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 testing whether a remote sensing model transfers during multisensor field campaigns. The immediate decision is to align observations across measurement scales. Useful reference evidence includes coordinated timestamps, sample support and instrument reference records. The case is an illustrative planning context, with assumptions to check against the actual sensor, landscape and project requirements.
The setting: Multisensor field campaigns
A multisensor field campaign brings together observations that may represent different surfaces, times and spatial scales. A ground measurement can describe a small sample, a drone product a local scene and a satellite observation a wider footprint. Their values should not be expected to match automatically. Define the variable of interest and the area represented by each method before treating the campaign as a direct comparison between interchangeable measurements.
Coordination should account for acquisition timing, crop or environmental change, positioning and the conditions relevant to each instrument. Record whether a sample represents a plot interior, an edge mixture or another setting that influences its relationship with imagery. A common coordinate reference helps align the observations, but it does not remove differences in measurement support. Aggregation and comparison rules should be chosen to address those differences explicitly rather than only making the maps look aligned.
The evaluation should preserve uncertainty in both the reference data and the remotely sensed products. Include representative ordinary conditions as well as unusual sites, and keep development samples separate from independent assessment where a model is being tested. A successful relationship within one campaign does not establish transfer to all other settings. The useful outcome is a traceable account of how the observations relate, where they disagree and what those disagreements reveal about the proposed application.
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.
Deliver quality flags that users can interpret
Quality information should describe the condition of an observation in terms that remain useful through the workflow. A single valid or invalid label may be sufficient for some calculations, but operational users often need to know why an observation was qualified. Distinguish acquisition limitations, processing failures, uncertain reference matches, and conditions that make interpretation ambiguous. These categories point towards different responses and should not be merged without a clear purpose.
Define a small, documented set of flags with stable meanings. Record whether a flag applies to a complete scene, a geographic unit, an individual measurement, or a derived result. A unit containing a mixture of qualified and unqualified observations may also require a coverage measure. Include the rule used to combine flags when summarising data so that a favourable average does not conceal a substantial fraction of unusable input.
Test the flags with the people receiving the product. Ask what they would do differently for each state and whether the terminology supports that choice. Keep the original flags available even if the interface presents a simpler summary. When flag definitions change, version the rules and review their effects on historical comparisons. Useful quality communication makes the condition of the evidence visible at the moment when a user decides how much confidence to place in the result.
Budget the complete processing workflow
A processing budget should cover the complete path from source observations to a maintained product. Data access and computing are only part of that path. Include preparation, geographic updates, reference collection, expert review, delivery, and support where those activities belong to the service. A calculation that runs cheaply on a small demonstration can still require substantial effort when the surrounding work is repeated across many locations.
Measure representative tasks rather than assuming that every observation costs the same amount to process. Record data volume, processing time, retry frequency, manual interventions, and the conditions that create unusually demanding cases. Separate fixed setup work from work that grows with the number of observations or users. These records support decisions about batching, caching, storage policy, and the point at which a different processing design becomes worthwhile.
Connect the budget to a defined level of service. Faster delivery, more frequent observation, broader coverage, or more detailed review can each change the resources required. Describe these trade-offs before promising a delivery pattern that the workflow has not demonstrated. Review the budget using operational evidence and preserve enough detail to understand changes between processing versions. The aim is a sustainable service whose resource choices remain aligned with the information that users value.
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.



