AI-Powered 4D City Maps Predict Which Buildings Will Fall and What Materials They Will Release

In the industrial city of Kitakyushu, Japan, researchers have built something remarkable: a model that can predict which buildings will be demolished, when, and how many tons of concrete, steel, and timber will come raining out of the urban fabric. The study, published in the Journal of Industrial Ecology, treats an entire city not as a static collection of structures but as a living reservoir of materials whose slow release can be forecasted building by building, block by block, all the way to 2040.

The work matters because most existing material stock and flow analysis, the field that tracks the resources accumulated in our buildings and infrastructure, operates at coarse administrative scales. National or prefectural averages can tell you roughly how much concrete a country holds, but they blur out the dramatic differences between a thriving station district and a hollowing-out hillside neighborhood. That blindness has real consequences: planners cannot target recycling infrastructure, anticipate waste flows, or identify emerging vacancy hotspots when their data is smeared across entire municipalities.

The research team, led by Masatoshi Hasegawa of Nagoya University together with Hiroaki Shirakawa, Marianne Faith Martinico-Perez, Osamu Higashi, and Hiroki Tanikawa, attacked this problem with four-dimensional geographic information systems, essentially three-dimensional city models extended through time. By overlaying building footprint datasets from 2010 and 2018 and matching records across the two snapshots, they identified which structures had vanished. Every unmatched building was flagged as demolished, creating a demolition record for thousands of individual structures in a shrinking industrial city.

Onto that spatiotemporal foundation, the researchers layered a statistical machine borrowed from reliability engineering: a Weibull accelerated failure time model. This survival analysis framework, routinely used to predict when machine parts fail or patients relapse, treats building demolition as an event whose timing depends on measurable covariates. Each building was assigned structural attributes such as construction type and use, spatial attributes including land use zone and slope angle, and demographic attributes capturing the aging rate of the surrounding population mesh.

The model’s parameters yield intuitive and sometimes counterintuitive insights. Apartment buildings in Kitakyushu are expected to last roughly thirty percent longer than the reference category, with median lifespans around 66 years, while detached houses cluster between roughly 51 and 53 years. Steel structures showed shorter expected lifespans than reinforced concrete in this dataset, a result the authors caution may partly reflect locational and redevelopment pressures rather than material durability alone. Commercial and industrial zone buildings outlived their residential-zone counterparts by more than twenty percent, echoing earlier findings that zoning shapes building longevity.

The most striking result concerns demographic aging. Neighborhoods with a higher share of residents aged 65 and older were associated with longer expected building lifespans, an effect that at first sounds benign but carries a warning. When populations shrink and age, buildings are replaced less often, not because they are wanted but because demand has collapsed. The researchers interpret this as a growing risk of vacancy: structures persist on the map even as they empty out, inflating apparent housing stock while the city’s population, which peaked at about 1.068 million in 1979, is projected to fall to roughly 729,000 by 2050.

Translating demolition probabilities into material tonnage, the team multiplied projected demolished floor area by material intensity factors derived from a Japanese government construction survey. Aggregate dominates the flows in every zone, followed by cement, with steel contributing a larger share in industrial areas and bitumen barely registering. A Monte Carlo simulation placed cumulative material outflow between 2018 and 2040 at 38.23 million tons, with a confidence interval spanning 37.63 to 38.95 million tons, a narrow 3.5 percent spread that speaks to the framework’s statistical stability.

The spatial forecasts reveal a city divided. Areas within 500 meters of railway stations are projected to generate an average of 55.2 thousand tons of demolition material per grid mesh from 2022 to 2040, more than double the 22.3 thousand tons expected beyond that radius. Inside the city’s Residential Induction Zones, designated for compact urban consolidation, material output is dominated by apartment buildings and peaks between 2041 and 2045 at up to 1.143 million tons annually. Outside those zones, detached houses and industrial facilities dominate, with output peaking earlier, between 2036 and 2040, and detached house demolition alone cresting at 233 thousand tons per year in the late 2020s.

Perhaps the most practically powerful feature of the approach is its flexibility. Because probabilities are estimated for individual buildings, results can be aggregated to any geography the user chooses: school districts, census meshes, station catchments, or administrative boundaries, without the sample-size collapse that afflicts methods that subdivide observations into ever finer categories. The authors argue this makes the framework ideal for small-start planning, where municipalities begin with a limited pilot district and expand incrementally, a mode well suited to cities with constrained data and budgets. It also positions buildings as urban mines, letting recyclers and policymakers anticipate where streams of recoverable steel, cement, and timber will surface decades ahead.

The researchers are candid about limitations. The dataset does not distinguish occupied homes from vacant ones, and with Kitakyushu’s vacancy rate climbing from 16.8 percent in 2013 to 19.1 percent in 2023, some surviving buildings are surely empty shells that overestimate true service life. Year-of-construction data, essential to the model, remains scarce in many Japanese municipalities due to privacy restrictions, and the team hopes their work will encourage broader data sharing. Still, the message resonates far beyond one Japanese city: as populations age and shrink across the developed world, the buildings left behind are not merely a planning problem but a vast, forecastable store of materials waiting to re-enter the economy, and the tools to map that future are now proven.

Subject of Research: Building-level demolition forecasting and construction material output prediction using 4d-GIS in a shrinking Japanese city

Article Title: Building-level demolition and material output forecasting using 4d-GIS: a case study of Kitakyushu City, Japan

Article References: Building-level demolition and material output forecasting using 4d-GIS: a case study of Kitakyushu City, Japan. (n.d.). https://doi.org/10.1007/s44498-026-00178-x

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00178-x

Keywords: 4d-GIS, material stock and flow analysis, Weibull AFT model, urban shrinkage, building lifespan, demolition probability, circular economy, construction waste, Kitakyushu, Japan, survival analysis, urban planning