Sound energy investment plans and decisions depend on good data. An old grid layer, for example, can make an already-connected settlement appear unelectrified. A location data error can place a health facility outside its service catchment, making an underserved community appear to have access to health care. These errors affect where planners direct investment and which communities get prioritized.This matters when 655 million people don’t have electricity access and 2 billion people still cook using polluting fuels. Achieving universal energy access by 2030 requires roughly $45 billion in annual investment, and smart, targeted investment is needed to deploy that funding effectively. To close this energy access gap, governments and philanthropies depend on energy planning or modeling tools to help decide where capital goes.Energy Access Explorer (EAE) is a geospatial planning tool and the first Digital Public Good supporting climate-compatible energy transitions. It overlays population data, grid infrastructure, energy resource potential and socioeconomic indicators to identify where interventions are needed most. Analysis from EAE can be used to prioritize interventions such as grid expansion or distributed renewable energy and to show where investment is needed for those interventions. To build an investment case based on this analysis, governments and investors need assurance: How good is the data underneath? This is why EAE builds data quality checks into the data pipeline itself rather than inspecting for it at the end.Problems with poor data can have real-world consequences. For example, researchers modeled travel time to every public health facility in sub-Saharan Africa using six widely used population datasets while holding facilities, roads and travel assumptions constant. In the Republic of the Congo, the modeled share of people living within 30 minutes of a facility ranged from 28.8% to 88.9%, depending on the dataset. A gap that wide is enough to determine whether a country meets or misses an international accessibility target, and at least 13 countries met the target under one population dataset and missed it under another. Published methodologies and peer-reviewed models are not enough if the underlying data is outdated, fragmented, inconsistently processed or spatially misaligned. EAE’s visible analysis rests on data layers, which rest on a foundation of standards and quality control. Note: EAE’s Data QA/QC uses more standards which can be found here ; ISO 19157 establishes principles of geographic data quality; FAIR principles are for metadata and QA4EO are for earth observation products. Catching Problems Early, Not at the EndQuality cannot be an afterthought applied once an analysis is complete; it has to be engineered into the pipeline that produces the analysis. The output of a flawed analysis looks much like the output of a sound one, which is why quality control has to sit early in the pipeline. A facility recorded in degrees, minutes and seconds, and the same record read as though the minutes were a decimal fraction. The two readings sit about 25 km apart, well outside any service catchment. Illustrative example. As a Digital Public Good, EAE’s value grows as partners create localized versions of the tool for their geographies. However, with these decentralized versions, data quality is harder to control. The proliferation of versions quickly outgrows what any single reviewer can track. No one can catch every layer that might age, drift or arrive in the wrong projection. Users can’t judge data quality by looking: The map renders cleanly even if the data degrades. This hidden risk is why shared standards are essential; they ensure quality remains high even when EAE’s data integrity lead can’t monitor the data directly. Data moves through five stages, from collection to archiving, and then re-enters review. This is an iterative process. For this reason, EAE puts its data standards, processing tools and review checklists at the center of the work. A partner adopting EAE inherits the same acceptance criteria, pre-processing steps and review routine that every other instance follows, so a dataset prepared in one geography is trustworthy for the same reasons as one prepared anywhere else. EAE structures that pipeline (Figure 3) as a data lifecycle, sustaining quality at every stage rather than at a single gate.From Fragmented Data to Standardized AcceptanceThe raw material of energy planning is uneven. Data arrives inconsistent, outdated and fragmented, and geospatial layers rarely share a resolution, a coordinate system or a collection date. EAE addresses this by publishing clear acceptance standards for all incoming data. It evaluates every dataset against explicit criteria: positional accuracy, attribute accuracy, completeness, temporality. These criteria are mapped to internationally recognized standards on data quality, metadata and quality assurance for Earth observation. Rather than leaving ‘recent enough’ to individual judgment, EAE accepts data no more than five years old across most categories, with documented exceptions, such as census data, where no more recent alternative exists.This is because energy access conditions change quickly. A grid extension, a new mini-grid or uptake of solar home systems, for example, can redraw the map of who is served within a single year. A decade-old layer loses precision and can direct investment toward a community that has since been connected, or away from one that has since grown. Geography administrators, in coordination with their EAE working groups, are encouraged to track and refresh their datasets as often as the source allows, well before the five-year threshold is reached.Because EAE belongs to a wider Digital Public Good and open-source ecosystem, its dataset must be highly interoperable. Even an individually accurate dataset carries a cost in time and formatting effort if it cannot align, join or transfer seamlessly across the broader ecosystem. EAE therefore aligns with the FAIR principles, under which data should be Findable, Accessible, Interoperable and Reusable by both people and machines. Traceability to recognized standards, coordinate reference system and FAIR-compliant metadata allow EAE’s analysis to move into a national planning process or a partner’s model without losing the context a user needs to interpret it.Data EAE acceptsData EAE flagsCurrent: within five years, or with a documented reason for an older vintageStale: older than the threshold with no justificationDocumented provenance and data licenseUnknown or unverifiable sourceDeclared, consistent coordinate reference system and unitsMixed or undeclared projections and unitsComplete coverage of the study geographyGaps or records silently truncated in processingClipped to analytical and administrative boundariesUnclipped, producing edge artifacts in overlaysTerms and categories that carry the same meaning across geographiesIndicators whose definitions differ from what users assumeModeled layers labeled as modeled, with cautions detailing uncertaintyModeled or proxy layers presented as though observedFAIR metadata (e.g., description of the data, why it’s used, source, license), with cautions and attributionNo metadata, no attribution, not reproducibleTraceable to a recognized standard (ISO 19157, QA4EO)No traceable basis for quality claimsStandards define what EAE accepts. For example, resampling1 process (Figure 4) shows why data processing decisions must also be verified and thoroughly documented in the metadata. Moving a layer from a fine resolution to a coarser one requires choosing an aggregation rule. For categorical data, such as land cover, the usual choice is majority rule, which assigns each output cell the most common class among the cells it replaces. The rule is sound, and it discards information every time it runs. Features that occupy a minority of any output cell disappear, and the resulting map carries no trace of what it lost. Resolution and minority-class loss How aggregating to coarser cells removes small features A land cover grid containing five settlements and a river. Increasing the cell size aggregates the grid by majority rule. Small settlements and the river disappear as the cell size grows, because they occupy a minority of every output cell that covers them. Cell size 100 m Settlements visible 5 of 5 River Retained Source data, 100 m After aggregation Agricultural land Built-up Water Every settlement is distinct at source resolution. Illustrative grid. Aggregation is by majority rule: each output cell takes the most common class among the cells it replaces. This figure needs JavaScript. A static version showing 100 m, 400 m and 1,000 m cells is published alongside the article. The same land cover data aggregated by majority rule to progressively coarser cells, over a 6 km by 6 km extent. Small settlements fall below the share of any output cell needed to array their class, and the river follows. EAE’s aggregation checklist therefore requires an explicit check for minority-class loss; a mass-balance test for additive variables; and a record of the original resolution, the target resolution and the method applied, with the residual uncertainty carried into the layer’s cautions. This is the difference between a transformation a colleague can audit and one that has to be taken on trust.EAE’s Extract, Transform, Load (ETL) Tool: PaverStandards decide what counts as acceptable; software makes applying them repeatable. Before integration, a dataset needs pre-processing into a specific form: reprojection to the common coordinate reference system, clipping to boundaries and, for distance-type layers, conversion into Euclidean distance surfaces measured in kilometers. Doing this manually for each geography would reintroduce the inconsistencies the standards exist to remove. Paver turns fragmented inputs with mixed projections, resolutions and units into standardized, comparable layers. The tool enforces consistency rather than each analyst applying it separately. Paver, our own open-source extract, transform, load (ETL) tool developed within EAE, performs this process consistently across geographies and dataset types. It serves two purposes: It makes cross-geography comparison reliable, because a layer prepared in one country passes through the same steps as its equivalent in another; and it lowers the barrier to entry, letting an administrator with basic GIS knowledge integrate a dataset without rebuilding the processing chain.Sustaining Quality After IntegrationA monitoring and deprecation step (Figure 3) flags when data has aged past its threshold or otherwise risks distorting multicriteria analysis. Archiving is deliberate rather than destructive: EAE moves a superseded dataset to an archive repository instead of deleting it and updates the record to point to the newer version, so that analyses run on the older layer remain reproducible. A FAIR metadata requirement underpins this. No layer is published in EAE without metadata carrying the necessary cautions, context and attributions; a downstream user needs to understand what they are looking at and how far to rely on it.Meeting the standard is not all-or-nothing. When a dataset falls short, the review flags it to the geography administrator rather than dropping it silently, and EAE attaches an additional caution to the layer so that the limitation is visible when the data is used. Removing a layer outright is a last resort: in many geographies an imperfect layer carrying an honest caution is more useful to a planner than no layer at all. Working Groups and the Review Cycle In EAE, geography administrators are responsible for their datasets throughout the data lifecycle. At an agreed frequency, the EAE working group reviews whether data remains suitable for the geography’s needs and whether licensing conditions still hold. Integrating these reviews into existing working group meetings makes the practice easier to maintain through staff turnover and competing priorities.A reviewer, usually the data quality assurance/quality control (QA/QC) lead, alongside the product owner, samples four to five datasets from different categories rather than checking each one. This provides a spot check of whether the QA/QC process is working while keeping responsibility with the geography administrator. Access rights and permissions are reviewed quarterly to ensure the terms of nondisclosure agreements and memorandums of understanding stay current as staff and partners change. A broader monitoring and maintenance review runs every six months.To keep reviews comparable across geographies and over time, each review is recorded in a standard monitoring report, described in the table below. Each report notes which datasets were examined, by whom, when and what was found.Table 1: Data monitoring report formatWhat it capturesFields recordedIdentificationName and long name; geography; category; data type (raster or vector); sourceCurrencyLast updated; next scheduled updateVerificationMetadata verified; QA check completed; access permissions reviewed; governance restrictions reviewedFindings and sign-offNotes and issues found; reviewed by; date of reviewAdapted from the EAE data standards and QA/QC framework, Appendix 1, which sets out the monitoring report format used to log each review.The Constraint: Finding Standardized Data and Data GapsData quality is one part of the puzzle. Often the harder part is finding relevant, standardized data. In many geographies, the layers that would inform a decision do not exist, sit behind incompatible licenses or come only at a resolution too coarse to be useful.Grid infrastructure shows the gap. High-voltage transmission lines are often mapped, while medium- and low-voltage networks are frequently missing or inaccessible, a real constraint for anyone working on electricity access. One study modeled these networks from nighttime lights and OpenStreetMap data, and its validation across 14 countries returned roughly 75% accuracy. That makes it a useful input, not a disqualified one, and it captures the real trade-off: In many places, the choice is not between a surveyed layer and a modeled one, but between a modeled layer and nothing. EAE prioritizes working with data shared by local working groups and falls back on OpenStreetMap (OSM) data and modeled datasets where local data is not yet developed or must remain restricted due to sensitivity and security concerns.Good data does not accumulate on its own. It takes sustained investment in collection, curation and validation, frequently in places where institutional capacity is thinnest. A standards framework does not remove that cost, but it makes the investment last, so that consistent handling preserves the effort of collecting one good dataset rather than eroding it further down the line.Distributed Ownership of the StandardsThis work is shared, not centralized. EAE’s data standards live with a distributed community of practice made up of geography administrators and partners, each holding local ownership of the tool in their own context and applying the standards to the data they know best. This shared ownership is the point: Because the credibility of every analysis rests on the data behind it, keeping to the shared standards is something the community has a direct stake in, not a rule handed down to it. For data quality, what matters is that the standards are a common commitment the community maintains together, not a constraint imposed from outside.Administrators setting up EAE in a new geography have the EAE data QA/QC guidelines and the data-addition manual, which translates these standards into a step-by-step process for evaluating, preparing and publishing a dataset.Looking Ahead: Confidence ScoringKnowing how much data quality matters, EAE has begun researching a confidence score that would attach to multicriteria analysis outputs. The score would draw on two properties of the underlying data: temporality and source type, which distinguishes primary or survey data from secondary, modeled or proxy data. Recent, directly observed data should carry more weight than an older modeled estimate, and the tool should show that difference rather than presenting both with equal visual authority. This would allow the analysis to record uncertainty rather than absorb it silently, especially when a modeled layer is the only option available or when swapping one population product for another moves a coverage estimate by tens of percentage points.A confidence layer would replace that binary with a calibrated judgment in which the strength of the evidence sits alongside the recommendation. A planner could then see both where the analysis points and how far the underlying evidence supports it, and could steer data-collection effort toward areas where low confidence constrains a decision.We will also research AI-assisted data stewardship, which could help parse metadata, compare a dataset’s documentation against EAE’s acceptance criteria, summarize caveats and flag data that does not meet the temporal threshold. This would support geography administrators, QA/QC leads and working groups.The Foundation of a Data-Informed PlanningData lifecycle and quality are not specific to EAE. The data standards and QA/QC framework can be adopted or adapted by any team running a geospatial planning tool. Teams in other sectors face the same lifecycle, the same acceptance decisions and the same problem of sustaining quality after launch, and can reuse (or adapt and contribute to) the checklists directly rather than building their own from scratch.Data quality is not a feature sitting behind the product. For a tool used to plan infrastructure and allocate capital, it is the product’s foundation. By building quality into the pipeline through published standards, consistent processing in Paver, an active review cycle run through working groups, FAIR metadata and, in the future, explicit confidence scoring, EAE gives governments and investors what a transparent methodology alone cannot: a reason to trust the map behind the map before they stake a decision on it.1 Resampling is the process of adjusting spatial data to a new resolution, such as downscaling for more detail or aggregating for a broader, coarser view.