A method that produces a plausible result has not automatically demonstrated that it works for its intended users. Validation requires a defined question, suitable reference evidence, and an evaluation design that reflects how the method will be used. This perspective examines which observations should be compared, how they should be sampled, and which errors matter in practice. It also considers the difference between evidence used to develop a method and evidence reserved for evaluating it. The objective is a test that can reveal weaknesses as well as confirm useful performance.
This working guide considers reading uncertainty in remote sensing products within ground-station data delivery. The immediate decision is to test delivery of usable observation products. Useful reference evidence includes transfer records, acquisition metadata and complete product checks. The case is an illustrative planning context, with assumptions to check against the actual sensor, landscape and project requirements.
Identify what the uncertainty describes
A satellite product is the result of a measurement chain, rather than a direct copy of conditions on the ground. Detector noise, calibration, atmospheric correction, geolocation and retrieval assumptions can all affect the final value. Their importance changes with the target. A small location error may matter greatly beside a narrow river, while an imperfect aerosol correction may dominate a water reflectance estimate.
Start by reading the variable definition, units, spatial support and quality documentation. An uncertainty layer may describe an estimated standard deviation, a confidence interval, disagreement among models or another quantity entirely. These measures are not interchangeable. A model's class probability also does not automatically represent a calibrated probability that its label is correct. Check how the quantity was obtained and whether it has been evaluated against independent observations.
Carry uncertainty into the decision
Separate systematic differences from random variation. Averaging many observations can reduce independent random noise, but it does not remove a shared calibration bias. Nearby pixels and repeated observations may share errors, so treating every value as independent can produce confidence intervals that are too narrow. When comparing dates, consider whether the same processing method, sensor characteristics and environmental conditions apply.
Quality flags identify conditions that deserve screening or special treatment. They may indicate clouds, saturation, poor retrieval geometry or inputs outside a method's tested range. Preserve the original flags and record the filtering rule. A strict mask can improve reliability while leaving a sample that overrepresents clear weather or easier terrain. Report missing coverage alongside the measurements that remain.
For a practical assessment, perturb influential inputs within defensible ranges and examine how the result changes. Compare alternative valid processing choices, stratify errors by landscape or season and use independent reference data. Classification summaries should include class specific errors rather than relying on one overall accuracy figure. For continuous estimates, examine bias and the distribution of residuals as well as an aggregate error statistic. A useful final product states the tested domain, describes meaningful limitations and shows whether the uncertainty is small enough for the intended decision.
The setting: Ground-station data delivery
A ground-station workflow connects acquired observations with the data products a user eventually receives. Successful communication is one part of that chain, while completeness, timing, processing and interpretation determine whether the observation is useful. Define what the delivery system must preserve, including acquisition identifiers, instrument settings, timing and the metadata needed for later calibration or geolocation. A received file should not automatically be treated as a complete and ready-to-use image product.
A test sequence should follow data from a known input through transfer, storage, processing and delivery. Record missing segments, failed attempts and changes in processing state instead of hiding them behind a simple success label. The operational question may concern when an agricultural user receives a current observation or whether a researcher can reproduce a product. These uses require traceable timestamps and a clear distinction between preliminary material and the version accepted for analysis.
Validation can use controlled test data and documented end-to-end checks before any independent satellite service becomes operational. Assess whether the delivered product retains its expected content and whether failures are represented clearly enough for a user to respond appropriately. A concept timetable cannot establish actual service availability. The practical output is evidence about the delivery workflow, its measured constraints and the information users need to judge the age, completeness and processing status of an observation.
Use observation labels with consistent meanings
A reference label should describe an observation that another person could recognise or reproduce. Broad terms such as good, damaged, or unusual can conceal substantial differences between collection teams. Begin by defining the feature or condition being recorded, the observation unit, and the evidence needed to assign a label. Where possible, distinguish a direct measurement from an interpretation made by the observer.
Provide a short collection guide with examples, boundary cases, and a way to record uncertainty. Train observers using shared cases before relying on independently collected labels. If two observers disagree, preserve the disagreement long enough to understand its cause. It may reveal an unclear definition, a genuinely ambiguous condition, a difference in observation scale, or a problem in matching the reference record to the remote measurement.
Retain the context that makes the label interpretable. Location, time, collection method, observer role, and relevant surface conditions can be essential for later review. Changes to the label scheme should be documented and related to earlier versions where a defensible correspondence exists. Consistent labels do not eliminate uncertainty, but they make it possible to investigate disagreement, evaluate analytical methods, and identify where better definitions or additional measurements would improve the evidence.
Design an evaluation that reflects deployment
A model evaluation should resemble the conditions in which the model will be used. Start by identifying which relationships the development data share: nearby locations, repeated observations of the same unit, common collection teams, or measurements from a single period. These relationships can make records less independent than they appear. An evaluation that ignores them may test the ability to recognise familiar conditions rather than the ability to operate in a new setting.
Choose a split strategy that matches the intended transfer. Holding out locations, seasons, operational units, or acquisition conditions can answer different questions. Document which question each split is designed to address, and keep the held-out evidence separate while selecting features and tuning the method. Repeatedly inspecting the same evaluation set to improve a model gradually makes that set part of development, even if it retains its original label.
Report results at a scale that is meaningful to the user. A strong average can hide poor performance for important subgroups or unusual conditions. Examine error patterns and retain examples of both successful and unsuccessful predictions. Where the evidence covers only a limited range of conditions, describe that range explicitly. The evaluation then becomes a practical guide to suitable use, further testing, and the collection of additional reference observations.
Bring the evidence back to the decision
The validation record should explain the population represented by the test, the reference observations used, and the limitations of the comparison. Report meaningful errors alongside useful successes, and retain examples that challenge the method. Treat wider deployment as a new opportunity to evaluate transfer across relevant conditions. A well-designed validation programme gives users a defensible basis for deciding where the method is useful, where additional evidence is needed, and where another approach may be appropriate.
Sources & further reading
Explore the bhumisat spacecraft concept and our vision towards 2030, or return to the journal.



