Detecting Illegal Urban Construction in Jakarta From Orbit

Jakarta is sinking faster than almost any other big city on Earth, while millions of new homes, shops and shophouses appear on its fringes every year. Much of this building work happens without permits, in floodplains and along river buffers where construction is supposed to be banned. City planners have long relied on tip-offs and slow field inspections to catch offenders, but the pace of growth now overwhelms that approach. A new generation of Earth observation tools, combined with machine learning, is starting to change how authorities spot unauthorised urban construction.

For Australian readers who watch their own urban fringes balloon each year, the Jakarta story is uncomfortably familiar. Outer suburbs across NSW, Victoria and Queensland are pushing further into bushland and farmland, with permit disputes a recurring headache for local councils. The technical lessons emerging from Jakarta therefore matter well beyond Indonesia. They point to a way of handling development pressure that any growing city in the Asia-Pacific could borrow.

Why Jakarta's Building Boom Defies Easy Monitoring

Jakarta's metropolitan area now holds more than 34 million people, spread across hundreds of kelurahan and densely packed kampung settlements. Ground teams can only realistically audit a fraction of new structures each year, and informal builders rarely lodge paperwork before pouring slabs. Add chronic land subsidence, with some northern districts dropping more than ten centimetres a year, and the picture becomes genuinely hard to track.

Compounding the problem, much of the metropolitan region sits across multiple administrative boundaries, from DKI Jakarta proper to Bogor, Depok, Tangerang and Bekasi. Each local authority keeps its own records, and the national permitting system does not always talk to the provincial ones. By the time an inspector learns about a new building, walls and roofs are already in place, and demolition becomes politically costly.

How Satellite Imagery Detects Unauthorised Construction

Modern Earth observation platforms give planners a much sharper view than aerial photography ever did. Sub-metre optical satellites, such as the WorldView and Maxar constellations, can pick out individual rooftops, drainage works and access roads from more than 500 kilometres up. Synthetic aperture radar, mounted on platforms like Sentinel-1 and ICEYE, sees through cloud cover and monsoon rain, returning consistent imagery through Jakarta's long wet season.

The real breakthrough comes from change detection. By comparing images taken weeks or months apart, analysts can flag new footprints that did not exist in the previous capture. Because most illegal construction involves clear ground disturbance such as cleared land, fresh concrete and new roofing, these changes stand out clearly against surrounding vegetation or older rooftops. Some teams now combine optical and radar layers, so a building going up under heavy cloud still shows up in the next pass.

AI and the Machine Learning Layer Above the Pixels

Raw imagery only becomes useful once algorithms can interpret it. In Jakarta, researchers have trained convolutional neural networks on labelled examples of legal and illegal structures, drawn from historical permits and field surveys. The models learn to recognise roof shapes, colours and materials common to informal housing, and to flag anything that appears inside a designated no-build zone such as a river setback or a flood retention basin.

Accuracy has improved quickly. Recent pilots in the Jabodetabek region report precision rates above 85 percent when models are trained on locally gathered data, rather than off-the-shelf datasets built for other cities. False positives still happen, particularly with temporary market stalls, but human review of the shortlist keeps the workload manageable. Cities like Ho Chi Minh City and Phnom Penh are now trialling similar pipelines, often with support from regional development partners.

Lessons From Down Under

Australian planners face their own version of the same challenge. Western Sydney keeps pushing outwards through new neighbourhoods rising around Marsden Park, Schofields and the Aerotropolis, and councils such as Blacktown and Liverpool spend heavily on compliance teams chasing unpermitted granny flats, sheds and secondary dwellings. Melbourne's urban growth boundary and Brisbane's southern corridor raise similar enforcement questions. The Australian approach has traditionally combined cadastral data with aerial imagery flown by contractors, but a move towards frequent, high-resolution satellite passes could cut inspection costs dramatically.

Local capability is already strong. Geoscience Australia and CSIRO both run Earth observation programs, and Sydney-based firms such as Arlula and Tasmanian-founded LISTmap are pushing commercial satellite analytics for planning and environment work. Brisbane's growing space precinct around the Queensland University of Technology is another piece of the puzzle, training the analysts who can turn raw pixels into actionable evidence. For anyone used to a bit of Aussie straight talk, the message is fair dinkum: the technology is here, and the main barrier is now data sharing between agencies, not the satellites themselves.

Scaling the Approach Across the Asia-Pacific

Jakarta's experience is increasingly being treated as a template. The same pipeline that flags unauthorised housing along the Ciliwung can be retuned for cyclone-prone coastlines in the Pacific, or for informal expansion around rapidly growing cities in Bangladesh and the Philippines. What matters is a shared framework for handling imagery, permits and field validation, so that one country's pilot does not sit on a shelf gathering dust.

This is where multi-stakeholder platforms earn their keep. Coordinated efforts through bodies such as the ICTD-ASP platform help align donor funding, government priorities and private-sector capability, so that a model trained on Jakarta rooftops can be handed to a planning agency in Suva or Port Moresby without reinventing the wheel. For Australia, joining that conversation brings both commercial opportunity for local space startups and practical knowledge for councils wrestling with growth on their own fringes.

Technique	Best Use Case	Strengths	Limitations
Sub-metre optical imagery	Detailed rooftop mapping in clear weather	Sharp visual detail; easy human interpretation	Blocked by cloud and monsoon haze
Synthetic aperture radar	Monitoring during wet season or at night	Sees through cloud; detects ground disturbance	Less intuitive for non-specialists
Hyperspectral imaging	Distinguishing roofing materials and land cover	Identifies materials used in informal builds	Higher cost; smaller archive of historical data
AI-driven change detection	Flagging new structures between time slices	Scales to whole-city reviews	Needs quality baseline imagery and ground truth

For Australian planners and policy folk watching Jakarta, the next practical step is small but concrete: commission a six-month pilot using openly available Sentinel-1 radar and a commercial sub-metre provider, focused on a single high-pressure growth corridor, and compare the cost per detected unpermitted structure against the council's current field inspection budget. The data will quickly show whether orbiting sensors are ready to join the compliance toolbox.