Field Survey Methodology for Geodata: Sampling, Routing, and QA

Field survey methodology for geodata is the documented set of decisions that determines which locations get visited, in what order field staff reach them, and how the collected data gets checked before it enters a production dataset. It sounds administrative. It isn't. Get the sampling frame wrong and you under-represent rural areas. Get the routing wrong and your field cost per record doubles. Get QA wrong and a client ships a dataset that looks complete and is quietly 8% inaccurate.

The market context explains why this keeps coming up. Independent estimates put the Europe geographic information system market at USD 3.0 billion in 2025 , with other analysts placing the 2024 base closer to USD 3.63 billion, projected to reach USD 8.44 billion by 2033 . The spread between estimates is itself a lesson: market-sizing methodology varies as much as field methodology does, and neither is settled science. What's not in dispute is the direction — more GIS spend, more automated pipelines, and still, underneath all of it, someone has to go look.

This piece lays out the three disciplines that make a field geodata program defensible: sampling design, route planning, and quality assurance. It leads with the EU, where address and cadastral infrastructure is unusually fragmented across France, Germany, Switzerland, Belgium, and the Netherlands, and flags where other markets diverge.

Why field verification still earns its budget

Satellite imagery, address-matching algorithms, and crowdsourced edits have all gotten better. None of them answer "is this pharmacy still open on Tuesdays" or "does this charger actually have a CCS connector." Those are attribute questions, not geometry questions, and attribute truth is exactly what remote sensing struggles with. Address and POI data decays continuously, which means a one-time scrape is already stale by the time it ships.

Three independent market analyses illustrate how differently this space gets measured — useful context before trusting any single figure in a vendor pitch.

Source	Base year value	Forecast	CAGR
IMARC Group	USD 3.0 Billion in 2025	through 2034	5.53%
Renub Research	USD 3.63 billion in 2024	USD 8.44 billion by 2033	9.82%
Market Data Forecast	USD 3.11 billion in 2025	USD 5.34 billion by 2034	6.20%

None of these figures agree exactly, because each report scopes "GIS market" differently — hardware versus software versus services, surveying versus telematics. Treat any single market number as directional, not precise. The underlying trend across all three is real growth, which means more datasets to verify, not fewer.

Sampling design: deciding what "enough ground truth" means

A sampling frame refers to the full list or map of locations eligible for selection — every address, every POI, every charging station — before you pick which ones a field team actually visits. Ground truth means a value confirmed by direct, on-site observation rather than inference. Most field programs can't visit every eligible location, so the sampling design decides where the budget goes.

Method	How it works	Best for	Key limitation
Simple random	Every location in the frame has equal selection probability	Small, homogeneous areas	Can leave entire regions unvisited by chance
Systematic / grid	points created in a nonrandom pattern across hexagonal, square, or triangular grids so the study area is evenly covered	Even geographic coverage, no known hotspots	Can alias with a regular real-world pattern (e.g. a road grid)
Stratified random	the study area is divided into strata, with points placed randomly inside each one so every stratum is represented	Known sub-groups (urban/suburban/rural, building density) that correlate with the target variable	Needs a reliable stratification variable up front
GRTS (spatially balanced)	spreads sample effort evenly over the target region with a stochastic rather than fixed-grid pattern	Long-running monitoring programs that add/drop sites over time	More complex to implement than a simple grid

Stratification is where most of the real gains sit. Research into stratified versus systematic designs for land-cover surveys found the stratified segment-based sampling design reduced the number of sample plots compared to the systematic design by 28% while holding precision roughly constant. For POI or address verification, the equivalent strata are usually geography type (dense urban, suburban, rural), building density, or known data-source reliability. Tools like the R package SamplingStrata take this further, using a genetic algorithm that explores possible stratifications and finds the one giving the global minimum sample size for a target precision — useful when you're justifying a sample size to a client who wants to know why it isn't larger.

Worked example: sampling and routing a nationwide infrastructure survey

Consider a mandate to verify a country's public EV charging network against the operator-published registry: station name, operator, connector types, charging capacity, and pricing. A naive approach samples major cities, where chargers are dense and easy to reach. That design systematically misses rural and highway-corridor stations, which is exactly where registry errors cluster because they get the least maintenance attention.

Teampl ran a field program on exactly this shape of problem, covering 1,500+ EV charging stations across the Netherlands, deliberately sampling urban, suburban, and rural areas rather than restricting coverage to major cities. Every record carried station name, operator, address, number of charging points, connector types, charging capacity, and pricing, backed by on-site photographic evidence for each location. The stratification variable there wasn't arbitrary — it was geography type, because registry drift correlates with how recently a location was last touched by anyone.

Routing: turning a sample into a feasible field plan

Once the sample is fixed, routing decides whether the plan is affordable. This is a vehicle routing problem (VRP): a generalization of the traveling salesman problem where multiple field agents, each with a daily capacity and start/end point, need to cover a set of stops at minimum total cost. Pure distance minimization is the wrong objective function for most geodata surveys, though. Business hours matter more.

A POI that's only verifiable as "open" between 9am and 6pm constrains the route more than the distance between stops does. Lone-worker safety policies add further constraints: no isolated rural visits after dark, mandatory check-in intervals, a maximum shift length before a mandatory break. Daily throughput caps — say, 40 to 60 verified locations per surveyor per day for a straightforward POI check, fewer for a multi-attribute infrastructure audit — set the real ceiling on what a sample size actually costs to execute.

Practical route planning for field geodata work typically layers three things: geographic clustering (group stops into compact zones before optimizing within each), constraint-based sequencing (open the VRP solver — commercial options like Esri's Network Analyst or open tools like Google's OR-Tools both handle this — with time windows and capacity limits as hard constraints), and a revisit queue for anything that fails on the first pass (locked gate, business temporarily closed, access denied). That last piece is routinely underbudgeted. A realistic field plan assumes 5-15% of first-pass visits will need a second attempt, depending on the location type.

QA: catching errors before they ship

ISO 19157-1:2023 is the current international standard for describing geographic data quality. It establishes principles for describing the quality of geographic data , and it works alongside ISO 19115 for metadata. Crucially, the standard does not attempt to define minimum acceptable levels of quality for geographic data — it gives you the vocabulary, not the pass/fail line. Your organization has to set that threshold itself, which is the part most field QA plans skip.

The standard's keyword set covers positional accuracy, completeness, logical consistency, thematic accuracy, temporal quality and usability as the core quality dimensions. Mapped onto a field verification program, they translate roughly like this:

Quality dimension	What it measures	Typical field QA check
Positional accuracy	Is the point where it says it is	GNSS capture tolerance, cross-check against known landmarks
Completeness	Are locations missing or duplicated	Reconcile field count against sampling frame count
Logical consistency	Do attribute combinations make sense	Rule-based validation (e.g. a "closed" status can't carry current opening hours)
Thematic accuracy	Is the classification correct	Spot-check against photo evidence or a second surveyor
Temporal quality	How old is the observation	Timestamp every record, flag anything past its refresh window
Usability	Fit for the client's specific purpose	Client-defined acceptance test, not a generic pass rate

In practice, the cheapest QA lever is photographic evidence tied to every record, because it turns a disputed attribute into something reviewable after the fact instead of a surveyor's word against the client's expectation. The same evidentiary logic that makes mystery-visit data defensible applies directly to POI and infrastructure audits. A second, independent re-survey of a 5-10% sub-sample is the standard way to estimate your own error rate rather than assume one.

EU address infrastructure: same question, different plumbing

The INSPIRE Directive, in force since May 2007 to support an infrastructure for spatial information across Community environmental policies , harmonizes how EU member states document and share spatial data across 34 spatial data themes , one of which is Addresses. It defines an address, for INSPIRE purposes, as the "location of properties based on address identifiers, usually by road name, house number, postal code" . What INSPIRE does not do is force every member state onto one database. The result is real divergence in how you'd actually go verify an address on the ground.

Country	Address infrastructure	Governance model	Notable fact
France	Base Adresse Nationale (BAN)	Single national open database, co-piloted by national agencies with municipalities as the authoritative source	26,104,003 records , published under an open licence
Germany	State-level cadastral registers	Each federal state's surveying authority maintains its own address and parcel data	No single national open address file; access and format vary by state
Switzerland	Cantonal surveying offices plus federal coordination	Cantons hold primary authority; the federal topographic agency coordinates standards	Verification programs typically need canton-by-canton data-sharing agreements
Belgium	Regional registers (Flanders, Wallonia, Brussels)	Each region runs its own address authority rather than one national body	A nationwide field program effectively means three separate data regimes
Netherlands	Basisregistratie Adressen en Gebouwen (BAG)	Single mandatory national base registry	Closer to France's model than to Germany's or Belgium's

The practical takeaway: a field program budgeted on France's single-source model will underestimate the coordination cost of running the same survey in Germany or Belgium. That cost shows up as extra lead time negotiating data access, not as extra field hours — easy to miss when scoping a multi-country project.

What this looks like outside the EU

The United States has no federal equivalent to the BAN or BAG. Address authority sits with county and municipal governments, and most usable national coverage comes from private aggregators (postal, telecom, or commercial data vendors) rather than a single public registry. That pushes more of the sampling-frame work onto whoever runs the survey — there's no ready-made national list to stratify against, so field teams often have to build or reconcile their own frame from multiple commercial sources before sampling can even start. The UK sits somewhere in between, with Ordnance Survey holding a comprehensive national address base but licensing it commercially rather than as fully open data. None of this changes the sampling, routing, or QA mechanics described above — it just changes how expensive and how reliable your starting sampling frame is before a single surveyor leaves the office.

What to check before you commission a field survey

• Sampling frame source: confirm whether it's a single authoritative register or an assembled composite, and how recently it was refreshed.
• Stratification variable: pick one that actually correlates with your error risk (geography type, data age, source reliability), not an arbitrary administrative boundary.
• Routing constraints: business hours, lone-worker rules, and a realistic revisit allowance — budget 5-15% of stops needing a second attempt.
• QA threshold: decide your acceptable error rate per quality dimension before fieldwork starts, since ISO 19157 won't set it for you.
• Resurvey cadence: set a refresh interval matched to how fast the attribute in question actually changes, not a fixed annual default.

Frequently Asked Questions