Wait, what? A satellite-data catalogue says Scene Cloud Cover: 10%. A learner is interested in a small reservoir near one corner of the image and says, “Great. That means my reservoir is 90% clear.” Then the browse image opens—and almost every cloud in the scene happens to sit over the reservoir.
The catalogue and the learner can disagree without the catalogue being wrong. The mistake is a spatial-denominator mistake. A percentage calculated over a whole satellite scene describes the whole scene. It does not automatically describe every smaller place inside it. Ten per cent of a large image can be cloudy while one small target is completely cloud-covered, completely clear, or anything in between.
This is the real-world evidence-transfer job in PSLE Science Reality Lab Vol No.433: learn how to evaluate an aggregate spatial percentage without pretending that the same percentage applies locally. The object is satellite metadata, but the habit transfers to biology quadrats, class surveys, damaged leaves, shaded areas, contamination maps and any scientific statement where a whole-area percentage is mistaken for a local condition.
Current USGS Landsat guidance explains that Scene Cloud Cover is an estimate of percentage cloud cover calculated over an entire Landsat scene. USGS also distinguishes it from Land Cloud Cover, which is calculated over land pixels, and provides pixel-level quality information that can classify cloud, cloud shadow and other conditions. The 2026 PSLE Science assessment objectives require interpreting and analysing information and evaluating observations, information and methods. Here, the key question is simple: What area is the percentage actually about?
Quick Answer
No. “Scene Cloud Cover = 10%” does not mean every target inside the scene is 90% clear. It means the cloud-cover estimate is about the scene as a whole. Clouds may be spread thinly, clustered in one corner, concentrated over land, concentrated over water, or positioned directly over the feature you care about.
To evaluate whether a particular target is usable, inspect evidence at the target: a cloud or quality mask, a browse image, relevant pixel classifications, acquisition time and the product’s quality documentation. An aggregate percentage is useful for screening, but it is not a local guarantee.
The Owned Learner Job
This page owns one narrow job: distinguishing a whole-scene cloud-cover percentage from the cloud condition at one chosen target area.
It does not become a generic percentages lesson, a satellite-physics owner, a remote-sensing tutorial or a weather-prediction page. Existing eduKateSengkang owners continue to own graph reading, sampling, observation versus inference, evidence selection and comparison. Reality Lab applies those skills to a spatial metadata claim.
Useful routes include How to Tell Observation, Inference, Prediction and Explanation Apart in PSLE Science and How to Compare PSLE Science Counts When Groups Are Different Sizes. The Reality Lab question is what happens when the denominator is an area.
Build the Scene From Scratch
Imagine an original 100-cell scene. Each cell represents the same area. Ten cells are classified as cloudy and ninety as clear. Scene cloud cover is therefore 10% in this simplified teaching model.
Now define a small target called Lake A that occupies five cells in one corner. There are many possible arrangements of the same ten cloudy cells:
| Arrangement | Cloudy cells in whole scene | Cloudy cells over Lake A | What Scene Cloud Cover still says |
|---|---|---|---|
| Clouds nowhere near Lake A | 10 of 100 | 0 of 5 | 10% |
| Two clouds over Lake A | 10 of 100 | 2 of 5 | 10% |
| Lake A fully cloud-covered | 10 of 100 | 5 of 5 | 10% |
The scene percentage is identical in all three cases. The local usability of Lake A is completely different. That is the evidence gap the learner must notice.
Observed, Claimed and Inferred
Observed
You can observe the catalogue value, scene identifier, acquisition date, browse image, metadata fields and quality-mask information supplied with the data product.
Claimed
The metadata claims an estimated fraction of the scene covered by cloud under the product’s stated cloud-detection method. USGS describes Scene Cloud Cover as a full-image estimate of percentage cloud cover for the Landsat scene.
Inferred too strongly
“My reservoir must be 90% clear” is a local inference. It assumes that clouds are distributed uniformly or proportionally across every sub-area. Scene Cloud Cover does not make that promise.
The Denominator Is the Scene
A percentage always belongs to some whole. In “10% scene cloud cover,” the whole is the defined scene used by the metadata calculation. The number becomes misleading when a reader silently changes the denominator from “all relevant pixels in the scene” to “pixels over my chosen school, forest, reservoir or farm.”
This is exactly like saying 10% of a class is absent and then claiming every friendship group must have 10% of its members absent. The class-level percentage does not fix the composition of every small group.
Scientific reasoning therefore keeps two questions separate:
- Whole-scene question: How cloudy is the scene overall?
- Target question: Is the feature I need clear enough for my purpose?
Worked Case 1: Ten Per Cent Cloud, One Hundred Per Cent Trouble
A researcher wants to compare the colour of a small wetland between two dates. Scene A has 8% cloud cover and Scene B has 15% cloud cover. The learner immediately chooses Scene A because 8 is smaller than 15.
Then the browse images are inspected. In Scene A, one cloud sits directly over the wetland. In Scene B, the wetland is clear and most clouds are far away.
Which scene is better for the wetland question? Scene B may be more useful even though its overall cloud percentage is higher. The relevant evidence is not only the global ranking; it is the cloud condition at the target.
Worked Case 2: Scene Cloud Cover and Land Cloud Cover
USGS Landsat metadata can distinguish Scene Cloud Cover from Land Cloud Cover. Why might those differ? Imagine a coastal scene containing large areas of ocean and land. Clouds could be concentrated over the ocean while most land pixels are clear, or the reverse.
An original teaching example:
| Metadata field | Practice value | What it summarises |
|---|---|---|
| Scene Cloud Cover | 20% | Cloud estimate across the full scene |
| Land Cloud Cover | 5% | Cloud estimate across land portion |
The numbers are not contradictory because their denominators differ. A student must read the field definition before comparing or using it.
Worked Case 3: “0% Cloud” Does Not Mean “Perfect Image”
Suppose a catalogue reports 0% cloud cover. Can the learner conclude, “Every pixel is scientifically perfect”?
No. Cloud cover is one quality dimension. Other issues may still matter: cloud shadows, haze, sensor saturation, missing data, edge effects, snow or ice classification, water glint, atmospheric conditions, geometric accuracy, or whether the target was even inside the useful part of the scene. A low cloud percentage supports a narrow claim about detected cloud, not a universal quality certificate.
This is a recurring Reality Lab rule: a good score for one property is evidence about that property, not automatic proof that every other property is good.
Worked Case 4: “Cloud Cover = −1”
USGS data dictionaries use special metadata codes in some contexts; for the relevant cloud-cover field, a value such as −1 can indicate that cloud cover was not assessed rather than “minus one per cent cloud.”
A learner who treats every number as an ordinary measured quantity can produce nonsense. Scientific tables sometimes use special codes for missing, unavailable, not assessed or fill values. The legend or data dictionary is part of the evidence.
This does not turn this article into a missing-data owner. The point is local to the object: metadata values need definitions before interpretation.
Representation Check: Metadata Is a Summary of a Spatial Pattern
“10% cloud cover” compresses a complicated two-dimensional pattern into one number. Compression is useful: a researcher can quickly screen thousands of scenes. But compression throws away location information. The same 10% can be arranged in countless spatial patterns.
Whenever a scientific representation compresses many measurements into one summary, ask what information was lost. A mean loses individual values. A total loses distribution. A percentage can lose location. An index can combine multiple inputs. A map class can hide within-class variation.
The correct response is not to reject summaries. It is to use the summary for the job it can do, then return to more detailed evidence when the question requires detail.
Method Check: How Does a System Decide Which Pixels Are Cloudy?
Satellite cloud-cover metadata are not produced by a person counting white blobs with a finger. Modern products use algorithms that classify pixels using spectral and other information. USGS describes cloud-cover assessment and supplies validation datasets containing reference masks for cloud, clear sky and cloud shadow so algorithms can be evaluated.
For a Primary 5/6 learner, the important method lesson is this: the percentage depends on a classification method. Thin cloud, bright surfaces, snow, haze and shadows can be difficult cases. Therefore the number is an evidence product with a method behind it, not an infallible count of obvious cartoon clouds.
Alternative Explanations When the Browse Image Looks Cloudier Than the Metadata
Suppose the catalogue says 10% cloud, but the learner feels the image “looks half cloudy.” Several possibilities exist:
- The cloudy area is visually bright and attracts attention even though it occupies a smaller area.
- The browse image is cropped or displayed differently from the full scene used for the percentage.
- Cloud shadow or haze is being mistaken for cloud.
- The learner is looking at a land-only region while the metadata describes the full scene.
- The cloud-detection algorithm and human visual judgement disagree on difficult pixels.
- The metadata field belongs to a different product version or acquisition than the image being viewed.
A strong scientific response checks identifiers, definitions and masks before declaring one source wrong.
Evidence That Strengthens the Claim “My Target Is Clear”
- The exact target location is confirmed inside the correct scene.
- A quality mask classifies target pixels as clear.
- The browse image independently appears clear over the target.
- Cloud-shadow and haze information do not contradict the claim.
- The acquisition date and time match the intended comparison.
- The product documentation explains the cloud-cover method and quality fields.
Evidence That Weakens the Local-Clearness Claim
- Only a scene-level percentage is known.
- The target is tiny compared with the scene.
- Clouds are spatially clustered.
- The target falls under a cloud or cloud-shadow mask.
- The metadata field is not assessed or uses a special code.
- The scene identifier or acquisition date is mismatched.
Tempting but Invalid Reasoning
| Tempting statement | Why it fails | Scientific repair |
|---|---|---|
| “10% scene cloud means my target is 90% clear.” | The scene percentage does not fix the local spatial distribution. | Inspect target-level mask or image evidence. |
| “Scene A has less cloud overall, so it is always better for my target.” | Cloud placement matters. | Check where the clouds are. |
| “0% cloud means every pixel is perfect.” | Cloud is only one quality dimension. | Check other quality flags and the scientific task. |
| “−1 means minus one per cent cloud.” | Metadata may use special codes. | Read the data dictionary. |
| “The algorithm says clear, so it cannot be wrong.” | Classification methods have limits. | Use quality documentation and independent visual or validation evidence when important. |
How Far Can the Conclusion Travel?
Scene Cloud Cover can help screen scenes quickly. A lower value can suggest that less of the overall scene is classified as cloudy. It can be useful for sorting and filtering large archives.
It cannot by itself prove that a particular school, lake, crop field, forest plot or coastline is clear. Nor does it prove that a clear pixel is scientifically suitable for every possible analysis. The conclusion should travel only as far as the denominator and method allow.
PSLE-Style Transfer Case: Ten Per Cent of Leaves Are Damaged
This is an original transfer case, not an examination question.
A gardener examines 100 leaves across ten plants and finds 10 damaged leaves. A learner says, “Therefore every plant has exactly one damaged leaf.”
The conclusion is not supported. All ten damaged leaves could belong to one plant, or they could be spread across several plants. The overall fraction tells us the total pattern across all sampled leaves, not the distribution among individual plants.
This is the same reasoning as scene cloud cover. The scientific habit survives the change of topic because the evidence structure is the same: aggregate proportion does not determine local distribution.
Delayed Independent Return
Tomorrow, draw a 10 × 10 grid. Shade exactly ten squares as cloudy. First spread them evenly. Then erase and place all ten in one corner. In both drawings write “Scene Cloud Cover = 10%.” Now choose a five-square target in that corner and explain why its cloud condition can change even though the scene percentage remains the same.
If you can explain that without using the word “satellite,” you have learned the transferable part.
Explained Practice
- A scene is 5% cloudy overall. Can a tiny island inside it be completely cloud-covered? Explain.
- Why might Scene Cloud Cover and Land Cloud Cover differ?
- What local evidence would you seek before saying a reservoir is clear?
- Why does a whole-scene percentage lose information about location?
- If an algorithm and a human disagree about a thin cloud, what should a scientist do rather than simply choose the preferred answer?
Suggested reasoning: yes, because the small target may contain most or all of the scene’s cloudy pixels; the two metadata fields have different spatial denominators; inspect quality masks and target imagery; one summary number compresses many possible arrangements; and use definitions, validation evidence and additional observations to investigate disagreement.
Parent and Tutor Teaching Guide
Use coloured counters on a sheet divided into four boxes. Place 10 red counters and 90 white counters across the whole sheet. Ask, “What percentage of all counters are red?” Then move all ten red counters into one box without changing the total. Ask, “Did the overall percentage change? Did the local box change?”
Next use a familiar science setting. Say that 20% of the total leaf area in a tray is shaded. Ask whether every leaf must be 20% shaded. Then ask what extra evidence would be needed to describe one chosen leaf. This makes the denominator visible before satellite metadata is introduced.
The teaching goal is not calculation speed. It is the reflex to ask “percentage of what area?” and then “does my question concern the same area?”
Authoritative Sources and Further Reading
- U.S. Geological Survey: Landsat Collections Land Cloud Cover — explains Scene Cloud Cover as an estimate over the entire scene and distinguishes Land Cloud Cover.
- U.S. Geological Survey: Landsat Collection 2 Data Dictionary — defines metadata fields, ranges and special values used in Landsat products.
- U.S. Geological Survey: Cloud Cover Assessment Validation Datasets — describes pixel-level cloud, clear-sky and cloud-shadow reference masks used to validate algorithms.
- Ministry of Education: Primary Science Syllabus 2023 — evidence, scientific inquiry, communication and healthy scepticism.
- SEAB: 2026 PSLE Science Syllabus — interpreting, analysing and evaluating scientific information and communicating reasoning.
Quiet Return: A Percentage Has a Place
“10% cloud” sounds precise, and it can be a useful scientific summary. But the precision belongs to the defined scene-level calculation. It does not magically attach itself to every smaller place inside the image.
When a real-world claim gives a percentage over space, keep one question close: Where is the denominator? Then look at the place you actually care about. That is how a summary becomes evidence instead of a shortcut.
