{"id":90469,"date":"2026-10-09T15:16:08","date_gmt":"2026-10-09T07:16:08","guid":{"rendered":"https:\/\/wp-productionenv-bjg9h2g2bgg5b8aa.southeastasia-01.azurewebsites.net\/?p=90469"},"modified":"2026-10-09T15:21:49","modified_gmt":"2026-10-09T07:21:49","slug":"can-satellite-imagery-detect-early-stage-construction","status":"publish","type":"post","link":"https:\/\/starpath.global\/faqs\/can-satellite-imagery-detect-early-stage-construction\/","title":{"rendered":"Can Satellite Imagery Detect Early-Stage Construction?"},"content":{"rendered":"

Satellite imagery can reveal construction activity before a building is completed\u2014and sometimes before its structure appears. Land clearing, exposed soil and expanding work areas can provide early signals. But reliably identifying a small foundation or building extension requires much finer imagery than detecting a large construction site. The result depends on the size of the change, image resolution, observation frequency and how the imagery is interpreted.<\/p>\n

A question on Reddit\u2019s GIS forum illustrates the challenge. A user working with a local government agency wanted to detect suspected unauthorized construction at an early stage, including activity at a scale of 2\u20135 meters. Commercial imagery was beyond the prototype budget, leaving free Sentinel data as the starting point. Would Sentinel-2\u2019s 10-meter imagery be enough?<\/p>\n

For reliably identifying individual features that small, Sentinel-2 alone is generally insufficient. For screening larger areas of ground disturbance and directing closer inspection, it can still be useful.<\/p>\n

What Does \u201cEarly-Stage Construction\u201d Look Like from Space?<\/h2>\n

Early construction monitoring does not necessarily begin with detecting a building. It may begin with detecting changes to the land where a building will eventually stand.<\/p>\n

A 2024 study published in Science of Remote Sensing<\/em> investigated broad-area construction screening using Landsat and Sentinel-2 time series. Its approach tracked changes in surface composition, including increases in exposed soil during ground preparation. The researchers described the screening results as a way to narrow down candidate locations for subsequent examination with very-high-resolution imagery.<\/p>\n

This distinction matters. A large cleared plot can be visible even when the individual foundations within it are too small to identify. Conversely, an extension built on an already paved urban parcel may create little detectable change in coarse imagery.<\/p>\n

The monitoring question should therefore be specific: are we looking for newly disturbed land, a construction site, an individual structure or a small modification to an existing building? Each requires different evidence.<\/p>\n

Why 10-Meter Imagery Struggles with Small Targets<\/h2>\n

Sentinel-2 has four spectral bands at 10-meter resolution, six at 20 meters and three at 60 meters. Its visible red, green and blue bands and broad near-infrared band are sampled at 10 meters; the shortwave-infrared bands are sampled at 20 meters.<\/p>\n

A simple geometric example explains the problem. A 5 \u00d7 5-meter foundation occupies 25 square meters, compared with 100 square meters for a nominal 10 \u00d7 10-meter pixel. Even if it falls entirely within one pixel, it occupies only a quarter of that pixel\u2019s area. Its signal is mixed with the surrounding ground, vegetation or paving.<\/p>\n

That calculation is an illustration, not a detection threshold. Contrast, image sharpness, pixel alignment and the surrounding landscape all affect visibility. A small feature may alter a pixel\u2019s brightness without providing enough information to establish its shape or identify it as construction.<\/p>\n

Detecting a signal is different from recognizing an object.<\/strong> A changed pixel may indicate disturbance, but it does not necessarily reveal whether the cause is a foundation, soil removal or another activity.<\/p>\n

Which Resolution Fits Which Task?<\/h2>\n

The following comparison is practical selection guidance, rather than a set of guaranteed detection limits. Landsat provides 30-meter multispectral imagery, Sentinel-2 includes 10-meter bands, and commercial products offer finer detail. Research using 0.3\u20130.5-meter WorldView imagery has demonstrated construction-area classification in a Berlin case study.<\/p>\n

\n
\n\n\n\n\n\n\n\n\n\n
Imagery resolution<\/th>\nChanges it is better suited to assess<\/th>\nMain limitation<\/th>\n<\/tr>\n<\/thead>\n
30 m<\/td>\nExtensive land conversion, large earthworks and broad development patterns<\/td>\nSmall sites and individual structures are poorly resolved<\/td>\n<\/tr>\n
10 m<\/td>\nLarger areas of clearing, exposed soil and sustained site disturbance<\/td>\nSmall foundations, narrow trenches and building extensions are generally unsuitable for reliable identification<\/td>\n<\/tr>\n
3\u20135 m<\/td>\nSmaller disturbed plots, developing access routes and changes across larger building footprints<\/td>\nFeatures only a few meters wide still contain very little spatial detail<\/td>\n<\/tr>\n
1\u20132 m<\/td>\nIndividual building changes and more detailed site layouts<\/td>\nSmall or low-contrast features can remain ambiguous<\/td>\n<\/tr>\n
0.3\u20130.5 m<\/td>\nDetailed review of visible foundations, roof changes, site equipment and localized disturbance<\/td>\nVisibility still depends on shadows, viewing angle, obstruction and image quality<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Higher resolution provides more evidence, but the advertised pixel size needs careful interpretation. For example, Planet documents SkySat orthorectified products sampled at 50 centimeters per pixel, while its technical FAQ distinguishes that output sampling from the resolution of basic products. Buyers should check native sensor detail, processing and positioning accuracy alongside the delivered pixel size.<\/p>\n<\/div>\n

Observation Frequency Can Be as Important as Resolution<\/h2>\n

A detailed image acquired after construction is completed cannot provide an early warning.<\/p>\n

ESA lists a five-day revisit for Sentinel-2, but revisit frequency is not a guarantee of a usable, cloud-free observation every five days. Optical monitoring requires an unobstructed view of the site. Cloud cover can therefore lengthen the interval between useful images.<\/p>\n

For a monitoring project, the relevant interval is the time between usable observations, followed by the time needed to process, review and deliver an alert.<\/p>\n

Consider a hypothetical case: one clear image shows an unchanged parcel, while the next clear image, three weeks later, shows excavation. The imagery brackets the visible change between those observations; it does not establish the exact day work began.<\/p>\n

Synthetic aperture radar, or SAR, can help maintain observations through cloud cover and at night. However, it measures radar responses rather than producing an ordinary photograph. It adds another source of evidence without automatically making small construction features easy to interpret.<\/p>\n

Change Detection Needs Context<\/h2>\n

A construction monitoring system must answer more than \u201cwhat changed?\u201d<\/p>\n

The 2024 broad-area screening study reported false positives that included vegetation clearing and mining projects. These activities can produce surface changes resembling construction preparation. A bare-soil alert therefore needs contextual review before it is classified as a construction event.<\/p>\n

Image alignment also matters. A small displacement between two images can create apparent changes along existing buildings or roads. A technical account published by CIGRE describes accurate image co-registration as a prerequisite for reducing false alarms in infrastructure monitoring.<\/p>\n

AI can help compare images and prioritize candidates, but its usefulness should be measured against the actual task. A model that finds large cleared sites has not necessarily demonstrated that it can detect a small urban extension. Enlarging an image or generating a sharper-looking version also does not, by itself, establish that a suspected feature was observed.<\/p>\n

Finally, physical change and authorization status are separate questions. Identifying potentially unauthorized construction requires linking the observed activity to parcel boundaries, permit records and the relevant administrative review.<\/p>\n

A Practical Example: Construction Near Power Infrastructure<\/h2>\n

Construction detection also matters beyond urban planning. Groundworks near electricity infrastructure can require prompt review.<\/p>\n

A CIGRE technical article describes monitoring for France\u2019s electricity transmission operator, RTE, using Sentinel-1 radar, Sentinel-2 multispectral imagery and commercial high-resolution optical data. Radar and multispectral observations identified candidate activities; detailed optical imagery helped interpret them. Analysts reviewed the findings before operators assessed the events and arranged field visits.<\/p>\n

The same account highlights an operational limitation: many detected activities were already known or did not pose a relevant threat. Finding construction and identifying construction that requires action are different tasks.<\/p>\n

How to Choose Data for a Construction Monitoring Project<\/h2>\n

A sensible pilot starts with a measurable requirement: the smallest change that matters, the area to cover and the time available to respond.<\/p>\n

For broad screening, free Sentinel-2 imagery can help test whether larger, known construction sites produce recognizable changes. For small suspected structures, high-resolution imagery should be evaluated from the beginning. Waiting for a coarse-resolution alert may miss sites that never produce a sufficiently large disturbance.<\/p>\n

A useful pilot should:<\/p>\n