How Simplification Can Distort Boundaries

Geometry simplification reduces vertices by replacing detailed portions of a line or polygon boundary with a less detailed approximation. Because the replacement geometry does not necessarily follow the original coordinates exactly, simplification can move boundaries, change measurements, remove narrow features, and alter spatial relationships.

At an appropriate tolerance, those changes may be negligible for the intended use. A world map does not need every bend in a highly detailed national coastline. At an inappropriate tolerance, however, a simplified boundary can exclude areas that were previously inside a polygon, include areas that were previously outside it, or stop matching the boundary of a neighbouring feature.

This does not mean simplification is inherently unreliable. It means that a simplified geometry is a derived approximation, and the amount of acceptable distortion depends on what the geometry will be used for.

Simplification changes where the boundary is

A polygon boundary can be thought of as a sequence of vertices:

•──•─•──•────•─•──•─•

A simplification algorithm decides that some of those vertices contribute insufficient detail at the chosen tolerance and removes them.

The new boundary might become:

•────────•─────────•

The replacement segments connect retained vertices directly.

Where the original boundary curved or deviated between those retained points, the new segment can pass somewhere else.

That creates small regions where the simplified polygon differs from the source:

original boundary    ~~~~~~~~~~~
simplified boundary  -----------

Some original area may fall outside the simplified polygon. Elsewhere, the simplified polygon may include area that was outside the original.

The feature can still look almost identical at ordinary map scale while having a mathematically different footprint.

That is the fundamental source of simplification distortion.

A simplified polygon is an approximation, not merely a smaller file

It is easy to think of simplification as geometry compression.

The analogy is useful only up to a point.

Reducing the number of coordinates usually reduces storage and processing requirements, but unlike lossless compression, the original coordinates cannot generally be reconstructed from the simplified output.

Suppose a coastline originally contains:

A ──╲_╱──╲__╱─╲_╱── B

and simplification replaces that detail with:

A ───────────────── B

The omitted inlets and projections no longer exist in that geometry.

The dataset may still be perfectly appropriate for a continental overview. It is simply no longer an exact geometric representation of the source dataset.

What Is Geometry Simplification in GIS? explains the underlying operation and the role of tolerance in more detail.

Area can change

When a polygon boundary moves, its area can change too.

Consider a small outward projection:

+------------+
|            |____
|                |
|            |‾‾‾
+------------+

If simplification removes that projection, the resulting polygon loses its area.

Elsewhere, replacing an inward bend with a straight segment can add area instead.

There is no general rule that simplification always increases or always decreases polygon area. The effect depends on the geometry and algorithm.

Across a complex boundary, some differences may cancel while others accumulate.

This matters when area is part of the analysis.

A simplified administrative polygon may be entirely suitable for drawing a small map while being a poor choice for calculating official land area. Likewise, a simplified habitat polygon may be useful for web display without being appropriate for estimating habitat extent.

If the analytical output depends on precise area, test the effect rather than assuming that a visually similar boundary produces an equivalent measurement.

Perimeter and line length can change substantially

Length is often even more sensitive to simplification.

A highly irregular line:

/\/\/\/\/\/\/\/\/\

travels a greater distance than a straight segment between the same endpoints:

------------------

Remove the smaller bends and the measured length generally changes.

This is particularly relevant for features such as:

coastlines;
rivers;
trails;
administrative boundaries;
contour lines.

The phenomenon is not limited to GIS simplification. Measured length itself depends on the scale and resolution at which a complex line is represented.

A simplified coastline can therefore remain recognisably the same coastline while having a noticeably shorter calculated length.

Any stored length or perimeter attributes also need scrutiny: simplifying the geometry does not guarantee that previously stored measurements are recalculated automatically.

Narrow projections can disappear

Simplification tends to affect details that are small relative to its tolerance.

Imagine a polygon with a narrow peninsula:

+--------------+
|              |------+
|              |      |
|              |------+
+--------------+

At a sufficiently aggressive tolerance, the vertices describing that projection may no longer survive.

The resulting boundary might become:

+--------------+
|              |
|              |
|              |
+--------------+

For a small-scale map, that may be desirable generalisation.

But suppose the projection represents:

a territorial corridor;
part of a parcel;
access to a road;
a river channel;
a protected habitat connection.

Its geometric size may be small while its geographic significance is large.

Simplification algorithms do not know that significance. They work from geometry and their mathematical criteria.

Narrow connections can be altered or lost

A polygon can contain two larger areas joined by a narrow neck:

+-------+       +-------+
|       |-------|       |
|       |       |       |
+-------+       +-------+

Aggressive simplification may change the geometry around that connection.

Depending on the algorithm, implementation, and topology constraints, the connection can become much narrower, change shape, or in non-topology-preserving workflows contribute to an invalid or otherwise unsuitable result.

This is one reason there are simplification methods specifically designed to preserve topology.

PostGIS, for example, distinguishes ordinary ST_Simplify from ST_SimplifyPreserveTopology. Its ordinary Douglas–Peucker simplifier may alter topology or even produce invalid polygonal geometry.

The fact that a simplification algorithm accepts polygons does not therefore mean that every tolerance produces a structurally acceptable polygon.

Small features can disappear entirely

Distortion does not always mean a feature becomes slightly misshapen.

Sometimes the entire feature is small relative to the tolerance.

An ordinary simplifier may reduce a very short line or small polygon so aggressively that it collapses or disappears, depending on the implementation and settings.

This can matter in datasets containing:

small islands;
tiny administrative enclaves;
short road segments;
small water bodies;
narrow habitat patches.

At a world-map scale, suppressing tiny features can improve legibility.

At a local analytical scale, silently losing them may be unacceptable.

This illustrates why a tolerance has meaning only relative to both the geometry and the purpose.

Holes can also be affected

Polygon interiors may contain holes.

For example:

+-------------------+
|                   |
|      +------+     |
|      | hole |     |
|      +------+     |
|                   |
+-------------------+

Those rings are geometry too.

A sufficiently aggressive simplification can alter their shape. Depending on the simplification method, topology rules, and size of the hole, some operations may preserve it while others may allow small structures to collapse or disappear.

A topology-preserving simplifier can impose stronger guarantees.

PostGIS's ST_SimplifyPreserveTopology, for example, preserves the number of polygon rings and ensures that resulting rings do not cross when the input is valid.

That is an important improvement over unconstrained simplification—but it does not solve every boundary problem.

Topology-preserving does not mean boundary-preserving

This distinction is easy to miss.

Suppose two polygons share the same border:

Polygon A    |    Polygon B
             |
             |
             |

The boundary is stored as part of both polygon geometries.

Now simplify A independently and simplify B independently.

Even if both resulting polygons remain individually valid, their replacement boundaries are not necessarily identical.

You might get:

Polygon A      Polygon B
      \        /
       \      /
        \    /

creating a gap, or:

Polygon A
        \__
           \__
          __/
      ___/
Polygon B

creating an overlap.

PostGIS explicitly notes that ST_SimplifyPreserveTopology preserves the topology of each input geometry but does not preserve shared boundaries between separate polygons. For that requirement, it directs users to coverage-aware simplification.

This is the difference between preserving a polygon's own topology and preserving the topology of a collection of adjacent polygons.

Adjacent polygons need coverage-aware simplification

Consider an administrative dataset in which municipalities tile an entire region.

Ideally:

A | B | C

has neither gaps nor overlaps.

The shared boundary between A and B should remain exactly the same line for both features, even after generalisation.

Simplifying each municipality independently can break that property.

A coverage-aware simplifier instead treats matching polygon boundaries as a connected system.

QGIS's Simplify coverage operation is designed for polygon coverages whose edges match exactly. It simplifies the edges while retaining a valid coverage, preventing the simplified edges from crossing or touching in ways that would destroy the coverage structure.

PostGIS's ST_CoverageSimplify addresses the same problem: shared edges are simplified consistently so adjacent polygons continue to form a valid coverage.

This is usually the more appropriate model when the input represents a complete polygon partition rather than unrelated standalone features.

Coverage-aware simplification assumes the coverage was valid first

A specialised coverage algorithm cannot safely infer relationships that were already inconsistent.

Suppose two polygons appear to share a boundary on the map but actually contain a narrow gap:

A | | B

or a small overlap:

A || B

A coverage simplifier may not treat them as one shared edge because geometrically they are not one.

PostGIS warns that invalid coverages can produce unexpected artefacts during coverage simplification, including boundary intersections or boundaries separating that appeared to be shared.

QGIS likewise provides a separate Validate coverage operation alongside its coverage processing tools.

So a coverage-aware workflow generally assumes that the input already represents a valid polygon coverage.

Simplification should not be expected to repair topology that was wrong beforehand.

Gaps and overlaps may be tiny but still matter

A gap created by independent simplification may be almost invisible.

At normal zoom, two boundaries can still seem aligned.

Yet later operations can detect the difference.

Suppose simplified administrative polygons are intersected with population points.

A point falling into a tiny gap may belong to no municipality.

If polygons overlap instead, a point may appear to belong to two municipalities.

The visual error may be a fraction of a millimetre on screen while the analytical consequence is categorical.

This is why visual inspection alone is not enough when simplified geometry will participate in subsequent spatial analysis.

Containment relationships can change

Imagine a small lake lying close to an administrative boundary.

Before simplification:

Boundary      Lake
~~~~~~~       ○

the lake lies inside Municipality A.

If the boundary shifts during simplification, it can pass across the lake or move to its other side.

A later point-in-polygon or containment operation can then produce a different answer.

Similar issues can affect:

buildings close to parcel boundaries;
observations near protected-area edges;
roads following administrative borders;
points near electoral boundaries.

A seemingly minor positional deviation matters whenever another feature lies close enough to the original boundary.

The acceptable tolerance therefore depends partly on what else will be compared with the simplified geometry.

Intersections can move

Suppose a river crosses an administrative boundary.

Simplify that boundary and the crossing point may shift.

If the river itself is also simplified independently, the intersection can shift further or even change structure.

For display purposes, the difference may be negligible.

For a workflow identifying the exact location at which responsibility changes between jurisdictions, it may not be.

This is another reason a geometry can be suitable for one purpose and unsuitable for another without either version being inherently “bad”.

Simplification can affect distance calculations

If a boundary moves, distances measured to it change.

Suppose a building lies 12 metres outside the original polygon boundary.

After simplification, the replacement line passes 8 metres from the building.

A subsequent query such as:

Which buildings lie within 10 metres of this boundary?

can now return a different result.

The effect becomes especially important when the simplification tolerance is of similar magnitude to the distance thresholds used later.

Using a 100-metre simplification tolerance before performing a 20-metre proximity analysis deserves obvious scrutiny.

The simplification error budget and the analytical threshold should be considered together.

CRS units affect the tolerance

Many simplification tolerances are expressed in the coordinate units of the data.

For a layer in a projected CRS using metres:

tolerance = 100

may represent a 100-metre geometric threshold.

For longitude/latitude geometry processed using a method that interprets raw coordinate units:

tolerance = 0.01

represents degrees, not kilometres or metres.

PostGIS documents the tolerance of ST_Simplify and ST_SimplifyPreserveTopology in the units of the input spatial reference system.

That makes CRS interpretation relevant to distortion. A seemingly small numeric tolerance can represent a very different physical distance depending on the coordinate system.

This article is not another general CRS guide, but the rule remains important: know what the tolerance measures before judging whether it is small.

Simplification for visualisation has different requirements from simplification for analysis

A useful distinction is between two outputs.

A display derivative

A simplified dataset used only to draw a map can tolerate positional approximation if the differences are below the map's useful visual resolution.

The authoritative source geometry remains elsewhere.

An analytical derivative

A simplified dataset used in:

overlays;
measurements;
containment tests;
proximity analysis;
area statistics;

needs stronger validation because its altered boundary directly influences the result.

This does not mean analytical simplification is forbidden. It can be valuable when the source geometry contains far more detail than the analytical resolution requires.

It means the acceptable tolerance should be justified against the analysis.

The same simplified geometry should not automatically be promoted from “good enough for the basemap” to “good enough for measurement”.

Legal and authoritative boundaries deserve particular caution

Some boundaries carry consequences that far exceed their cartographic appearance.

Examples include:

property parcels;
concessions;
protected-area limits;
administrative jurisdictions;
electoral districts;
legal setbacks.

A simplified representation can be useful for public-facing overview maps while the authoritative geometry remains the basis for decisions.

The simplified version should not silently replace that source.

Even a very small positional shift could move a parcel, building, observation, or other feature from one side of a legally meaningful boundary to the other.

When precision carries legal, regulatory, or contractual consequences, use the authoritative data and appropriate domain rules rather than assuming that a visually faithful generalisation remains operationally equivalent.

Boundary distortion is scale-dependent

A ten-metre displacement can be imperceptible on a world map.

The same displacement can be enormous in a cadastral map.

This means there is no universal statement such as:

A 20-metre simplification tolerance is safe.

Safe for what?

For a continental overview, perhaps.

For mapping individual properties, probably not.

For a regional ecological model with kilometre-scale inputs, it may or may not matter depending on how the boundaries are used.

Tolerance should be evaluated against:

intended display scale;
analytical resolution;
source accuracy;
smallest important features;
neighbouring datasets;
downstream operations.

The number becomes meaningful only within that context.

More detailed geometry is not automatically more correct

Caution about distortion should not turn into the opposite mistake.

The original high-vertex geometry may itself be approximate.

A boundary containing 100,000 vertices could have been traced from:

a low-resolution raster;
noisy GPS observations;
an old paper map;
interpolated data;
another already-generalised dataset.

Preserving every vertex does not create positional accuracy that the source never had.

Simplification should therefore be evaluated relative to source quality and required use, not merely relative to vertex count.

A carefully generalised boundary can be entirely appropriate at its target scale.

How can you measure simplification distortion?

Visual comparison is useful, but quantitative checks can provide a better picture.

Depending on the use case, you might compare:

vertex count before and after;
area change;
perimeter or length change;
maximum or typical positional deviation;
number of features lost;
number of invalid geometries;
gaps and overlaps in polygon coverages;
changes in important intersections;
changes in containment classifications.

No single metric is sufficient for every dataset.

A 0.1% area difference can hide a large local displacement along one part of a boundary. Conversely, a noticeable perimeter reduction may be perfectly acceptable for a generalised coastline.

Measure the properties that matter to the intended use.

Test several tolerances rather than guessing once

Simplification is easier to evaluate when several candidate outputs are compared.

For example:

Original
↓
Tolerance 10 m
↓
Tolerance 50 m
↓
Tolerance 100 m
↓
Tolerance 500 m

For each level, compare both the benefits and losses.

Perhaps moving from the original to 50 metres removes 70% of the vertices while producing negligible visible and analytical change.

Moving from 50 to 100 metres may save relatively little additional geometry while beginning to remove important narrow features.

That point is often a better operating tolerance than the largest value that still produces recognisable polygons.

The target is not maximum vertex reduction. It is an acceptable trade-off.

Preserve the source geometry

Because simplification is lossy, an authoritative source should generally remain available.

A common production pattern is:

authoritative geometry
        ↓
simplified derivative for regional use
        ↓
more simplified derivative for small-scale use

Different representations can then serve different map scales and processing requirements without pretending that one geometry is ideal everywhere.

This approach also makes it possible to regenerate simplified derivatives when algorithms, tolerances, or product requirements change.

Replacing the only copy of the source with a heavily simplified result removes that flexibility.

Simplification vs smoothing does not solve the distortion problem

Smoothing can make a boundary appear more natural after simplification, but it does not restore the original geography.

As Simplification vs Smoothing in GIS explains, smoothing constructs another derived shape and can even add vertices.

A workflow such as:

simplify → smooth

may produce an attractive boundary, but it applies two geometric transformations.

The result should therefore be judged against its purpose rather than assumed to be closer to reality because it looks less angular.

How to simplify boundaries more safely

There is no single safe simplification workflow, but several principles are broadly useful.

Preserve the original geometry.

Choose a tolerance based on the intended scale or analytical precision rather than merely on desired file size.

Inspect small and narrow features.

Compare area, length, and other measurements when they matter.

If polygons form a coverage, use a coverage-aware approach rather than simplifying shared boundaries independently.

Validate the simplified output before using it in later overlays.

And most importantly, distinguish a geometry designed for display from one intended to support authoritative or precise analysis.

Those decisions matter more than the name of the simplification algorithm.

The real risk is not that the boundary looks different

A simplified boundary is expected to look different when examined closely. That is the point of simplification.

The more important question is whether the differences change something the use case cares about.

A coastline losing tiny bends on a world map may be harmless.

A parcel boundary shifting across a building is not.

Two administrative polygons developing a microscopic gap may be invisible on screen but still break a containment operation.

A small island disappearing may improve a generalised map while invalidating an analysis that counts islands.

So the right way to assess distortion is not:

Did the geometry change?

It necessarily did.

Instead ask:

Did it change more than this representation or analysis can tolerate?

Once that threshold is explicit, simplification becomes a controlled form of generalisation rather than an arbitrary loss of geographic detail.

References

PostGIS — ST_Simplify. Documents ordinary Douglas–Peucker simplification, including tolerance units, possible disappearance of small geometry, loss of topology, and the possibility of invalid polygonal outputs.
PostGIS — ST_SimplifyPreserveTopology. Documents topology-preserving simplification and clarifies that preserving each geometry's topology does not preserve boundaries shared between separate polygon features.
QGIS Documentation — Simplify coverage. Documents coverage-aware simplification for edge-matched polygon datasets, reducing boundary complexity while retaining a valid coverage.