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 radiometric calibration and reflectance consistency in cassava production plots. The immediate decision is to select plots for crop-stage assessment. Useful reference evidence includes planting records, canopy cover and dated plot visits. The case is an illustrative planning context, with assumptions to check against the actual sensor, landscape and project requirements.
Define the quantity that the instrument records
Radiometric calibration establishes a documented relationship between an instrument's recorded output and the physical quantity it is intended to represent. A stored pixel value may be a digital number, a scaled radiance or reflectance value, or another processed representation. Those alternatives should not be treated as interchangeable. The calibration chain explains how the recorded signal becomes an interpretable measurement and which corrections have already been applied. Understanding that chain is essential before comparing dates, combining bands or transferring a model between instruments.
Instrument response can vary across detectors, channels and operating conditions. Calibration therefore concerns more than selecting one conversion coefficient. A study may need to consider offsets, response differences, noise, saturation and changes in behaviour over time. The appropriate characterisation depends on the instrument and its intended use. For a future payload concept, these are requirements to investigate rather than capabilities established by an attractive render. The presence of a multispectral camera alone does not demonstrate that its channels produce consistent, quantitatively comparable observations.
Track consistency through the complete chain
At the product level, begin with the supplier's definitions and metadata. Check scale factors, additive offsets, fill values and valid ranges before deriving an index or statistical summary. Preserve the processing version so that an output can be reproduced later. Inspect whether extreme values correspond to real surfaces or to invalid observations. When a dataset is updated, compare representative old and new results rather than assuming that a familiar file name guarantees an unchanged numerical meaning throughout the archive.
Validation should cover the full chain that supports the stated application. A stable response in a laboratory test is valuable evidence, while operational imagery introduces additional effects from atmosphere, geometry and scene complexity. Use independent references suited to the measurement scale and report the relevant uncertainty. Maintain a record of corrections and their limitations when delivering results to users. A traceable radiometric workflow supports meaningful comparisons and helps separate a change in the observation system from a change in the landscape that a monitoring service is trying to describe.
The setting: Cassava production plots
Cassava plots may have different planting dates, harvest strategies and mixtures of neighbouring vegetation. The commercial product develops below ground, while an optical satellite primarily observes the surface and canopy. This distinction is fundamental to the monitoring objective. A visible vegetation pattern can support investigation of plot development or cover, but a claim about root production needs a demonstrated relationship with representative harvest measurements for the relevant production setting.
Start with boundaries that describe the area actually planted and a calendar that captures establishment and any partial harvesting. Small plots can contain substantial mixed pixels along their margins, particularly where paths, trees or other crops surround them. Track the proportion of valid observations and the spatial support used for each summary. Repeated measurements should be compared in a way that accounts for plot age rather than assuming all cassava fields share a single seasonal cycle.
Field visits should record crop stage, canopy continuity, intercropping and management events that might explain the image pattern. Sampling locations need enough context to identify whether they represent the plot interior or an edge mixture. If an analytical model is developed, reserve independent plots or periods for evaluation and communicate its limits. A dependable first service may be a plot inventory and inspection workflow, with more specific production claims added only when the evidence supports them.
Design field observations around the information need
Field observations are most useful when they answer a defined question within the monitoring workflow. Begin by identifying the variable, condition, or management event that needs to be recorded and the geographic unit to which it applies. Distinguish direct measurements from an observer's interpretation. A consistent record of crop stage, management timing, and visible conditions can provide valuable context even when the field team is not collecting a complex set of measurements.
Prepare a collection protocol that reflects the variability of the site. Define where observations are made within a unit, how many locations are represented, and how the position and time are recorded. Avoid selecting only the most dramatic or convenient examples. Include apparently normal conditions and records that might challenge the remote interpretation. Where a measurement method has known limitations, preserve the information needed to assess whether those limitations affect the intended comparison.
Link field records to the corresponding remote observations without losing their original detail. Geographic matching, timing, and the scale represented by each record should remain visible. Review disagreements with the collecting team before changing a label or excluding an observation. A disagreement may reveal a timing mismatch, a mixed surface, or an incomplete interpretation rather than a simple error. The field programme should improve the understanding of the monitoring process as well as provide evidence for evaluating its outputs.
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.



