Ground Truth for Maps: What Can Only Be Verified on Site

Ground truth, in a geospatial context, refers to information collected through direct, physical observation at a real-world location, used to verify or correct data gathered remotely or compiled from secondary sources. It is the reference layer against which satellite imagery, open registries, and crowdsourced map edits get checked for accuracy. The term borrows its name from remote sensing, where analysts needed field measurements to calibrate what a sensor thought it was seeing from orbit — and the same logic now governs how delivery platforms, navigation apps, and public mapping agencies decide whether a point of interest is actually where the database says it is.

The European GIS market is not a niche concern. IMARC Group puts the Europe geographic information system market size at USD 3.0 billion in 2025 , with mapping as the single largest function, holding around 50.2% of market share in 2025 [1]. Renub Research frames growth even more steeply, projecting the Europe GIS market to grow from USD 3.63 billion in 2024 to USD 8.44 billion by 2033, at a CAGR of 9.82% [2]. Behind both figures sits the same operational problem: software and satellites can display a map, but they cannot confirm that the shop on the corner still exists, that the delivery entrance is where the facade suggests, or that the address assigned to a rural plot actually resolves to a driveway.

This piece works through what ground truth verification covers, what the open and public data ecosystem — INSPIRE, national address registries, OpenStreetMap — already solves reasonably well, and where it structurally cannot substitute for someone standing at the location with a device and a checklist.

What Does "Ground Truth" Actually Mean in a GIS Context?

Ground truthing refers to the process of collecting direct field observations that serve as the reference standard against which remote, derived, or crowdsourced geodata is evaluated. The concept comes from the ISO 19157 data quality standard, which splits geographic data quality into distinct, separately measurable elements: positional accuracy, thematic accuracy, completeness, logical consistency, and temporal quality (IPGH, 2021) [via INSPIRE technical guidance]. Positional accuracy defines the closeness of locational information to the true position, split into absolute accuracy — how close a coordinate is to its true value — and relative accuracy, how close features are to each other's true relative positions . That distinction matters more than it sounds. A POI can sit within a few meters of its true coordinate and still be wrong in every way that matters to a courier, if the listed entrance, floor, or operating hours no longer reflect reality.

INSPIRE — the EU directive (2007/2/EC) that harmonizes spatial data across member states — defines addresses formally as the "location of properties based on address identifiers, usually by road name, house number, postal code." The specification requires that the position of the address be represented by the coordinates of the actual location with the best available accuracy [4]. "Best available" is doing a lot of work in that sentence. It concedes, structurally, that accuracy is a moving target set by whatever data collection method produced the record — a geocoded centroid, a cadastral survey, or a technician with a GNSS receiver standing at the front door.

Ground truth is not a stylistic choice but a structural requirement: without a physically verified reference point, there is no way to know whether a positional or thematic error exists at all.

Why Do Public Address Registries Still Need On-Site Verification?

Every EU country covered by this analysis runs its own address infrastructure, and none of them update at the same pace or through the same mechanism. France's Base Adresse Nationale is a crowdsourced-plus-authoritative national database maintained jointly with local communes. The Netherlands runs the Basisregistratie Adressen en Gebouwen (BAG), a mandatory municipal registry with legal status as a "base registry." Switzerland maintains address and building data through the Federal Register of Buildings and Dwellings, administered at municipal level and consolidated federally. Belgium splits the job regionally — Flanders, Wallonia, and Brussels each run separate address authorities. Germany manages cadastral and address data through sixteen separate state-level surveying administrations, with harmonization still in progress under AdV coordination. INSPIRE standardizes the schema these systems expose, not the frequency or rigor with which any one of them gets re-checked on the ground.

A peer-reviewed study testing geolocated address quality in urban areas found that statistical results show sufficient positional accuracy, about 10–40 meters, when there is a match between address datasets , and that this positional accuracy value is similar to other related studies that confirm similar accuracy ranges [3]. Ten to forty meters is workable for a regional heat map. It is not workable for a last-meter delivery instruction, a wheelchair-accessible entrance flag, or an emergency dispatch address. The same research notes that the OSM spatial data source is unique and does not share address data with geocoders directly, though OSM can sometimes be completed using authoritative address data from government organizations [3]. That interdependency — open data feeding proprietary geocoders feeding navigation apps — means an error introduced at the registry level can propagate for years before anyone physically checks it.

Consider a rural property outside Lyon where the cadastral address point sits at the parcel's legal centroid rather than the actual driveway entrance, a known artifact of how many European cadastral systems assign coordinates. A geocoder built on that registry will route a delivery vehicle to a field boundary, not a door. No amount of re-processing the same dataset fixes this — the correction has to come from someone standing at the real entrance with a GNSS reading.

What Can Satellite and Aerial Imagery Not Tell You?

Remote sensing answers "what is there geometrically" extremely well: building footprints, road centerlines, land cover, and large infrastructure are all visible from above and improving steadily with higher-resolution commercial satellite constellations. What imagery cannot answer is anything behavioral, operational, or interior. A storefront's facade looks identical whether the business inside closed eighteen months ago or opened yesterday. A loading dock's physical outline gives no information about whether it is currently accessible, under renovation, or restricted to a different carrier. Indoor layout — which floor a tenant occupies, where an entrance actually connects to a corridor — falls entirely outside what any aerial or satellite pass can resolve; our piece on indoor positioning and its real accuracy covers why GNSS signal itself breaks down the moment a device crosses a threshold.

This is also where completeness and logical consistency, the other two ISO 19157 quality elements, become the binding constraint rather than positional precision. A dataset can be geometrically flawless and still catastrophically incomplete if half the POIs it lists have closed. Completeness describes the degree to which geographic features, their attributes, and their relationships are included or not included in a spatial dataset [5] — and completeness, unlike position, decays continuously rather than being fixed at the moment of survey.

How Do Delivery and Navigation Platforms Handle the Verification Gap?

Logistics and navigation platforms operationalize ground truth differently than public agencies do, because their error cost is immediate and measurable: a failed delivery, a misrouted courier, a stranded rider. These companies run their own field verification layers on top of licensed and open base maps — photographing storefronts, confirming access points, checking operating hours against posted signage, and flagging duplicate or defunct listings. The operational logic mirrors what our field survey methodology work covers in more depth: sampling strategy and routing determine whether a verification program catches the errors that matter or just re-confirms the easy cases near depots and city centers.

Verification Method	What It Confirms	What It Misses	Typical Refresh Cadence
Satellite / aerial imagery	Building footprint, road geometry, land cover, large infrastructure changes	Operational status, interior layout, signage, accessibility	Weeks to months, provider-dependent
National address registry (BAN, BAG, GWR, regional cadastres)	Legal address existence, administrative boundaries, parcel linkage	Whether a business at that address is open, exact entrance location	Varies by commune/municipality; often irregular
Crowdsourced mapping (OpenStreetMap)	Road network detail, POI tagging at volunteer density	Consistency across regions, freshness in low-edit areas	Continuous but geographically uneven
On-site field verification	Current operational status, exact position, photographic evidence, attribute accuracy	Scale without recurring cost; requires re-visit for currency	Program-defined, typically weeks to quarters

The table makes the trade-off explicit: nothing in the open or automated layer captures operational status directly, and nothing in the field-verified layer scales for free. The two are structurally complementary rather than competing approaches.

What Does a Field Verification Program Actually Record?

A well-run on-site data collection program does not just confirm "yes, this exists." It produces a structured, repeatable record built to the same rigor the ISO 19157 framework expects of any spatial dataset: a stable identifier, a position with a confidence measure, a timestamp, and photographic evidence tying the record to a specific visit. Teampl ran a nationwide field program across the Netherlands covering more than 1,500 EV charging stations, spanning urban, suburban, and rural areas rather than sampling only major cities. For each station, the program recorded station name, operator, address, number of charging points, connector types, charging capacity, and pricing as a structured dataset, with on-site photographic evidence attached to every location. That structure is what turns a field visit into usable geodata rather than a one-off confirmation: every record carries the same schema and the same evidentiary backing, which is what lets a client trust the 1,500th record as much as the first.

For instance, a charging operator's public API might list a station as "available" when, on the ground, two of its four connectors have been physically disconnected for months pending a grid upgrade — a thematic accuracy failure invisible to any remote check but obvious within thirty seconds of arriving on site. Freshness compounds the same way indoor maps do; our piece on how fast indoor maps decay walks through the re-survey economics in more detail, and the same decay curve applies to any attribute — connector type, pricing, accessibility — that changes independently of the building it's attached to.

Where Does Ground Truth Verification Hit Its Limits?

Field verification is not infinite-resolution truth either. A technician visiting a location records a single moment; a business can close the following week, a charging station can go offline the next day, and a seasonal retailer can shift hours by month. Verification establishes a confidence interval anchored to a visit date, not a permanent fact. This is why serious programs treat sampling design, re-visit cadence, and routing as core methodology rather than afterthought — covered at length in our dedicated piece on field survey methodology — because a single comprehensive sweep degrades in value from the day it finishes.

Cost is the other real constraint. Verifying every address in a country at monthly intervals is not economically rational for most use cases; the discipline is in deciding which subset of locations — high POI-density urban cores, high-churn commercial strips, safety-critical infrastructure like EV charging or emergency access points — justifies the recurring field cost, and which can rely on periodic open-data refresh instead. Germany's GIS market alone illustrates the scale involved: Germany was the prime player in the Europe geographic information system market with a 25% share in 2024 [7], and a market that size cannot run exhaustive, continuous field verification everywhere at once. Prioritization, not universality, is the operating principle.

Global demand for the analytics layer that sits on top of verified geodata keeps expanding regardless. The Business Research Company's geospatial analytics market report notes that the geospatial analytics market size has grown rapidly in recent years [6], a trend that increases pressure on the underlying data pipeline rather than reducing it — better analytics on stale or unverified inputs just produces more confident wrong answers, faster.

Verdict

Public registries and crowdsourced maps solve coverage. Satellite imagery solves geometry. Neither solves currency, operational status, or the gap between what a database says and what a front door looks like today. That gap is not a temporary technology limitation waiting on better sensors — it is structural, because operational reality changes faster than any remote system can re-observe it, and someone has to go check.

Your next hire isn't a vendor. It's a data team.

Frequently Asked Questions