. The normalised model is the one used here.

The report includes a comparison of four transformation strategies.

Data

Not included in this repository, and too large for git. Both sources are public and available from Global Fishing Watch:

• AIS fishing effort: Kroodsma et al. (2018), Science
• SAR vessel detections: Paolo et al. (2024), Nature

https://globalfishingwatch.org/data-download/

Bathymetry and distance-to-shore and distance-to-port rasters are resampled to the same 0.1 degree grid. To reproduce, download the sources and place them under R/Data/ following the paths in fishing_figure.qmd.

Running it

Requires Quarto and R with dplyr, ggplot2, maps, scales, patchwork and knitr for the report, plus data.table, raster and randomForest to build the panel data and model metrics from source.

quarto render fishing_figure.qmd

The first run does the raster join, the prediction and the model evaluation, then caches the results to R/Data/figure_cache.Rdata. Later runs load the cache and redraw in seconds. Delete that file to rebuild.

Report template

The Quarto_template/ folder holds the styling: a stylesheet plus two HTML fragments referenced from the YAML header. Swapping the colour tokens and the footer produces a different identity without touching any analysis code. It is documented separately at quarto-report-template.

Status

This figure supports a manuscript in preparation on global elasmobranch conservation prioritisation. The code is shared for transparency. Please get in touch before reusing it in published work.

Author

Théophile L. Mouton, quantitative ecologist and data consultant. theophile-mouton.com