Geospatial Data ANNOTATION AND Validation.
A Buyer’s Guide to geospacial data annotation and validation services.
That foundation is largely invisible until it fails, at which point a misclassified transmission tower or a building footprint offset by two metres propagates into every downstream decision the model makes.
This guide is written for the person who has to buy that work. It explains what geospatial data validation and annotation actually consists of, sets out the criteria that separate a capable provider from a plausible one, and profiles the vendors that genuinely compete in this space.
It also states plainly where the evidence for each provider comes from, and distinguishes throughout between what a company claims about itself and what can be independently verified. Where we could not verify something, we say so.
What is geospatial data annotation and validation?
Geospatial data annotation is the process of adding structured, machine-readable labels to data that has a position on the earth. Geospatial data validation is the separate and equally important process of verifying that those labels, and the geometry they sit on, are correct, consistent and fit for the purpose the data was commissioned for.
Annotation creates the training signal; validation is what makes it trustworthy. Buyers routinely purchase the first and assume they are getting the second.
The source data
The raw material arrives in several distinct forms, and each imposes different demands on the workforce handling it. Satellite imagery offers wide coverage at resolutions ranging from sub-metre commercial products to tens of metres for public missions, and typically includes multispectral bands that carry information invisible in a photograph.
Aerial and drone imagery provides much higher resolution over smaller areas, which makes it the standard input for utility inspection and construction monitoring. LiDAR point clouds record millions of three-dimensional returns per scene and are the basis of terrain models, vegetation encroachment analysis and autonomous navigation.
Street-level and vehicle sensor data underpins high-definition mapping. Existing GIS and map databases need continuous verification rather than initial labelling, because the world changes faster than the map does.
The annotation tasks
The labelling itself takes a limited number of geometric forms, deployed according to what the model needs to learn.
Task | What it produces | Typical application |
|---|---|---|
Semantic segmentation | Every pixel assigned to a class such as road, roof, water or vegetation | Land use and land cover mapping, impervious surface analysis |
Instance segmentation | Each individual object separated from others of the same class | Counting buildings, vehicles, solar panels or trees |
Polygon and building footprint extraction | Precise vector outlines of features | Cadastral mapping, property intelligence, insurance risk |
Object detection and bounding boxes | Located and classified objects | Asset inventories, vehicle and vessel detection |
3D point cloud classification | Each LiDAR return labelled as ground, vegetation, structure, wire | Terrain models, vegetation management, digital twins |
Keypoint and landmark annotation | Discrete positioned points | Utility pole and insulator marking, intersection modelling |
Point of interest and address verification | Confirmed or corrected attribute data against ground truth | Navigation, logistics, local search |
Change detection | Differences between two epochs of imagery | Construction monitoring, deforestation, disaster damage assessment |
Semantic segmentation
Instance segmentation
Polygon and building footprint extraction
Object detection and bounding boxes
3D point cloud classification
Keypoint and landmark annotation
Point of interest and address verification
Change detection
Why the geospacial data validation half matters more than buyers expect
Two standards govern this field, and the most important thing about both of them is what they refuse to do.
ISO 19157-1:2023, “Geographic information Data quality Part 1: General requirements”, published in April 2023 by ISO technical committee TC 211, establishes the principles for describing the quality of geographic data, defines the components for doing so, and sets out procedures for evaluation and reporting. It replaces ISO 19157:2013.
It also contains this sentence, which every buyer should read twice:
The ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2, Version 2, published on 24 June 2024 by the American Society for Photogrammetry and Remote Sensing, takes the same position from the other direction.
It is deliberately sensor-agnostic and data-driven, offering what its authors describe as unlimited accuracy levels without hardware limitations, and it uses root mean square error as its single accuracy measure.
Edition 1 had expressed accuracy through both RMSE and a 95 per cent confidence level; the confidence level was removed because it offered no additional benefit while causing considerable confusion. Crucially, the accuracy class is an input from the buyer, not an output from the standard: once a user specifies that a project requires, for example, five centimetres, that figure becomes the accuracy class.
The practical consequence is that neither standard will tell you whether a deliverable is good enough. You have to specify that, and then insist on being able to test it. This single fact should reorganise how you evaluate vendors: the question is not who claims the highest accuracy, but who will accept a written, testable specification and be measured against it.
What to look for in a geospatial annotation Provider ?
The criteria below are ordered roughly by how much damage getting them wrong will do.
A quality specification you write and they accept
Ask the provider to commit to an accuracy class expressed as RMSE under the ASPRS standard, and to report against section 7.16 of that standard, which prescribes the formal reporting statements. Ask how many independent checkpoints they will use.
The standard requires a minimum of thirty for horizontal accuracy and non-vegetated vertical accuracy on projects up to 1,000 square kilometres, rising with project size, and it requires that checkpoints be independent of the control points used in calibration and be at least twice as accurate as the expected product accuracy.
A provider who has never heard of any of this is not necessarily incapable, but they are telling you that quality on their projects has so far been a matter of opinion.
Understanding of vertical accuracy and vegetation
If your work involves elevation data, note that the ASPRS standard separates non-vegetated vertical accuracy from vegetated vertical accuracy, and that only the former is used to accept or reject data. Vegetated vertical accuracy has no threshold and is reported as found unless you agree otherwise in advance.
For anyone doing vegetation encroachment work on power lines or forestry analysis, that is precisely the number you care about, and it is precisely the number the standard leaves to negotiation. Raise it before contracting, not afterwards.
Segmentation metrics, and which one they quote
For segmentation work, providers commonly quote Intersection over Union, also known as the Jaccard index, or the F1 score, equivalent to the Dice coefficient.
Both measure overlap between prediction and ground truth, but IoU penalises under-segmentation and over-segmentation more heavily, which makes it the stricter test of boundary fidelity.
A vendor quoting F1 where you expected IoU is not lying, but the numbers are not comparable, and boundary quality is exactly what matters when you are extracting building footprints or parcel outlines.
Coordinate reference system discipline
A coordinate reference system or EPSG code mismatch between datasets causes spatial misalignment, distorted distances and areas, and analysis that is confidently wrong rather than obviously broken. It is a mundane failure mode and a frequent one.
Ask which coordinate reference system deliverables will be returned in, who is responsible for reprojection, and how mismatches are detected before delivery rather than after.
Domain knowledge in your specific vertical
Generic annotation skill does not transfer automatically to geospatial work, and geospatial skill does not transfer automatically between verticals.
An annotator who is excellent at labelling vehicles in street imagery may not be able to distinguish a distribution transformer from a capacitor bank, or recognise the difference between a substation and an industrial site from overhead.
Ask for work examples in your domain, and ask who trains the annotators on the domain rather than on the tool.
Security, certification and data residency
High-resolution imagery of critical infrastructure is sensitive by nature, and in some jurisdictions it is restricted.
Establish which certifications the provider actually holds rather than mentions: ISO 27001 for information security management, SOC 2 for service organisation controls, and documented GDPR compliance where European personal data is implicated.
Ask where the work is physically performed, whether annotators work from secured facilities or from home, and whether the workforce is employed or crowdsourced. For defence and government work, ask specifically about cleared personnel and accredited networks.
Workforce model and attrition
Attrition is the hidden cost in annotation. Geospatial labelling has a genuine learning curve, and a provider that loses trained annotators every few months is repeatedly rebuilding the domain knowledge your project depends on.
Ask for a trailing twelve-month attrition figure, ask whether annotators are employees or contractors, and treat a refusal to answer as an answer.
Pilot terms, minimum commitment and exit
Most providers in this market do not publish pricing, and most will want to run a pilot on your data before quoting, which is reasonable given how much task complexity varies.
What matters is the terms: whether the pilot is free or chargeable, what volume triggers a commitment, and what the minimum engagement is.
Minimums can be substantial and are not always prominent. Mindy Support, for example, publishes a minimum threshold of 735 productive man-hours per month, equivalent to roughly five annotators, to start a project.
That is a perfectly legitimate business model and a serious constraint if you needed a two-week proof of concept.
Vendor continuity
This market consolidates and companies disappear. Orbital Insight, once among the better-known names in geospatial analytics, was acquired by Privateer in 2024 and its website now redirects; it no longer operates independently.
Appen, a long-established annotation provider with genuine geospatial and sensor-fusion capability, lost a major contract with Google and has since reported significant revenue decline.
Neither fact makes a provider unusable, but both belong in a due diligence file, and both argue for retaining the ability to move your work.
Whether you should be buying annotation at all
One honest question before you procure: do you need labelled data, or do you need the finished map layer? Some companies sell the output rather than the labour.
Ecopia AI, based in Toronto and founded in 2013, converts high-resolution imagery into finished high-precision vector map data such as building footprints, roads and land cover, and lists clients including the World Bank, Airbus, NOAA and the Government of Canada.
If you want a building footprint layer rather than a training set, buying the layer may be faster and cheaper than commissioning annotation to produce it.
TOP 5 geospatial data annotation providers
The five below were selected on evidence of genuine, documented geospatial capability rather than general annotation reputation. Each was verified against the company’s own published material and, where available, third-party review platforms.
Claims that could not be independently confirmed are identified as the company’s own.
iMerit
Founded in 2012 and headquartered in San Jose, California, iMerit operates a dedicated geospatial technology practice covering satellite, aerial and drone imagery, with polygon annotation, instance and semantic segmentation, classification, object tracking, LiDAR annotation, point of interest marking and point annotation.
It holds SOC 2 Type 2, ISO 27001, ISO 9001:2015, HIPAA, GDPR and TISAX certifications, and states a workforce of more than 10,000 across over 60 countries. Pricing is not published; the company offers custom pricing and pilot projects. It carries a 4.8 rating from 12 verified reviews on G2.
Why it leads: iMerit publishes the strongest named geospatial credential we found anywhere in this market asset inspection work for Enel Group across a stated 1.5 million miles of electrical distribution network. A named client at that scale in a demanding vertical is worth more than any quantity of capability claims.
Caveat: its other flagship geospatial reference, high-definition map production for a leading robotaxi company, is unnamed and therefore unverifiable.
TELUS Digital
Formerly TELUS International, and before that the acquirer of Lionbridge AI, TELUS Digital is headquartered in Vancouver, was founded in 2005, and reports more than 75,000 team members globally. Its Ground Truth Studio platform supports 3D sensor fusion including object classification, 3D object tracking, 2D-to-3D linking, bird’s-eye-view and point cloud segmentation, alongside semantic segmentation, polygon and LiDAR annotation, landmark and keypoint annotation, point of interest tagging, address verification and routing.
It holds ISO 27001, SOC 2, GDPR and HIPAA, rates 4.9 from 35 reviews on Clutch, and was named a Leader in the Everest Group Data Annotation and Labeling PEAK Matrix Assessment 2024.
Why it leads: independent analyst recognition plus the deepest bench in the group. If your programme needs to scale to thousands of annotators across time zones without subcontracting, the list of providers who can do that is short.
Caveat: enterprise-oriented, with a minimum project size listed on Clutch at 50,000 US dollars and no published pricing. We found no named geospatial client references.
Scale AI
Founded in 2016 and headquartered in San Francisco, Scale AI is distinguished less by breadth of annotation services than by its position in government and defence work, including operation on classified networks and a dedicated defence platform.
Its self-service data engine publishes pay-as-you-go pricing from 0.05 US dollars per unit after a free tier, though managed annotation rates are not disclosed.
Why it leads: for overhead imagery work with a national security dimension, accreditation to operate on classified infrastructure is a qualifying requirement that most competitors simply cannot meet.
Caveat: enterprise sales-driven with limited pricing transparency, which is a real barrier for smaller teams needing predictable costs. We could not locate Clutch or G2 ratings.
Keymakr
Founded in 2015, Keymakr states that it creates complex geospatial models from satellite imagery and GIS or GPS data, covering satellite, aerial and LiDAR sources with semantic segmentation, object detection, polygon annotation, 3D point cloud annotation, bounding boxes, cuboids and skeletal tracking, applied to agriculture, urban planning, disaster response and autonomous drone navigation.
It works through its own Keylabs platform, originally built for internal use, and uses an in-house workforce under non-disclosure agreements rather than crowdsourcing.
Named clients include SeeChange, Evinced, Cognex, BlueWhite Robotics, SoundAware and Agwa. It carries a 4.8 rating from 46 reviews on G2, the highest review count of the five, and offers an hourly pay-as-you-go entry tier with custom pricing above it after a pilot on the client’s data.
Why it leads: the strongest review evidence in the group combined with named clients and an in-house workforce model.
Caveat: its published workforce figure of more than 400 specialists dates from January 2019 and may be stale. Its certification claims for ISO 27001, SOC 2 Type II, HIPAA and GDPR appear in blog and directory material rather than on a formal trust page, so ask for the certificates. Its own site lists a New York address while third-party sources indicate Israeli operations; clarify where your data would actually be processed.
Cogito Tech
Cogito Tech offers satellite, aerial, GIS and LiDAR annotation as part of a broad data-labelling portfolio and competes directly for geospatial work.
We verified that it is a genuine competitor in this space, but our verification of its specific geospatial credentials was less complete than for the four providers above, and we would recommend requesting domain-specific case studies and current certification evidence before shortlisting.
Also worth evaluating: Label Your Data and Sama
Two further providers deserve mention for specific reasons. Label Your Data maintains a dedicated geospatial practice covering LiDAR segmentation, building footprint extraction, road network mapping, land use classification, crop and soil analysis and flood impact assessment, states a workforce of around 1,000 across 22 countries, claims ISO 27001, SOC 2, HIPAA, PCI DSS Level 1, GDPR and CCPA, and rates 5.0 from 27 reviews on Clutch.
Unusually in this market it publishes indicative pricing from 0.015 US dollars per object for keypoint annotation and 6 US dollars per annotator hour and offers a free pilot, although geospatial work specifically requires a custom quote.
Sama combines enterprise-grade 3D and sensor data annotation with a Certified B Corporation impact sourcing model built on a full-time in-house workforce, which makes it the closest philosophical comparator to providers that employ rather than crowdsource, at the cost of some flexibility on small ad-hoc projects.
Provider | HQ / Founded | Certifications (as stated) | Review evidence | Distinctive strength |
|---|---|---|---|---|
iMerit | San Jose, USA / 2012 | SOC 2 Type 2, ISO 27001, ISO 9001, HIPAA, GDPR, TISAX | G2 4.8 (12) | Named utility client at continental scale |
TELUS Digital | Vancouver, Canada / 2005 | ISO 27001, SOC 2, GDPR, HIPAA | Clutch 4.9 (35) | Everest Group PEAK Matrix Leader 2024; scale |
Scale AI | San Francisco, USA / 2016 | Classified network accreditation (CUI, SIPR, TS) | Not found | Government and defence geospatial work |
Keymakr | New York, USA (see caveat) / 2015 | ISO 27001, SOC 2 Type II, HIPAA, GDPR (claimed) | G2 4.8 (46) | Proprietary Keylabs platform; in-house NDA workforce |
Cogito Tech | India / USA | Request evidence | Not verified | Broad annotation portfolio including geospatial |
Label Your Data | Wilmington USA / Nicosia CY | ISO 27001, SOC 2, HIPAA, PCI DSS L1, GDPR, CCPA (claimed) | Clutch 5.0 (27) | Published pricing and a free pilot |
Sama | San Francisco, USA | Certified B Corporation | Not verified | Impact sourcing with full-time in-house staff |
iMerit
TELUS Digital
Scale AI
Keymakr
Cogito Tech
Label Your Data
Sama
Another Option: OWorkers
We publish this guide, so treat what follows with the scepticism you would apply to any vendor writing about itself.
What we do claim is delivery experience in many demanding geospatial verticals.
We have completed to name of few :
- Electrical grid annotation work, the category of task that involves identifying and classifying transmission and distribution assets from aerial and satellite imagery.
- Smart city data annotation for Dubai port.
- Data annotation of Forest trees and agriculture fields.
- Data annotation of supermarket shelves from droides.
ALL are genuinely difficult: grid annotation requires annotators who can tell one piece of hardware from another at low resolution and in poor light, and smart city work requires three-dimensional consistency across large areas. The ASPRS standard added a three-dimensional positional accuracy class in its 2024 edition explicitly because of digital twin and smart city applications.
We work in that environment.
What we can bring
We operate 5 delivery centres in Bulgaria, Egypt, Madagascar, the Philippines and Kenya, holding ISO 27001 / 27701 and ISO 9001:2015, with GDPR compliance across centres.
We hire a full time DPO, Data Protection Officer based in Europe, bring compliant DPA (Data Processing Agreement) , SCC (Standard Contractual Clauses) and DPIA (Data Protection Impact Assessment) support available on request.
On the workforce criterion set out earlier in this guide, our position is straightforward.
Every annotator is an employee rather than a freelancer, we pay social taxes, and we are rated above 4.6 out of 5 on Glassdoor.
Our employee turnover was 1.7 per cent in 2025 against a BPO industry average we understand to be around 20 per cent on an annual basis.
It matters here for the reason given above: geospatial annotation has a learning curve, and retained annotators keep the domain knowledge your project paid to build.
Geospatial work sits alongside our wider data annotation services, image annotation, multilingual and document processing practices, which means attribute verification, address validation and document-based ground truth can be handled by the same provider as the imagery labelling.
Frequently Asked Questions
Annotation is the creation of labels on positioned data, producing the training signal a model learns from.
Validation is the separate verification that those labels and their underlying geometry are correct, consistent and fit for the intended purpose.
Buyers frequently purchase annotation and assume validation is included; it should be specified and priced explicitly.
ISO 19157-1:2023 establishes the principles for describing and reporting the quality of geographic data and replaces ISO 19157:2013.
The ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2, Version 2, published in June 2024, defines positional accuracy using root mean square error and prescribes assessment and reporting procedures.
Notably, neither standard defines what counts as acceptable quality: that must be specified in the data product specification by the buyer.
Positional accuracy is measured as root mean square error against independent checkpoints that are at least twice as accurate as the product being tested, with a minimum of thirty checkpoints under the ASPRS standard for projects up to 1,000 square kilometres.
Label quality on segmentation tasks is usually measured with Intersection over Union or the F1 score, of which IoU is the stricter test of boundary fidelity.
Most providers in this market do not publish pricing and will run a pilot on your data before quoting, because task complexity varies enormously between, say, bounding boxes on vehicles and point cloud classification of vegetation against conductors.
Label Your Data is an exception, publishing indicative rates from 0.015 US dollars per object and 6 US dollars per annotator hour, with geospatial work quoted separately.
Ask about minimum commitments early: they can be significant.
It depends on the task. Bounding boxes on aerial vehicles are within reach of any competent annotation workforce.
Distinguishing utility hardware classes, classifying LiDAR returns, or maintaining three-dimensional consistency across a city model require domain training and quality processes that not every vendor has.
Ask for work examples in your specific vertical rather than in geospatial generally.
A question worth asking before procurement. Companies such as Ecopia AI sell finished vector map data building footprints, road networks, land cover rather than annotation labour.
If you need the layer rather than a training set, buying the output can be faster and cheaper than commissioning the process.
Which certifications they hold rather than mention, with certificates supplied on request; where the work is physically performed; whether annotators work in secured facilities or remotely; whether the workforce is employed or crowdsourced; and how imagery of critical infrastructure is handled.
For government and defence work, ask about cleared personnel and accredited networks specifically.
Discuss a Geospatial data Annotation Project
If you are scoping geospatial annotation or validation work, the productive first conversation is about specification: the source data, the classes, the accuracy class you need expressed as RMSE, the validation regime and the volumes.
Tell us those and we will tell you honestly whether we are the right provider for it, including when we are not.
Request a quote or contact our team. Lines are open 24 hours a day, seven days a week: United States +1 (833) 246 7264, United Kingdom +44 (800) 048 5681.
Read also : Outsourcing in bulgaria | Egypt Outsourcing | Madagascar BPO
[1] International Organization for Standardization, “ISO 19157-1:2023 Geographic information — Data quality — Part 1: General requirements”, Edition 1, published April 2023, ISO/TC 211. https://www.iso.org/standard/78900.html
[2] Qassim A. Abdullah, “Overview of the ASPRS Positional Accuracy Standards for Digital Geospatial Data, Edition 2, Version 2 (2024)”, LIDAR Magazine, 30 June 2025. https://lidarmag.com/2025/06/30/overview-of-the-asprs-positional-accuracy-standards-for-digital-geospatial-data/ — Note: the ASPRS standard itself is the primary source; asprs.org returned no extractable content at the time of writing, so this expert overview is cited in its place.
[3] Mindy Support, published minimum project threshold of 735 productive man-hours per month. https://mindy-support.com/
[4] Orbital Insight, acquired by Privateer in May 2024; orbitalinsight.com now redirects to Privateer.
[5] Appen, reported loss of a major Google contract and subsequent revenue decline. https://www.appen.com/
[6] Ecopia AI, company and product information. https://www.ecopiatech.com/
[7] iMerit, geospatial technology practice. https://imerit.ai/domains/geospatial-technology/ — certifications, workforce figures and the Enel Group reference are as published by iMerit; G2 rating verified on g2.com.
[8] TELUS Digital, data annotation services and Ground Truth Studio. https://www.telusdigital.com/solutions/data-for-ai-training/data-annotation-services — Everest Group PEAK Matrix 2024 Leader designation and Clutch rating verified independently.
[9] Scale AI, platform, pricing and government capability. https://scale.com/
[10] Keymakr, geospatial annotation capability and Keylabs platform. https://keymakr.com/ — G2 rating verified on g2.com; certification claims sourced from company blog and directory material rather than a formal trust page.
[11] Cogito Tech, data annotation services including geospatial. https://www.cogitotech.com/
[12] Label Your Data, geospatial data annotation and published pricing. https://labelyourdata.com/industries/geospatial-data-annotation and https://labelyourdata.com/pricing — Clutch rating verified on clutch.co.
[13] Sama, enterprise annotation and impact sourcing model. https://www.sama.com/
