Inside the CEOS Land Cover Validation Guidelines: how scientists worldwide just agreed on a common way to check a satellite land cover map’s accuracy
Why a "good-looking" map can still be wrong
Satellite images now let us map forests, farmland, cities, and wetlands almost anywhere on the Earth. Governments use these maps to track deforestation, plan disaster response, and report progress on climate goals. But a map that looks colorful and detailed is not necessarily accurate — and the only way to know how much to trust it is to test it against real, independently collected ground evidence. That process is called accuracy assessment.
"A map that is lacking an accuracy assessment is just a prototype or an untested hypothesis and should not be used as a source of information."
— Land Cover and Change Map Accuracy Assessment and Area Estimation Good Practices Protocol, Tyukavina et al. (2025)
In other words: without this check, even a beautiful, high-resolution map is a guess. Analysis of the studies going back to the 1990s shows that the share of land-cover research papers that also discuss accuracy has climbed from about 40% to roughly 50% over the past two decades — yet many published maps still skip the check, describe their methods too vaguely to be verified, or never share the reference data behind their numbers.
A global science community sets a common standard
To close that gap, the Committee on Earth Observation Satellites (CEOS) — the international body that coordinates the world’s Earth-observing satellite agencies — has just finished updating its flagship guidance on how to properly check the accuracy of land cover maps. The update was conducted by the Land Product Validation (LPV) Subgroup, part of CEOS’s Working Group on Calibration and Validation (WGCV), shown below.
Figure 1. How the update fits into CEOS’s structure
The Land Product Validation Subgroup, which focuses on land cover among other themes, sits within CEOS’s Working Group on Calibration & Validation.
The revision brings the field’s foundational guidelines — last written in 2006, before free access to Landsat satellite imagery and before the Sentinel program existed — up to date with two decades of new science: practical instructions for estimating how much area a given land type actually covers (not just how accurate the map is), new sources of reference data such as drones and high-resolution imagery, and examples of real-life applications so non-specialists can follow the logic step by step.
This was very much a team effort, led by Alexandra (Sasha) Tyukavina (University of Maryland) and Sophie Bontemps (UC Louvain, Belgium), who co-lead the CEOS land cover validation effort. They worked alongside Nandika Tsendbazar (Wageningen University, the Netherlands), who is the current CEOS LPV co-lead, Stephen Stehman (SUNY ESF) and Giles Foody (University of Nottingham) — two of the most cited researchers in statistical map-validation methods, with a combined 30-plus years of publications on the topic — and Anna Komarova (University of Maryland) and Jaime Nickeson (NASA Goddard Space Flight Center), who focused on making the document clear for a broad audience. In total, 35 land cover mapping and validation experts from 18 institutions across nine countries contributed.
3
years in the making
35
contributing experts
18
institutions
9
countries
How the update came together
The work began with an open call for input in October 2022, followed by nearly two years of drafting and building consensus on best practices. A draft was shared publicly for community review in August 2024, drawing detailed and positive feedback from 12 outside reviewers across 11 organizations in seven countries. The finished document (Version 1.0) was published in September 2025, and a lightly revised Version 1.1 — now formally endorsed by CEOS’s calibration and validation working group — followed in November 2025.
What practitioners will actually find inside
The new guide is organized as a practical handbook rather than an academic text. It walks through:
- How to design a sound sampling plan — in plain terms, how to choose a statistically valid set of locations to check on the ground or with high-resolution satellite imagery
- How to collect trustworthy reference data — from satellite time-series and lidar to drone surveys and old-fashioned fieldwork – including establishing the rules for labeling a reference sample, ensuring the quality of reference data and accounting for potential reference data errors
- How to calculate accuracy and area estimates correctly, with appropriate formulas for specific sampling designs and common pitfalls flagged
- Real case studies from existing global and regional mapping efforts, so readers can see the methods applied in practice
To keep the material approachable for non-statisticians, the guide relies heavily on diagrams, real-life examples from published studies, and cross-references that let readers jump straight to the section, table, or figure being discussed.
Why this matters beyond the lab
This may sound like a technical, behind-the-scenes fix, but its effects reach far beyond research papers. Maps built and checked using these standards feed directly into decisions that affect real places and real people: tracking whether countries are meeting forest and climate commitments, guiding disaster response after floods or wildfires, informing land-use and agricultural policy, and underpinning biodiversity and conservation planning. The guide even recommends that at least 30% of any mapping project’s budget be set aside for accuracy assessment — a reminder, aimed at funders and program managers, that evaluating a map properly is not an afterthought but a core part of making it useful.
Endorsed — and already being used
Figure 2. Examples of illustrative diagrams from the guidelines
Top: methods for estimating subsample proportions of land cover class. Bottom: constructing a stratification for sampling – a real-life example from mainland Southeast Asia.
At SIT-41, held in Irvine, California on 14–16 April 2026, the CEOS Strategic Implementation Team formally endorsed Version 1.1 of the Land Cover Validation Protocol, as requested and without amendments. The endorsement decision reaffirmed the guide’s central message: that map accuracy assessments should be prioritised, planned, and properly resourced as a core part of any mapping project, not an optional extra.
The meeting discussion pointed to signs that the guidelines are already gaining traction. The Italian Space Agency (ASI) said it intends to adopt them for its own national land cover mapping work — an early, concrete example of the kind of uptake by national space agencies that the authors had hoped for. Further discussion on whether the protocol is ready for the age of artificial intelligence pointed to the challenges of keeping the reference datasets independent from AI-based map training, to ensure independent accuracy evaluation of the newly created maps. According to the lead author Alexandra Tyukavina, the core accuracy assessment and area estimation methods described in the guidelines will remain valid regardless of the way the new maps are produced.
CEOS agencies are now being encouraged to promote the protocol as the standard for newly published land cover maps and to help publicise it through official CEOS and member-agency channels, so its benefits reach mapping teams worldwide.
Alexandra (Sasha) Tyukavina, University of Maryland, USA
On behalf of the CEOS WGCV · Land Product Validation Editorial Team
(LPV Land Cover Focus Area Co-leads, Land cover mapping experts)
