Research Article  ·  PLoS One  ·  Published March 13, 2026

Vector Fields as a Framework for Modelling the Mobility of Commodities

Sima Farokhnejad  ·  Angélica S. da Mata  ·  Mariana Macedo  ·  Ronaldo Menezes

PLoS One, 21(3): e0340109  ·  DOI: 10.1371/journal.pone.0340109

Study Region 853 municipalities in Minas Gerais, Brazil
Data Resilience 36% of areas keep <15° directional error after 60% data removed
Time Span 4 yrs of monthly cattle trade records analysed
Spatial Autocorrelation 0.258 Moran's I, indicating statistically significant spatial clustering

The Problem

Tracking how goods (animals, crops, manufactured products) move across a country is harder than it sounds. The standard tool is an origin-destination (OD) matrix: a table that records where each shipment starts and ends.

But OD matrices have a blind spot. They only know about the towns that appeared as origins or destinations. Any municipality that neither sent nor received a recorded shipment simply disappears from the map, even if cattle pass through it every day, even if a disease outbreak could ignite there.

This matters enormously for animal health surveillance. Foot-and-mouth disease, for example, travels along livestock trade routes. If your model cannot see half the network, you cannot protect half the country.

"Locations that are not explicitly recorded as origins or destinations are ignored, even though they may play significant roles in the overall flow of commodities."

Beyond missing locations, the sheer scale of livestock trade, with millions of individual movement records, makes traditional network graphs computationally expensive to build and nearly impossible to visualise meaningfully at a national level.

The Approach

This study borrows a concept from meteorology: the vector field. Just as weather maps show wind direction and speed at every point on a grid (including over the open ocean where no weather station exists), a vector field can represent commodity flow at every location in a region, including unmonitored ones.

Key Findings

Applied to cattle trade data from Minas Gerais, Brazil, one of the country's largest cattle-producing states, the framework revealed patterns that traditional network analysis had missed.

Stable in the north, volatile in the south

Northern municipalities showed low entropy; their trade directions barely changed month to month. Southern municipalities, where long-distance trade is less established, exhibited far more unpredictable flows.

Rhythmic seasonal patterns

Cosine similarity between consecutive months followed a wave-like pattern across the four-year study period, with a consistent dip towards year-end, consistent with known breeding and slaughter cycles in Brazilian cattle farming.

Sinks shift with the seasons

Sink regions (where cattle accumulate) moved across the map with the seasons. During spring and summer, sink locations closely matched the known distribution of slaughterhouses across the state.

Spatially clustered trade distances

Moran's I confirmed that municipalities trading over similar distances cluster geographically. Spatial lag maps revealed "doughnut" and "diamond" patterns: towns with atypically short or long trade ranges surrounded by towns behaving oppositely.

Robust under data loss

When 60% of municipalities were artificially removed from the dataset, the majority still showed directional deviations under 45° and distance errors under 25 km, a small perturbation at the scale of Minas Gerais (approximately 1,250 km across).

Northern hubs for long-distance trade

Source regions were concentrated in the north, acting as supply hubs for long-distance movements. Central-eastern areas served as transit corridors linking north and south of the state.

Why It Matters

Vector fields are not just a prettier way to draw a network. They unlock analytical capabilities that are simply not possible with OD matrices alone.

Infographics

From sparse network to complete field
A standard origin-destination model only connects municipalities that actually traded — leaving every other region blank. The vector field uses triangle-based interpolation to estimate a flow vector for every cell, even those absent from the original records.
Origin–Destination Network sparse · gaps present
→
Vector Field (interpolated) complete · no gaps
Directional entropy: how stable is the flow direction over the year?
Each cell contains an arrow representing trade direction. In a stable region the arrows barely change across seasons. In a dynamic region they rotate continuously.
Static region e.g., northern Minas Gerais
Dynamic region e.g., central / eastern areas
≈ 0.08Normalised entropy
STABLEDYNAMIC
≈ 0.92Normalised entropy
Critical points: sinks and sources in the cattle trade field
Particles converge toward the sink (red) — likely a slaughterhouse or feedlot — and radiate outward from the source (green), a breeding region dispersing livestock. These critical points shift location between seasons.
Sink — Attractor cattle converge here
Source — Repeller cattle disperse from here

Paper & Authors

Authors

Sima Farokhnejad Department of Computer Science
University of Exeter, UK
Angélica S. da Mata Department of Physics
Federal University of Lavras, Brazil
Mariana Macedo Northeastern University (London & Boston)
Constructor University, Bremen
Ronaldo Menezes Department of Computer Science
University of Exeter, UK

Cite This Work

@article{farokhnejad2026vectorfields, title = {Vector fields as a framework for modelling the mobility of commodities}, author = {Farokhnejad, Sima and da Mata, Ang\'{e}lica S. and Macedo, Mariana and Menezes, Ronaldo}, journal = {PLOS One}, volume = {21}, number = {3}, pages = {e0340109}, year = {2026}, doi = {10.1371/journal.pone.0340109} }

Open access · CC BY 4.0