Satellites Reveal Malaysia’s Hidden Thirst for Sand: 5.8 Billion Tonnes and Counting

Sand is the most extracted solid material on Earth, and yet it is also one of the least understood. Unlike gold or copper, it is cheap, abundant and largely unregulated, which means that in many countries nobody keeps reliable records of how much of it is dug out of rivers, coastlines and quarries each year. A new study published in the Journal of Industrial Ecology has taken an unconventional route around this problem: instead of trying to measure sand extraction directly, a team of researchers led by the British Geological Survey counted what Malaysia built. By combining high-resolution satellite data with deep learning, they reconstructed nearly four decades of construction across roughly 9.1 million buildings, arriving at a striking figure — modern Malaysia’s built environment has consumed approximately 5.8 billion tonnes of construction materials, with sand at the heart of the story.

The research matters because sand has quietly become a global sustainability crisis. UNEP estimates that global construction sand consumption runs to 40–50 billion tonnes per year, a figure derived not from direct measurement but from a proxy: cement production. The logic seems sound, since sand and cement are typically consumed together in concrete. Cement is a high-value commodity, produced in capital-intensive plants and taxed on sales, so many countries track it carefully. Sand, by contrast, is often extracted with little state oversight, feeds largely local markets and supports a substantial informal sector with low barriers to entry. The result is chronic data blindness. Reported sand production in Vietnam, for example, falls well short of known consumption, and mismatches in import and export statistics are common worldwide.

The Malaysian study exposes just how fragile the cement proxy really is. In the United States, where both cement and sand-and-gravel production are well documented, the two track each other closely, with a regression R² of 0.76. But across other countries the relationship collapses. Germany shows almost no correlation at all, and no consistent pattern emerges when national pairs are compared, suggesting the link depends on industry structure, trade patterns and construction practices rather than any universal constant. When the researchers tested the proxy against Malaysia’s own official statistics, cement-based estimates of sand production came out far too high in some years and diverged sharply from reported production after 2012 — precisely the period when sand demand shifted away from buildings and toward roads and coastal land reclamation, sectors the cement proxy cannot see.

The team’s alternative approach works backwards from consumption. Using the Google Open Buildings dataset, generated by a U-Net deep learning model applied to high-resolution imagery from Maxar Technologies and CNES/Airbus, they mapped building footprints across the entire country with a detection confidence threshold of 0.75. Building heights came from a separate deep learning architecture, BHE-NET, which fuses optical data from Sentinel-2 and radar data from Sentinel-1 to produce height maps at 10-metre resolution — a considerable advance on the 90-metre World Settlement 3D product. Validated against roughly 1,000 buildings measured by the ICESat-2 satellite lidar, the height product showed a root mean square error of about 9 metres and a mean absolute error of about 5 metres.

Those error figures sound alarming, but the study’s authors explain why they matter less than they appear. Nearly 88 percent of Malaysian buildings stand below 6.6 metres — broadly one to two storeys — and fewer than 1 percent exceed 14 metres. Because the national total is dominated by low-rise structures, a 3-to-5-metre height error at the individual building level often represents less than a single storey, and with 9.1 million buildings in the sample, over- and under-estimations tend to cancel out. The net effect on national stock totals is therefore strongly reduced, even though uncertainty remains at the level of any single structure. It is a textbook illustration of how big data can tame noisy measurements.

To convert geometry into tonnage, the researchers needed to know what each building is made of. They classified buildings along two axes: urban versus rural setting, using a population density threshold of 500 persons per square kilometre from NASA’s Gridded Population of the World dataset, and residential-commercial versus industrial use, derived from land cover maps produced by a retrained U-Net model applied to Landsat composites from 1988 through 2018, achieving overall accuracies above 85 percent. The land cover time series also served a second purpose: by intersecting building footprints with the year built-up land first appeared, each building could be assigned an approximate construction decade. Material intensity coefficients — tonnes of sand, gravel, brick clay, tile and cement per unit of floor area — adapted from a comprehensive Chinese database were then applied, with floor heights of 3 metres for residential and commercial buildings and 3.5 metres for industrial ones.

Buildings alone, however, do not exhaust a nation’s appetite for sand. The team added nearly 1.5 million linear infrastructure elements from OpenStreetMap, covering almost 400 million kilometres of roads and railways, dated by intersecting each segment with the land cover time series and assuming a 20-year maintenance cycle for roads. They also accounted for land reclamation, a major sand consumer in a country actively expanding its coastline for ports and urban development. Between 1991 and 2021, Malaysia reclaimed or plans to reclaim 82.64 square kilometres of land, more than half of it in the final decade. Using a conservative loading of 2.25 tonnes per square metre and a sand density of 1.5 tonnes per cubic metre, the researchers folded these megaprojects into their national consumption model at decadal resolution.

The results reveal a country transformed. Built-up surface in 2020 covered roughly 1,830 square kilometres — less than 1 percent of Malaysia’s land area — but buildings now account for about 80 percent of that footprint, up from just 49 percent in 1988. Almost 40 percent of the entire building stock predates the 1990s, and nearly as much sand, gravel and crushed rock is locked into buildings erected after 1998 as in everything built before. Regional contrasts are sharp: in the urbanised state of Selangor, sand dominates consumption, while rural Sarawak, with older buildings, consumes proportionally more bricks and tiles. Across the country, brick and tile use has declined as concrete has taken over, and gravel and crushed rock use has risen — trends that only become visible with this kind of spatial and temporal resolution.

Perhaps the most consequential finding concerns the future. The model estimates that around 326 million tonnes of sand and 356 million tonnes of crushed rock are embedded in Malaysian buildings constructed before 1988 — structures now approaching the end of their typical 50-year lifespan. With annual production running at roughly 30 million tonnes of sand and 100 million tonnes of crushed rock, these ageing stocks could become a significant secondary resource if recycling and reuse policies are put in place. More than 680 million tonnes of sand and crushed rock sit in pre-1988 buildings in total, a latent urban mine waiting for the right infrastructure. The temporal dimension of the model, the authors argue, allows policymakers to forecast when materials will come back onto the market, supporting circular economy planning in a way that static national statistics never could.

The methodology is not without limitations. The material intensity coefficients, the largest source of uncertainty, are drawn from Chinese rather than Malaysian data — a defensible choice given documented similarities in modern concrete building typologies between the two countries, but one that could introduce errors for older and rural structures, where estimates can vary by 30 percent or more. The approach also captures only new construction on previously undeveloped land, which works for a country in an active phase of urbanisation but may not translate to mature cities building on brownfield sites. And some sand flows remain stubbornly opaque: Singapore’s reported imports of Malaysian sand in 2017 exceeded Malaysia’s entire reported production that year, hinting at unrecorded trade shrouded in political sensitivity. Even so, the study demonstrates that when the full value chain — buildings, roads, reclamation — is included, satellite-derived consumption estimates correlate well with official production figures, while cement-based proxies fail. For the dozens of countries where sand data simply does not exist, that is a genuinely scalable way forward.

Subject of Research: Earth observation-based dynamic material stock modelling of construction sand consumption in Malaysia from 1989 to 2020

Article Title: Using earth observation for national high resolution dynamic material stock modelling through time; assessing sand consumption in Malaysia from 1989 to 2020

Article References: Bide, T., Novellino, A., Watson, C. S., Cai, B., Holland, L., Shao, Z., Wang, S., & Othman, N. S. (2026). Using earth observation for national high resolution dynamic material stock modelling through time; assessing sand consumption in Malaysia from 1989 to 2020. Journal of Industrial Ecology. https://doi.org/10.1007/s44498-026-00109-w

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00109-w

Keywords: sand consumption, earth observation, material stock modelling, Malaysia, deep learning, building footprints, construction materials, material flow analysis, land reclamation, cement proxy, circular economy, remote sensing