However, there is one place in Europe that completely shatters every fundamental rule of spatial continuity and traditional mapmaking. Situated between the Netherlands and Belgium is the town of Baarle. To the untrained eye, it is a single, quiet European municipality. To a cartographer, it is the most beautifully complex cadastral anomaly in the world [1].

Here, the international border does not follow a river or a road. Instead, it fragments like broken glass across streets, gardens, restaurants, and living rooms. This is the story of Baarle-Hertog (Belgium) and Baarle-Nassau (the Netherlands), a geographic puzzle consisting of 30 distinct pieces of land, and how a medieval land dispute created a modern engineering marvel [1].

The Geometry of the 30 Enclaves

To understand Baarle, you must first understand the concept of an “enclave.” In geography, an enclave is a portion of territory completely surrounded by the territory of another state. For example, a piece of Belgium completely surrounded by the Netherlands is a Belgian enclave.

In Baarle, this concept is taken to the extreme. The town consists of exactly 30 fragmented territorial pieces. The mathematical breakdown is as follows:

• 22 Belgian Enclaves: These are isolated parcels of Belgian territory (Baarle-Hertog) floating like islands within the Netherlands.
• 8 Dutch Enclaves: These belong to the Netherlands (Baarle-Nassau).

If this were not complicated enough, seven of the eight Dutch enclaves are located inside the largest Belgian enclave (known in cadastral records as enclave H1). In spatial engineering, we call this a “second-order enclave” or a “counter-enclave.” Think of it like a set of geographical Russian nesting dolls: you have the Dutch mainland, which surrounds a Belgian island, which in turn surrounds a smaller Dutch island.

The eighth Dutch enclave (N8) is the only one that does not follow this rule; it sits completely outside the town, entirely surrounded by the main territory of Belgium. There is even a spot in a local cornfield where two Belgian enclaves (H1 and H2) touch at a single microscopic coordinate without overlapping, forming a rare “quadripoint” — a crossroads of four borders.

The Genesis of the Mess: The Year 1198

When you look at a map of Baarle, the most natural question to ask is: “Was this border drawn by mistake?” As an engineer, I can assure you that the border was not drawn badly. In fact, for a very long time, it was not drawn at all. The complexity is not the result of a surveying error, but rather the perfect preservation of a very old feudal contract.

The story begins in the year 1198. The land was divided between two powerful medieval figures: the Duke of Brabant and the Lord of Breda. They did not divide the land by drawing a straight line. Instead, they divided it based on land use and ownership. The Duke of Brabant kept the properties that were already cultivated, populated, and capable of generating tax revenue. The Lord of Breda was given the remaining uncultivated lands, such as forests and heathlands.

The titles of these two men are forever stamped on the maps. The Duke’s lands (Hertog, in Dutch) became Baarle-Hertog. The Lord’s lands later fell under the control of the House of Nassau, becoming Baarle-Nassau.

Over the centuries, empires rose and fell. The 1648 Peace of Westphalia established a formal border between the northern Dutch Republic and the southern Spanish Netherlands. However, instead of trading land to create a straight border, both sides simply kept their respective medieval properties. The fragmented checkerboard of 1198 survived intact.

The Surveyors’ Surrender: The 1843 Treaty of Maastricht

The true cartographic nightmare occurred when Belgium gained its independence from the Netherlands in 1830. An official international border had to be drawn. In 1843, a joint boundary commission was established under the Treaty of Maastricht. Surveyors methodically walked the border, placing cast-iron posts from the German border to the North Sea.

When they arrived at the Baarle region — specifically between border posts 214 and 215 — the commission was faced with an impossible task. The land was heavily populated, and residents refused to change their nationality or give up their traditional tax statuses. Rather than forcing a straight line and triggering a local rebellion, the international surveyors threw their hands in the air and essentially gave up.

Instead of drawing a continuous border, they meticulously analyzed 573 separate parcels of land, assigning a definitive Dutch or Belgian nationality to each one based on historical ownership. The map they created was not a line, but a complex mosaic. This 1843 cadastral registry officially transformed a medieval land dispute into modern international law.

The “Voordeurregel”: The Front Door Rule

How does a society function when an international border zigzags through a densely populated town? The border here does not respect urban planning. It cuts directly through streets, gardens, warehouses, and even private houses.

To solve the administrative chaos of a house sitting in two countries at once, local authorities rely on a legal geographic principle called the “Voordeurregel,” or the Front Door Rule. The rule states that a building’s nationality — and therefore where its residents pay taxes, vote, and register their vehicles — is determined entirely by the location of its main front door.

If the border slices right through the exact middle of the front door, the building is officially divided. It is assigned two different addresses, two different postal codes, and features two separate house number plaques, usually decorated with small Dutch and Belgian flags to avoid confusion.

Historically, this rule has led to highly creative spatial manipulation. If the Dutch government raised taxes, or if Belgian laws required restaurants to close earlier, property owners would simply physically move their front door a few meters to the left or right, legally changing the country their building belonged to.

You can walk into a local business, such as the famous liquor store ‘De Biergrens’, and see the international border painted on the floor. The loading dock is bisected by the border. A customer can stand in the Netherlands, reach across the line, and pick up a product sitting in Belgium. Similarly, the local printing company, Em. de Jong, actually contains an entire Belgian enclave (H9) entirely inside its commercial building.

To help tourists and residents navigate this invisible maze, the local municipalities have marked the borders on the town’s sidewalks and roads with a series of white crosses (+ + + +), accompanied by the letters “B” for Belgium and “NL” for the Netherlands.

The Final Measurement: The 30th Piece in 1995

You might assume that a border established in 1843 was completely finalized. But mapping is an evolving science. For over a century, local authorities believed there were only 29 enclaves in Baarle.

However, in 1995, a modern cadastral team set out to digitize the exact borders using highly precise coordinate geometry. During this process, they discovered a small, unassigned piece of land — a meadow known locally as “Hiltje” or Parcel H8. According to the 1843 documents, this small patch of grass had accidentally been left unallocated; it was legally stateless.

After deep archival research, geomatics experts determined that the parcel historically belonged to the Duke of Brabant. Consequently, in 1995, this tiny meadow was officially awarded to Belgium, becoming the 22nd Belgian enclave, and bringing the total number of enclaves in Baarle to exactly 30. That year, the border was finally, officially completed — down to the exact millimeter.

Conclusion: A Lesson in Geographic Harmony

From an engineering perspective, the borders of Baarle-Hertog and Baarle-Nassau represent a failure of standard spatial optimization. It is inefficient, legally complicated, and administratively heavy. The town requires two mayors, two municipal councils, two postal systems, and two police forces sharing a single building.

Yet, as a human experiment, Baarle is a remarkable triumph. It proves that borders do not have to be walls, fences, or militarized zones. In Baarle, an international border is just a white cross on the pavement that you step over while walking to the bakery. It is a testament to peace and cooperation, showing that even the most fragmented, complicated lines on a map can be managed when the people living across them choose to simply get along.

For mapmakers and engineers, Baarle will always be a beautifully complicated paradox: the most chaotic border in the world, functioning in absolute harmony.

References

[1] Baarle-Hertog and Baarle-Nassau Border Enclaves

[2] Baarle-Hertog

[3] Baarle-Nassau/Baarle-Hertog

[4] The Baarle Enclaves

[5] Baarle-Nassau/Baarle-Hertog

Those details are easy to miss in broad land-cover maps. They are also the sort of details that satellite imagery and spatial AI are getting better at finding. Two recent projects — one from Google in England, another from Amazon in Africa — show how that work could help conservation teams decide where to investigate next. Neither project counts wildlife directly. Both offer a more detailed way to look at the landscape.

A map that sees beyond “green”

In June 2026, Google Research described a new, vectorized map of small landscape features across England’s farmland. It distinguishes hedgerows, copses, linear woodland, woodland, and stone walls. That is a useful step beyond a map that simply labels a place “vegetation”: the shape and arrangement of a feature can matter as much as its color [1, 2].

The team used a vision-transformer model pre-trained on more than 300 million satellite images, then adapted it to recognize details of the English countryside. Sub-meter imagery and one-meter LiDAR helped the model handle features that can overlap — for example, a stone wall running beneath a hedgerow. The resulting detections were turned into vector shapes, which are easier to count and compare than a field of colored pixels [1].

Why care about a line of trees between fields? Hedgerows can help wildlife move between isolated habitat patches, though the benefit depends on the species and the landscape. The practical details matter too: a corridor that looks well placed on a screen may not work for every species, and any plan has to make sense to the people farming the land. Research on European agricultural landscapes points to those local differences as part of the conservation problem, not an afterthought.

There is a catch with Google’s map: it describes conditions in imagery captured between 2018 and 2020. It is a helpful baseline, not a live inventory. Google’s data catalog also warns that performance is weaker in dense urban and mountainous areas, and that stone walls are less accurately mapped than some other feature types. If a conservation decision depends on what is there now, someone still needs to check on the ground [2].

Searching a continent for restoration candidates

Google’s project looks closely at small features in one country. Amazon’s pilot takes a wider view. The company tested whether a geospatial foundation model could help find African landscapes that resemble places where afforestation, reforestation, and revegetation projects already exist. The model worked with Sentinel-2 satellite images and summarized each roughly 2.5-kilometer patch as a numerical representation — an “embedding” — that could be compared with other patches [3].

The team then used those comparisons to flag landscape look-alikes and intersected them with a separate map of reforestation potential. The result is a way to narrow a very large search area. It is not a list of sites ready for planting.

One reported result is easy to misread. The model scored about 0.93 AUC when ranking areas in the top quartile of reforestation potential. That does not mean that 93% of the map is correct. It means the model was usually able to rank a higher-potential location above a lower-potential one in that test. Its score for predicting where ARR projects actually exist was much lower, about 0.62. That makes sense: satellite images show aspects of the land, but they do not reveal who owns it, whether permits are possible, whether funding is available, or whether a project has local support [3].

The patch size is another reason to treat this as an early filter. A 2.5-kilometer view can help decide where to send analysts; it cannot tell a project team which parcel to choose. Amazon’s own case study says the look-alike areas need further ecological and social assessment. Similar-looking land is not proof of a suitable site — or of a successful restoration project [3].

A map is a starting point

There is a broader point here. A satellite can show where trees are, but “more trees” is not a complete conservation plan. Planting in the wrong place can compete with food production or damage an ecosystem that is naturally open. A 2025 study refining global reforestation maps used layers for cropland, wetlands, and peatlands, among others, to rule out areas where planting could create unwanted trade-offs [4].

That is why spatial AI is most useful as a way to focus attention. It can help researchers find patterns across a huge area, flag places that have changed, or identify patches that deserve a closer look. It cannot tell us, by itself, whether a species is present, whether a corridor works, or whether a restoration plan is welcome locally.

The useful question is not “Can the model pick the best place?” It is “Can this map help the people who know the place ask a better question?” Maybe that means checking whether a hedgerow connects two remaining habitats. Maybe it means ruling out a planting site before money and time are spent there. Either way, the map is doing its job when it leads to a better decision on the ground — not when it replaces one.

References

[1] From Pixels to Planning: Earth AI for Nature Restoration

[2] Farmscapes 2020 Vectorised — Earth Engine Data Catalog

[3] Applying Geospatial Foundation Models to Reforestation Site Selection

[4] Addressing Critiques Refines Global Estimates of Reforestation Potential

[5] Connecting Habitats in European Agricultural Landscapes