Field Survey Methodology for Geodata: Sampling, Routing and QA

Field survey methodology for geodata is the set of decisions a team makes before, during, and after a site visit: which locations to check, in what sequence, and how to verify the result before it enters a live dataset. It sits between desk research (crawling websites, scraping listings, pulling satellite tiles) and the moment a human or a sensor stands in front of a real address and confirms or contradicts what the database says. Done well, it's unglamorous and repeatable. Done badly, it produces datasets that look complete and are quietly wrong.

This matters more in Europe than the market-size numbers alone suggest, because the region runs on a patchwork of national mapping agencies, address registers, and coordinate systems rather than one federal standard. France, Germany, Switzerland, Belgium and the Netherlands each have their own cadastral and addressing infrastructure, and a survey design that works in one won't translate cleanly to the next without adjustment. The methodology below — sampling, routing, QA — is the same three-part skeleton regardless of country. What changes is who owns the ground truth you're checking against.

None of this is new science. Statisticians have argued about stratified sampling for a century and national mapping agencies have run field verification programs for longer than GIS software has existed. What's changed recently is the scale: cheaper mobile capture, routing software that used to require a logistics department, and ISO's 2023 rewrite of the core geographic data quality standard. The fundamentals haven't moved. The tooling around them has.

What a "field survey" actually covers in GIS work

A field survey, in the geodata sense, is any program where someone physically visits a location to record or verify spatial and attribute information — an address, a point of interest, a piece of infrastructure, a road segment. That's distinct from remote sensing, where satellite or aerial imagery stands in for a visit, and distinct from crowdsourced or scraped data, where the "verification" is really just another unverified claim from a different source.

Ground truthing refers to the act of confirming, on site, that what a dataset claims matches physical reality. It's the step that catches the gap between "a business is listed at this address" and "a business is actually operating at this address today." Desk research can flag candidates for a field visit. It can't replace the visit itself, because closures, relocations, and signage changes happen faster than most review cycles.

The Europe geographic information system market was valued at roughly USD 3.0-3.6 billion in recent estimates, with forecasts putting it at USD 5.0-8.4 billion by the early 2030s depending on methodology and scope (IMARC Group, 2026; Renub Research, 2026). Field services and data collection sit inside that figure as a cost center every GIS buyer eventually has to fund, whether they run it in-house or commission it.

Sampling: deciding where the field team actually goes

Every field program starts with a population — every address in a city, every charging station in a country, every storefront in a retail chain — and a budget that won't cover visiting all of it twice a year. Sampling is the discipline of choosing a subset that still answers the research question honestly.

Census (full enumeration) means visiting every unit in the population. It's the only approach that eliminates sampling error entirely, and it's the default when the population is small, high-value, or legally required to be complete — a national charging-infrastructure inventory, for instance, or a regulator's registry of petrol stations.

Stratified random sampling refers to dividing the population into groups — by region, category, or network type — and drawing a random sample within each group, so a small but important segment (rural fuel stations, say) isn't drowned out by a large one (urban ones). It's the standard choice when budget forces a subset but the analysis still needs to generalize to the whole population with a known confidence interval.

Risk-triggered or change-based sampling prioritizes locations flagged by a signal — a complaint, a long gap since last verification, inconsistent data from two sources — rather than sampling at random. It's cheaper per useful finding but biased by design, since it only revisits places already suspected of having a problem.

Convenience or route-of-opportunity sampling checks whatever's near wherever a team already has to be. It's the weakest method statistically but sometimes the only affordable one for low-priority, high-volume categories.

Method	Coverage logic	Typical use case	Cost profile	Main limitation
Census	Every unit, no sampling error	Legally mandated registries, small fixed populations	Highest, scales linearly with population	Expensive and slow for large populations
Stratified random	Random draw within defined subgroups	Freshness audits across a chain or network with known segments	Moderate, scales with sample size not population	Needs a reliable sampling frame to stratify against
Risk-triggered	Targets locations flagged by a signal	Complaint-driven or decay-aware re-checks	Low per visit, efficient at finding known problems	Can't estimate error rate for the whole population
Convenience / route-of-opportunity	Whatever's along an existing route	Low-priority categories piggybacking on another program	Lowest marginal cost	Not statistically representative

In practice, most mature programs blend two of these: a stratified random baseline to establish an accuracy rate, plus a risk-triggered layer that catches the worst decay between baseline rounds. A program that only ever revisits flagged locations will look increasingly accurate over time even as the unflagged majority quietly drifts — because nobody's checking it. This is the same trap that shows up in address and POI freshness programs more broadly: confidence without a baseline is just an assumption wearing a dashboard.

Routing: turning a sample list into a day's work

Once a sample is fixed, the operational question is sequencing: which locations in which order, by which mode of transport, inside which time windows. This is a variant of the vehicle routing problem, a well-studied class of optimization problems where the goal is minimizing total travel cost while respecting constraints like vehicle capacity, opening hours, and maximum shift length.

Naive routing — sorting by straight-line distance or postal code — tends to perform fine for dense urban samples and badly for anything spread across a country. Consider a national survey of roadside infrastructure: 80% of the sample sits inside five metro areas, and the remaining 20% is scattered across rural stretches with 40-60 minute drive gaps between stops. A straight-line router will send a field team back and forth across a region inefficiently; a proper routing pass clusters the rural 20% into dedicated loops and treats the urban clusters separately, because the two subsets have fundamentally different travel economics.

Three constraints matter more in practice than the optimization algorithm itself:

• Opening hours and access windows: a POI verification visit timed outside business hours produces a false negative, not missing data — the record gets marked "closed" when it's actually just closed for the night.
• GNSS accuracy by environment: consumer GPS typically resolves to 3-5 meters in open sky but degrades substantially in urban canyons, dense tree cover, and underground or indoor environments, where multipath reflection and signal blockage push errors well beyond that. Teams surveying metro stations, parking structures, or dense old-town cores need a capture workflow that doesn't assume open-sky accuracy.
• Legal and physical access: gated communities, private industrial sites, and some multi-tenant buildings require prior arrangement; a route plan that ignores this returns field staff empty-handed on a meaningful share of stops.

For indoor and GNSS-degraded environments specifically, accuracy figures quoted for outdoor consumer GPS simply don't transfer — a separate set of assumptions, covered in more depth in indoor positioning without GPS, applies once a survey moves underground or inside a large structure.

QA: the three layers that catch different errors

Quality assurance for field geodata isn't one checkpoint — it's a stack. Each layer is designed to catch a category of error the others miss.

Layer	What it catches	Typically performed by	Common trigger
Capture-time validation	Missing fields, implausible coordinates, no photo evidence	Mobile app logic, the field agent	Automatic, on every submission
Back-office review	Inconsistent attribute values, outlier flags, duplicate records	QA analyst or rules engine	Scheduled batch review, or statistical outlier detection
Cross-source reconciliation	Disagreement with an authoritative register (cadastre, national address file)	Automated matching against reference data	Periodic reconciliation run
Independent re-check / audit sample	Systematic bias in the field team or process itself	A second, independent visit to a random subset	Fixed percentage of total sample, e.g. 5-10%

The data quality standard underlying most of this structure, ISO 19157, defines the vocabulary: completeness (commission and omission), logical consistency, positional accuracy, thematic accuracy, temporal accuracy, and usability. ISO replaced the original 2013 edition with ISO 19157-1:2023 in April 2023, splitting the standard into parts and tightening the general requirements (International Organization for Standardization, accessed October 2026). The standard is explicit that it doesn't set a minimum acceptable quality level — that's left to whoever's buying or producing the data (ISO, 2026).

The EU's INSPIRE directive leans on that same ISO framework. It requires member states to report data quality using specific elements — completeness (commission and omission), logical consistency, and positional accuracy among them — but it doesn't mandate a pass/fail threshold (INSPIRE Technical Guidelines, 2026). That's a meaningful gap: a French dataset and a German dataset can both claim INSPIRE conformance while describing very different real-world accuracy, because the directive standardizes the reporting format, not the result.

Standards and coordinate systems: what's actually binding

INSPIRE recommends ETRS89 as the reference datum for pan-European spatial data, specified either as geodetic coordinates on the GRS80 ellipsoid or projected using the ETRS89 Lambert Conformal Conic or Transverse Mercator systems (INSPIRE Data Specifications, 2026). In practice, national agencies still run their own local systems for high-precision work — France's IGN, Germany's BKG, Belgium's NGI, and the Netherlands' Kadaster each maintain national address and cadastral registers that predate INSPIRE and remain the actual source of truth field teams reconcile against.

Switzerland isn't an EU member and INSPIRE doesn't bind it, so swisstopo runs its own coordinate reference system (CH1903+/LV95) and its own data specifications entirely outside the INSPIRE framework. A field survey spanning Switzerland and its EU neighbors has to handle that boundary explicitly rather than assuming one coordinate system works across the whole route.

Outside Europe, the picture is less harmonized still. The United States has no direct INSPIRE equivalent; address and parcel data quality varies by county and state, and there's no federal mandate comparable to what INSPIRE imposes on EU member states. Market researchers tracking the broader geospatial analytics space put the global market at roughly USD 103.75 billion in 2025, growing to an estimated USD 122.96 billion in 2026 — a scale that dwarfs Europe's GIS-specific figure and reflects a much more fragmented regulatory landscape across regions (The Business Research Company, 2026).

A worked example: mapping EV charging infrastructure nationwide

Consider a program built to document every public EV charging station in a country — not just the ones in the five largest cities. The sampling decision here tends toward census rather than stratification, because the research question ("is this register complete and accurate?") can't be answered by sampling if the register itself might have gaps a sample would never surface.

Teampl ran exactly this kind of program across the Netherlands, covering 1,500+ charging stations in urban, suburban, and rural areas rather than sampling major cities only. Routing had to account for the fact that rural stations are sparser and farther apart, so the loop structure differed meaningfully between dense city center sweeps and long-haul rural circuits. For each location, the team 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 record.

The QA layer standardized that output so every record carried the same field structure and a supporting photo, which matters because "charging capacity" and "connector type" are exactly the kind of attributes that get entered inconsistently when field notes aren't forced into a fixed schema at capture time. A national, all-terrain sample like this is also where routing software earns its keep: a rural EV station visited at the wrong time of day, or missed because a naive router assumed urban drive times, is a data gap that looks identical to "this station doesn't exist" in the final dataset.

Where the methodology changes by market

The three-part skeleton — sample, route, verify — holds everywhere. What differs is the authoritative register a field team reconciles against, and how open that register is by default.

Market	Authoritative address/cadastral source	Open-data default	Field survey note
France	IGN / Base Adresse Nationale (BAN)	BAN is open and widely used as a reconciliation reference	Strong open baseline simplifies cross-checking field results
Germany	BKG, federated by Land (state)	Partial — varies by state, less uniform than France or NL	Reconciliation often needs state-level, not federal, sources
Switzerland	swisstopo	Open but on a separate CRS, outside INSPIRE scope	Non-EU; coordinate conversion is a real step, not a formality
Belgium	NGI, split regionally (Flanders, Wallonia, Brussels)	Regional portals, not a single national one	Survey design should assume three regional data regimes
Netherlands	Kadaster / BAG (Basisregistratie Adressen en Gebouwen)	High — BAG is a frequently cited open-data benchmark	Strong reference data lowers reconciliation cost significantly

The practical consequence: a survey program built once for the Netherlands, where BAG gives a high-confidence reconciliation baseline, will need real adjustment before it works the same way in Germany or Belgium, where the authoritative source is fragmented by state or region. This is the piece that off-the-shelf survey software tends to underplay — the field methodology travels, the reference data doesn't.

What field survey methodology cannot fix

A well-designed sampling, routing, and QA stack still can't outrun data decay. Every verified location starts aging the moment the visit ends — a business closes, a charger gets decommissioned, a shop changes hands — and the only defense is a re-verification cadence built into the program from the start, not bolted on after the first complaint. Methodology answers "how do we verify accurately." It doesn't answer "how often do we need to," which is a separate, often underfunded, design decision.

It also can't substitute for access. No routing algorithm gets a field agent past a locked gate or into a building without a scheduled appointment, and no QA layer recovers data that was never collectable in the first place. Programs that budget for 100% visit success rates are budgeting against reality; a realistic plan assumes a non-trivial share of attempted visits will fail on the first pass and need a second routing cycle.

What to check before commissioning a field survey

• Sampling frame: do you have a reliable list to sample from, or does the research question actually require a census because the list itself might be incomplete?
• Reconciliation source: what authoritative register will field results be checked against, and how current is it in each country you're covering?
• Routing constraints: have access windows, opening hours, and GNSS-degraded zones (dense urban cores, underground, indoor) been built into the route plan, or just straight-line distance?
• QA layering: is there an independent audit sample — not just a back-office review of the same data the field team already submitted?
• Re-verification cadence: is there a funded plan for the second visit, or does the budget assume one pass is permanent?

Teampl runs field data collection programs — egocentric video, in-store audits, mystery visits, GIS surveys — designed around a specific research question, not a generic panel. If this topic touches a program you're planning, tell us what you're trying to learn.

Frequently Asked Questions