Skip to content
Back to search results

Flow Mapping Analysis

Introduced by
Flow cartography tradition (computer migration mapping by Waldo Tobler)
Year
1987
Type
Cartographic technique for visualizing movement between origins and destinations
Family
Movement / flow cartography
Font size

What does it do?

Flow mapping analysis visualizes movement between places — migrants, commuters, trade, traffic — by drawing the flows of an origin-destination matrix as lines on a map, with line width scaled to the volume moving along each link.

It is the cartography of interaction: where choropleths show what is in a place, flow maps show what travels between places, and the central challenge is to reveal the dominant patterns of movement without the map dissolving into an unreadable tangle of crossing lines. The technique was put on a computational footing by Waldo Tobler's 1987 experiments in computer migration mapping, and modern methods add edge bundling, smoothing, and statistical filtering to manage visual complexity.

The intuition behind the method

Picture every person who moved between two cities last year drawn as a thread connecting them; a few major routes carry thick ropes of movement while thousands of minor links form a faint haze. A flow map tries to show this so the eye immediately sees the big rivers of movement and ignores the trickles. But if you naively draw every origin-destination pair, the lines cross and overlap until the map is a hairball. The craft of flow mapping is therefore as much about subtraction and tidying as drawing: bundle threads that travel together so they read as a single route, hide or fade the flows too small to matter, and emphasize the ones that are unexpectedly large given how far apart and how big the places are.

History

Flow maps are among the oldest forms of thematic cartography, with Charles Joseph Minard's mid-nineteenth-century flow maps — above all his 1869 depiction of Napoleon's Russian campaign — celebrated as masterpieces of the genre, drawn entirely by hand. Henry Drury Harness had mapped Irish traffic flows even earlier, in 1837. The computational era began with Waldo Tobler, whose 1987 paper 'Experiments in migration mapping by computer' set out algorithms for drawing and simplifying flows from migration matrices and articulated the clutter problem that still defines the field. From the 2000s, information-visualization research contributed force-directed and hierarchical edge-bundling algorithms to reduce that clutter, and web mapping libraries such as flowmap.gl and kepler.gl brought interactive, large-scale flow mapping to a wide audience.

When to use it

Use flow mapping analysis whenever your data describe movement or interaction between located places and you want to show its geographic structure — migration between regions, commuting between home and work, trade between countries, traffic between zones, or transfers between accounts. It is the natural display for an origin-destination matrix when both the magnitude and the geography of flows matter. It works best for a manageable number of dominant links or after bundling and filtering reduce a dense matrix to its structure; it is less suitable when there are so many comparable flows that no design can declutter them, when direction or net versus gross flow is the real question (where specialized chord or net-flow displays may serve better), or when location is irrelevant and a non-spatial network diagram would communicate more clearly.

Limitations and considerations

  • Dense matrices produce crossing-line clutter that bundling can only partly resolve.
  • Bundling and smoothing distort the true paths of flows in exchange for clarity.
  • Representing direction and asymmetric two-way flows clearly is difficult on a static map.
  • Long lines between distant places can dominate visually regardless of their actual importance.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Shows movement and interaction geographically, which choropleth and symbol maps cannot.
  • Encodes flow volume directly through line width, making dominant routes immediately visible.
  • Edge bundling and smoothing can render even dense origin-destination matrices legible.
  • Statistical filtering against a gravity baseline highlights structurally significant, not merely large, flows.

Applications

  • Mapping internal or international migration flows between regions or countries.
  • Visualizing commuting patterns between residential and employment zones.
  • Displaying trade flows of goods or capital between economies.
  • Showing transport and traffic movements across a network of zones.
  • Tracing the movement of information, disease, or money between located entities.

How it works

How it works

1

The input is an origin-destination matrix whose cell T_ij records the volume moving from place i to place j, together with coordinates for each place. Each non-trivial cell becomes a line drawn from the origin to the destination, and the visual weight of that line — usually its width, sometimes its colour intensity — is scaled to the flow volume so that larger flows are heavier. This raw rendering is the literal flow map, faithful to the data but typically far too cluttered to read when the matrix is large.

Formula
T_{ij} \rightarrow \text{line}(i \to j),\quad \text{width} \propto T_{ij}
2

To tame the clutter, flows that run in similar directions through similar areas are pulled together into shared bundles, just as separate roads merge into a highway. Force-directed edge bundling treats edges as flexible springs attracted to nearby compatible edges, while kernel-based and hierarchical methods group flows geometrically; smoothing curves the lines so bundles read as coherent streams. Bundling dramatically reduces visual complexity and exposes the trunk routes of movement, at the cost of slightly distorting the exact path of any individual flow.

Formula
\text{force-directed or kernel bundling of nearby edges}
3

Not every large flow is interesting: big places exchange many people simply because they are big and close. To find the noteworthy flows, observed volumes are compared with those expected under a baseline such as a gravity model, and the residual — observed minus expected — highlights origin-destination pairs that flow far more (or less) than their sizes and separation predict. Mapping only the dominant or over-represented flows, rather than all of them, turns a cluttered everything-map into a focused picture of structurally important movement.

Formula
\text{residual} = T_{ij} - \hat{T}_{ij}
4

Finally the map is designed for legibility: line widths are scaled with a perceptually sensible function (often square-root) so very large flows do not blot out the rest, direction is shown with arrowheads, tapering, or animation, and origin versus destination ends are distinguished by colour. A clear legend ties width to volume, asymmetric flows between the same pair may be offset or curved so both are visible, and a basemap provides geographic context without competing with the flows. Good design is what separates an informative flow map from a decorative tangle.

Formula
\text{width scaling},\ \text{direction arrows},\ \text{origin/destination colour}

Common pitfalls

  • Drawing every origin-destination pair at once, producing an unreadable hairball.
  • Scaling line width linearly with volume, so one huge flow obscures everything else.
  • Treating large flows as significant without comparing them to a gravity or distance baseline.
  • Losing the direction or the origin-versus-destination distinction through ambiguous symbolization.

Frequently asked questions

What is edge bundling and why is it used in flow maps?

Edge bundling groups flow lines that run in similar directions through similar areas into shared, curved bundles, much as separate roads merge into a single highway corridor. It is used because a flow map of a large origin-destination matrix quickly becomes an unreadable hairball of crossing lines, and bundling collapses that tangle into a handful of legible trunk streams. The trade-off is that the bundled path of any individual flow no longer follows its true straight-line route, so bundling buys clarity at the price of geometric precision.

How do I decide which flows are significant enough to map?

Large flows are not automatically significant, because big, nearby places naturally exchange a lot simply due to their size and proximity. A standard approach is to fit a baseline model — most often a gravity model that predicts flow from the sizes of the two places and the distance between them — and then examine the residual, the difference between the observed and expected volume. Flows that are far larger than expected are the structurally interesting ones, and mapping only those dominant or over-represented links turns a cluttered map into a focused one.

How is flow mapping related to the gravity model of migration?

They are complementary. The gravity model is a statistical model that predicts how much flow to expect between each pair of places from their sizes and separation, while flow mapping is the cartographic display of the actual observed flows. The two combine naturally: the gravity model supplies the expected baseline, and the flow map can then show either all flows or, more powerfully, only the residual flows that the gravity model fails to explain, revealing the movement patterns that are surprising rather than merely predictable.

Related methods

Sources

  1. Tobler, W. (1987). Experiments in migration mapping by computer. The American Cartographer, 14(2), 155–163.

    Open DOI reference
    Copied.