A convincing image can still contain measurements that are unsuitable for a particular task. Quality control asks whether the inputs, transformations, and outputs meet the requirements of the intended application. This perspective follows evidence through the workflow, from acquisition metadata and reference observations to the final interpretation. It encourages teams to define acceptance conditions before seeing the result, rather than adjusting them to favour a preferred conclusion. The purpose is a repeatable way to distinguish usable observations, observations requiring qualification, and observations that should be excluded from a decision.
This working guide considers testing whether a remote sensing model transfers across wheat growing areas. The immediate decision is to review fields with unusual seasonal timing. Useful reference evidence includes sowing dates, crop stage and representative field observations. The case is an illustrative planning context, with assumptions to check against the actual sensor, landscape and project requirements.
The setting: Wheat growing areas
A wheat monitoring programme should begin with a clear description of its production setting. Sowing windows, variety choices, water supply and management practices can vary across a region. Fields at different developmental stages should not be judged against a single image threshold as though they followed one uniform calendar. A useful field register combines boundaries with sowing information and a record of the operations that can influence the observed canopy through the season.
Repeated observations can help describe the timing and spatial consistency of crop development. Inspect the quality of each acquisition and distinguish persistent patterns from a single questionable measurement. Field margins, tracks, bare patches and neighbouring crops can affect irregular or narrow units. Where comparisons involve different image products, make sure the processing and spatial support are suitable before interpreting a numerical difference as a crop response.
The field investigation should follow a defined question, such as whether a section has uneven establishment or a different development stage from its peers. Record the evidence used to resolve that question, including places where the image alert did not correspond to an actionable condition. If a model estimates harvest outcomes, evaluate it against appropriate production records and separate training from assessment. The reporting system should communicate uncertainty alongside the spatial pattern that prompted attention.
Identify what changes between domains
A model trained on one region or season learns relationships present in its development data. Applying it elsewhere assumes those relationships remain useful. New crop varieties, forest types, soil colours, management practices, illumination, or sensor characteristics can violate that assumption. Define the intended transfer explicitly: a new district, another year, another sensor, or a different class distribution. Each change creates a different test. A model that transfers across nearby fields has not thereby demonstrated reliable performance across climates or observation systems.
Keep the measurement pipeline consistent before attributing differences to the model. Reflectance scaling, band order, cloud masking, spatial resolution, composite windows, and missing value handling can all change predictions. Sensors with similarly named bands may have different spectral responses. Harmonization can reduce some differences but does not establish equivalence for every target. Examine predictor distributions, missing data rates, and examples from the new setting. Flag inputs outside the conditions represented during training so that extrapolation remains visible.
Test the destination, then adapt carefully
Design evaluation splits around the intended deployment. Hold out geographic regions to test spatial transfer and later years to test temporal transfer, keeping neighboring samples or repeated observations of the same site together when needed. Report performance for meaningful destination groups rather than only a pooled score. Class prevalence affects some metrics, so interpret changes alongside the underlying confusion matrix or residual distributions. A random pixel split can look excellent while providing little evidence about performance in an unseen region.
If local labels are available, reserve a destination assessment set before fine tuning or recalibration. Use the remaining observations to compare straightforward baselines with more complex adaptation methods. Check whether improvement comes from genuine transfer or simply from adding representative local data. Retain an option to withhold predictions when input quality or novelty makes them unreliable. Document the training domain, destination evidence, preprocessing versions, and conditions where use is unsupported. Continue monitoring once new observations arrive, since seasonal and environmental shifts can emerge after an initial evaluation. Transferability is a measured property of a particular model, workflow, target, and destination; it should not be inferred from architecture name or strong performance on the original test set.
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.
Translate the decision into observation requirements
Start with the decision that the information is intended to support. Describe the geographic unit, the time available for acting, and the consequence of an incorrect interpretation. A requirement such as detecting an unusual pattern across management blocks is different from identifying its cause. Keeping that distinction explicit helps a team select measurements that are relevant to the application and prevents later evaluation from drifting towards whatever result happens to be easiest to demonstrate.
Build a compact requirements matrix linking each decision to the evidence needed, the proposed observation source, and an acceptance condition. Include geographic accuracy, temporal relevance, quality screening, and the context required for interpretation. The matrix should also identify the evidence used to justify each requirement. An engineering preference or a convenient data source can inform a study, but neither should silently replace a documented information need.
Review the matrix with the people who will use the result. Ask how their response would change when information is delayed, incomplete, or uncertain. Record unresolved requirements as questions for a pilot rather than converting them into confident specifications. This produces a practical framework for comparing technical options, designing tests, and explaining why an observation is useful to the intended workflow.
Bring the evidence back to the decision
Quality control is most useful when its results travel with the information being delivered. Preserve the reasons for accepting or rejecting an observation, and expose limitations that could change the user's interpretation. Review recurring failures as clues about the design of the workflow. A clear quality record supports consistent comparisons between operators and processing versions, while helping the team decide whether a problem originates in acquisition, preparation, interpretation, or the underlying observation requirements.
Sources & further reading
Explore the bhumisat spacecraft concept and our vision towards 2030, or return to the journal.



