Relief
We start from elevation models and cartographic data whose source is documented.
Produse Relieva
Orașe și modele 3D Prototip & cercetare
A city you can explore in depth.
We are developing 3D representations of the city that combine terrain, maps and data about the built environment. The Relieva direction is keeping that model current through field observations, guided by the areas where change shows up.
Ce reprezintă straturile
Relieful oferă baza, clădirile descriu spațiul construit, iar observațiile indică locuri de verificat. Activează și dezactivează straturile.
Relieva
From information to work done.
Urban model
We build a navigable representation of the buildings, the routes and the surroundings.
Actualizare
We are researching field data collection and how satellite observations can prioritise where to look again.
Where it earns its place.
3D exploration
Understanding the terrain and the relationship between buildings, roads and land.
Planificare
Demonstrations for routes, observation and documenting change.
Field research
Prototypes for the link between local observations and the digital model.
Relieva is a research and prototyping project. The models may include estimated heights; they are not a cadastral survey, nor a robot fleet already deployed.
Relieva in detail
What you can do with this project.
Relieva asks a simple question about a city: where can something on wheels climb, and where can it not. Chișinău has 234 metres between its lowest and its highest point, and that decides whether an autonomous delivery, a wheelchair or a ramp has anywhere to exist. The prototype measures this from public data, not from estimates.
The base is Copernicus DEM GLO-30: two one-degree tiles read from the public AWS Open Data archive, reprojected into UTM zone 35N at 25 m per pixel, with slope computed through the Horn 3×3 kernel — the same method as the standard GDAL tool. On top of it sit the street geometry and the building footprints from OpenStreetMap.
Every block on the page carries a marker: measured, sourced or proposed. The distinction is necessary because otherwise every statement would sound alike, and the one that matters to a buyer is precisely the one saying "this does not exist yet".
The slope map of the municipality
Area analysed 555 km², urban core 230 km². The distribution across four bands: 21.2% of the core below 4% slope (48.7 km²), 31.0% between 4 and 8% (71.3 km²), 23.4% between 8 and 12% (53.7 km²), 24.5% above 12% (56.3 km²). The thresholds are not chosen for looks: 8.33% is the 1:12 accessibility ramp, and 12% is the point above which sidewalk robots rated at 10–15% stop climbing.
The real slope of the streets, not of the ground
7,929 street segments with the slope computed individually, median 3.7%. Of the 156 named streets longer than 400 m, exactly one exceeds 8% on average. The conclusion overturns the intuition: the ground is rough, the roadway is not — the engineers laid the roads along the contour lines.
The centre in 3D, with the provenance of the heights kept separate
9,506 building footprints from OpenStreetMap (snapshot 2026-07-02), each placed on the ground elevation beneath it. Of these, 1,128 have a height declared in OSM; the other 8,378 are estimated from the number of floors × 3.2 m. The two categories are coloured differently in the scene, precisely so that a measurement is not mistaken for an estimate.
The view from a robot's eyes
A camera 65 cm above the pavement, placed at the real coordinates of a street in Chișinău, pointed along the roadway, with the real buildings at their real elevations. Scale 1:1, with no vertical exaggeration — unlike the terrain scene, where the ×3.5 exaggeration is written under the image.
Sentinel imagery fetched on load
Three live layers from Copernicus over a fixed area of Chișinău: Sentinel-2 true color (10 m), Sentinel-2 bare soil / disturbed surface (10–20 m) and Sentinel-1 GRD VV radar (10 m, IW mode, DV polarisation, with GAMMA0 terrain correction and orthorectification). The window is the last 30 days.
An elevation profile of the city
A west–east section 22.9 km long through the centre, 300 samples, between 36.0 and 215.4 m. The full model covers 23.4 × 25.05 km, with terrain between 24.0 and 257.6 m.
Data and operation
What goes into the system. What has to be checked.
- The trap that matters: GLO-30 is a surface model
- Copernicus DEM GLO-30 is a DSM, not a DTM — in a dense block it returns the elevation of the roof or of the tree canopy, not of the roadway. It raises no error; the numbers simply come out wrong. Concretely, while this project was being built it produced a slope of 13.5% on Bulevardul Ștefan cel Mare, which is practically flat: two points 30 m apart, one of them landing on a cornice. For the same reason, the roads came out a median of 13 m below the buildings and disappeared between the blocks in the 3D scene.
- What we did about it, and what we chose not to hide
- For the 3D geometry we use an approximated ground surface (a minimum filter over a 5 px window at 25 m/px, that is 125 m — wider than a city block, so buildings stop counting as ground while the terraces survive). The published statistics, however, stay on the raw DSM, and the page says so: because the model includes the buildings and the canopies, in dense blocks the slope comes out higher than it really is on the roadway. The figure of 52.2% of ground below 8% slope is therefore a floor, not an optimistic reading.
- Slope is measured over a long baseline
- Two consecutive vertices of an OpenStreetMap line can be 4 metres apart — below the raster's 25 m cell. Differencing them measures the noise of the raster, not the road. That is why the gradient is computed over a window that looks forward and backward up to a fixed distance, and returns zero if the baseline stays under 8 m.
- Measured, sourced, proposed — three different things
- Measured: the terrain, the slope, the profile, the geometry of the buildings on the ground. Sourced: the OpenStreetMap footprints and street geometry, with contributor attribution and the ODbL licence. Proposed: the rescanning loop through courier fleets, the robots, the drones, the accessibility map. The rover illustration explains a concept; it is not evidence of equipment delivered.
- The Copernicus quota dictates the architecture
- The area of interest of the satellite layer is fixed, not chosen by the visitor — precisely to cap how much of the allowance of 30,000 processing units per month traffic can spend. On top of that, the response is cached for 6 hours at the origin and 24 hours in the CDN, with background revalidation for a week. The satellite scene changes every few days anyway, not every few minutes.
- How current the OSM data is
- The street geometry comes from the 2026-05-06 snapshot, the building footprints from 2026-07-02. OpenStreetMap coverage and currency differ from one area to another; it is not a cadastral register and cannot be treated as one.
From exploration to deployment
How we prepare a project with Relieva.
1. We pick the territory and the question
A city, a district, a route. The question has to be one that slope decides: where a device can travel, where a ramp is needed, which route has the smallest cumulative climb.
2. We check coverage before quoting
Copernicus DEM GLO-30 covers the globe at 30 m, but OpenStreetMap coverage and quality vary a great deal. We check the density of footprints and of street geometry in the requested area before promising a result.
3. We build the layers from source
Reprojection into the zone's UTM, Horn 3×3 slope, contour lines, profile, filtered ground for the geometry. For every published figure we keep the method and the base it was computed on — raw DSM or filtered ground — because mixing the two makes the result incoherent.
4. A pilot, if field collection is involved
Any image collection in the city starts with what gets blurred at source (faces, number plates), what is kept (the derived geometry) and an impact assessment before the first camera is mounted. Moldova's data protection law — Law no. 195/2024 — applies from 23 August 2026.
Questions worth settling.
Where does the terrain come from, and how accurate is it?
Copernicus DEM GLO-30, taken from the public AWS Open Data archive — two one-degree tiles, reprojected into UTM 35N at 25 m per pixel. The native resolution is 30 m, so one cell covers about half a city block. It sees terraces, hillsides and the valley floor; it does not see the kerb, the ramp or the pothole.
Is the elevation model a DSM or a DTM, and why does it matter?
It is a DSM — a surface model. That means that in dense blocks it returns the elevation of the roof or of the tree canopy, not of the roadway. It matters because it is a silent trap: nothing raises an error, the numbers just come out wrong. It gave us a slope of 13.5% on Bulevardul Ștefan cel Mare, which is flat, because one point landed on a cornice. The fix for the geometry is a minimum filter over 125 m; for the published statistics we chose to stay on the raw DSM and to say on the page that the 52.2% figure is a floor.
Are the building heights measured or guessed?
Of 9,506 buildings, 1,128 have their height declared in OpenStreetMap. The other 8,378 are estimated from the number of floors multiplied by 3.2 m. In the scene the two categories have different colours, exactly so you can see which is which. The ground elevation under each building comes from the elevation model, not from an estimate.
Is it a cadastral map? Can I use it in an official file?
No. It has neither the precision nor the legal standing of an authorised survey — the geometry comes from OpenStreetMap, a collaborative map, and the elevation from a global model at 30 m. What it gives you is context: slope, terrain, urban volume, in a form in which planning or service-zoning decisions can be taken, followed by verification on the ground where it counts.
Are your robots already out in the city?
No. There is no fleet, there are no sensors mounted and there is no rescanning loop. They are marked "proposed" on the page, alongside the accessibility map. What exists today is the measurement layer built from public data and the scene that shows it. A real pilot would require verified hardware, assessed routes, approved operating conditions and the data protection rules settled before the first camera.
Is the satellite image real, or an illustration?
Real, and fetched when the page loads. The server asks Copernicus for a Sentinel-2 L2A true color raster at 10 m over the Chișinău area, on a 30-day window, with a 40% cloud filter, and returns it with the headers that say exactly which product it is and which interval it covers. There is also a Sentinel-1 radar layer, which sees through cloud. It is not a live image: Sentinel-2 comes back over the same place every few days.
Can it be done for another city?
The elevation model covers the globe, so the terrain and slope part reproduces almost identically anywhere. What changes is OpenStreetMap: in cities with few building footprints, the 3D scene and the street slope come out thin. Coverage is checked before the quote, not after.
Illustrative example
Where a sidewalk delivery service could exist in Chișinău
A usage scenario, with no client data and no commercial results attributed.
The starting situation
An operator wants to know in which districts it is worth starting delivery with an autonomous device, and where there is no point trying.
How it works
Slope is computed over the urban core from Copernicus DEM GLO-30 (Horn 3×3, 25 m) and classified into four bands at 4%, 8.33% and 12%. Separately, the longitudinal slope of every OpenStreetMap street segment is computed over a long baseline, so that raster noise is not measured instead of the road. The two layers are then overlaid.
Rezultatul
52.2% of the urban core sits below 8% slope, but the useful result is not the percentage — it is its shape. The hard areas gather on the escarpments between terraces, while the plateaus and the valley floor stay usable. That turns "52%" into a zoning plan: service areas are drawn along a terrace, not on a radius of kilometres. And, separately, the roadway is far gentler than the ground — the median street slope is 3.7%, and of the 156 named streets over 400 m only one goes above 8% on average.
Ce este necesar:Un instantaneu OpenStreetMap cu acoperire decentă a geometriei stradale în zona vizată. Rezultatul e o hartă de fezabilitate, nu un traseu certificat: orice rută pentru un dispozitiv autonom trebuie verificată în teren, pentru că modelul nu vede borduri, rampe sau obstacole.
Ways of working together
Relieva, in the context of your organisation.
Planning in a geographic context
Geographic models and layers for exploring a territory and discussing scenarios. Estimated data is kept distinct from verified measurements.
Private companies
We define a pilot around one real process: users, data, integrations, costs and acceptance criteria. Expansion follows once the result has been assessed.
Public institutions and state-owned companies
We establish the requirements for accessibility, hosting, data protection and interoperability. Any connection to services run by Moldova's e-Governance Agency (AGE) or its state information-technology service (STISC) requires eligibility, access and approvals to be validated.
These are adaptation scenarios, not statements about existing contracts or partnerships. The proposed capabilities are confirmed within the project's scope of work.
Discută un pilotWe look beyond the map, too.
See how we use Copernicus observations and how we explore satellite orbits.
Part of an ecosystem.
Inteligență geospațială
FlowMind
FlowMind organizează hărți, surse publice și produse Copernicus într-un spațiu de analiză.
Implementare specializatăPlatformă civică
sesizari.md
sesizari.md organizează sesizările despre spațiul public: loc, categorie, descriere și materiale relevante.
Platformă publicăInteracțiune vizuală & agenți
World Agent
World Agent este un demonstrator pentru interacțiunea în limbaj natural cu o scenă vizuală.
Demonstrator de cercetareWhat would you want to work better?
Tell us about your process. Together we decide what is worth building, what we can connect, and how we check the result.