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 mixed pixels and field boundary effects across river floodplain areas. The immediate decision is to locate areas needing flood-impact review. Useful reference evidence includes dated water observations and representative field impact records. The case is an illustrative planning context, with assumptions to check against the actual sensor, landscape and project requirements.
The setting: River floodplain areas
A floodplain contains channels, low-lying fields, settlements, wetlands and surfaces that may be wet or dry under ordinary conditions. A flood-monitoring workflow should distinguish a new observation of water from the baseline distribution expected in that setting. Define the event period and the intended output, such as visible inundation extent or locations for field impact assessment. These products do not automatically establish water depth, damage or the cause of a particular local condition.
Cloud cover can limit optical observations during an event, while vegetation and structures may obscure water at the surface. Other observation types can contribute complementary evidence but require interpretation appropriate to their measurement physics. Compare suitable baseline and event acquisitions, check alignment and preserve the actual observation times. A map assembled from different dates should make those differences clear rather than implying that the whole floodplain was observed simultaneously.
Field validation should record when water or impacts were observed, the location and the area represented by each report. Conditions can change between an image acquisition and a visit, so apparent disagreement may reflect timing rather than a simple mapping error. Sample different land-cover settings and include places outside the mapped extent. A useful product communicates the evidence for observed inundation separately from estimates of damage or operational consequences that require further measurements.
A pixel can represent several surface components
A satellite pixel represents an observation over a ground area, influenced by the instrument's spatial response and the scene within it. That area may contain several materials, such as crop canopy, soil, shade, water or a road. A mixed response is therefore common, especially near boundaries and in fragmented landscapes. It should not be treated automatically as a processing defect. The analytical question is whether the mixture supports the intended interpretation and whether a model trained under one mixture remains applicable under another.
Field geometry affects the number and character of observations available. A compact large field can contain many interior pixels, while a narrow or irregular field may consist mostly of boundary observations. Simply counting pixels does not establish how much independent information a dataset provides. Resampling onto a finer grid also does not remove the original mixing. Before selecting a monitoring product, examine the relationship between field dimensions, relevant landscape features and the measured spatial characteristics of the imagery that will support the service.
Choose a reporting unit that the data can support
Boundary exclusions can help investigate sensitivity to neighbouring surfaces, but they need a consistent rule and a clear purpose. Excluding a fixed distance may remove much of a small field while leaving a substantial interior in a larger one. Report the area retained and consider whether the remaining pixels represent the management unit fairly. Compare full-field and interior summaries during method development. A large difference can reveal a boundary or mixture problem that deserves attention before an alert is presented as a crop condition.
Validation should use observations collected at a scale compatible with the satellite interpretation. A measurement from a single plant or a small isolated plot may not describe the larger mixed area observed from orbit. Document the spatial support of both datasets and consider stratified sampling across relevant components. If the application requires information finer than the data can support, seek a suitable complementary source or change the question. Recognising mixed pixels helps an analyst design a credible product instead of assigning field-level precision to an observation that contains several different surfaces.
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.
Connect observations within a catchment
Water-related observations often depend on processes beyond the immediate pixel or sampling location. Begin by identifying the geographic relationships that are relevant to the question: upstream areas, connected drainage paths, contributing land uses, and the receiving water body. A local pattern may be easier to interpret when placed within that wider setting. The catchment boundary is a useful organising unit, but its suitability should be checked against the process and observation scale.
Combine the remote observation with contextual records that can help explain its timing and spatial pattern. Rainfall, water level, land management, and known interventions can influence what is observed. Preserve the dates and geographic meaning of those records rather than treating them as timeless attributes. Where drainage connections are uncertain or change with conditions, state the limitation and consider whether additional mapping or reference observations would materially improve the interpretation.
Plan reference sampling across locations that distinguish plausible explanations. A set of measurements concentrated at an easily accessible outlet may provide a different view from one that also represents tributaries, source areas, and contrasting conditions. Document how each location contributes to the evaluation. Interpret spatial agreement with care, because connected observations can share influences and may not provide independent evidence. The resulting workflow should connect local measurements with a defensible account of the wider system they help describe.
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.



