PSLE-SCI-REALITY-0040
Wait, What? Two pixels can both have numbers, but one may carry much weaker evidence.
A scientific map shows two neighbouring squares. Pixel A has a value of 0.72. Pixel B has a value of 0.74.
If you only read the numbers, you might conclude that Pixel B definitely has slightly more of the measured property.
But many scientific datasets contain another layer: quality information. A quality flag can tell you whether clouds affected the observation, whether a correction was difficult, how many observations contributed to a pixel, or whether a measurement passed the product’s preferred quality tests.
NASA’s MODIS surface-reflectance quality products, for example, include information about cloud and cloud shadow, aerosols, correction states and the number of observations contributing to a pixel. The number you see is only part of the evidence. The quality information tells you how much confidence that number should carry.
Quick Answer
A displayed value answers “What number did the data product provide?” A quality flag answers a different question: “Under what conditions was that value obtained or processed, and how suitable is it for this use?” Before comparing values, check whether they have comparable quality. Availability is not the same as reliability.
Reality Lab rule: Read the value and the evidence about the value.
The learner job this page owns
This Reality Lab applies PSLE Science reasoning to one object: a scientific dataset in which values have separate quality-assurance information. It does not create a new owner for satellite science, cloud detection, calibration or statistics. It teaches a learner to stop treating all visible numbers as equally strong evidence.
Original case: two lake pixels
Imagine a fictional satellite map estimating a water-surface property across a lake.
| Pixel | Displayed value | Quality information |
|---|---|---|
| A | 0.72 | Clear conditions; 6 usable observations |
| B | 0.74 | Thin cloud present; 1 usable observation |
| C | 0.60 | Clear conditions; 5 usable observations |
A student writes: “Pixel B definitely has the highest value because 0.74 is the largest number.”
That conclusion ignores evidence quality. Pixel B does have the largest displayed value, but its observation conditions are weaker. The difference between 0.72 and 0.74 is also small. A careful learner would say that B has the largest reported value, but the comparison needs caution because B’s quality information is poorer.
Value and quality are different variables
A common mistake is to imagine quality as a secret correction to the number. It is usually better to think of two separate questions:
- Value: What result does the dataset report?
- Quality: What conditions, checks or limitations describe that result?
A low-quality flag does not automatically tell you the true value. It tells you that the reported value deserves more caution.
Why quality information exists
Scientific instruments work in real environments. Clouds can block a satellite’s view. Shadows can alter reflected light. Aerosols can change the path of light through the atmosphere. A sensor may obtain several observations for one cell and only one observation for another. Processing algorithms may succeed easily in one place and struggle in another.
A quality layer records some of this context so users do not have to pretend every number was produced under identical conditions.
Quality flag does not mean “good” or “bad” in one universal way
Different datasets use different quality systems. One product may mark cloud contamination. Another may record how many observations contributed to an average. Another may flag instrument saturation or whether a correction was applied.
Never memorise “flag 0 is good” or “flag 3 is bad” as a universal rule. Read the documentation for that dataset. The meaning of the flag is part of the evidence.
The hidden-layer problem
A public visualisation may show only the colourful value layer. Quality information may be stored in a separate file, switch, legend or metadata field. This can make all pixels appear equally trustworthy even when the scientific product itself keeps careful records of their different quality states.
That is not automatically deceptive. A map designed for quick viewing cannot display every technical layer at once. The learner’s job is to know when the hidden layer matters enough to inspect.
When should quality flags matter most?
- when two values are very close;
- when the conclusion depends on one unusual pixel;
- when a key location is cloudy, shadowed or otherwise difficult to observe;
- when different groups have systematically different quality;
- when a trend appears only after including low-quality observations;
- when the dataset documentation explicitly recommends filtering by quality.
A six-step quality audit
- Name the measured quantity. What does the displayed number mean?
- Find the quality information. Is there a flag, observation count, cloud mask or reliability layer?
- Decode it from the source. Do not invent flag meanings.
- Compare like with like. Are the values you are comparing of similar quality?
- Check whether the conclusion depends on weak-quality data.
- State the limit. If quality differs, reduce the strength of the conclusion rather than pretending the problem does not exist.
Worked reasoning case 1: a dramatic outlier with poor quality
A map shows most values between 18 and 21 units. One pixel reads 35 units. The quality layer says that pixel was cloud-affected and based on one marginal observation.
The correct move is not to delete it automatically and not to build the whole story around it. Mark it as a weak-quality outlier and seek independent evidence or a later high-quality observation.
Worked reasoning case 2: the pattern survives quality filtering
A regional pattern appears in all data. You then inspect only the high-quality observations and the same broad pattern remains.
That strengthens confidence that the pattern is not created mainly by poor-quality pixels.
Worked reasoning case 3: quality differs between the two groups
Region A is mostly clear and has many high-quality observations. Region B is mostly cloudy and has far fewer usable observations. A chart shows Region A slightly higher than Region B.
The learner should not treat the two groups as if evidence quality were identical. More data or another measurement source may be needed before making a strong comparison.
Quality and missingness are connected but not identical
A low-quality value still exists. A missing value does not. A gap-filled value may replace a missing or poor-quality observation. That is why Reality Lab Vol No.039 and this article belong next to each other without being the same job.
Think of the evidence states as a chain:
good observation → lower-quality observation → missing observation → possible derived replacement
Different products handle these states differently, so always follow the documented source.
Tempting reasoning that fails
- “If a number exists, it is trustworthy.” A displayed value can carry a warning or lower-quality state.
- “Low quality means false.” It means the evidence is weaker or more limited, not automatically wrong.
- “Quality is the same as size.” A large number is not necessarily high quality.
- “All datasets use the same flags.” Flag meanings are product-specific.
- “The colourful map is the whole dataset.” Scientific products often contain metadata and QA layers that are not visible in the default picture.
PSLE-style transfer case
Four groups measure water temperature. Groups A, B and C use thermometers that were checked before the investigation. Group D notices that its thermometer has a cracked scale and is difficult to read, but still records 31°C.
Does Group D’s number become useless because the instrument condition is poor?
Not automatically. But the reading should carry less confidence until the instrument is checked or the measurement is repeated with suitable equipment. The evidence quality belongs in the reasoning, not just the final number.
What would strengthen a claim?
- high-quality flags across the key observations;
- several usable observations rather than one marginal result;
- agreement with another sensor or ground measurement;
- the same conclusion after weak-quality values are excluded;
- clear product documentation explaining the quality criteria.
What would weaken it?
- the conclusion depends on a few low-quality pixels;
- quality differs systematically between comparison groups;
- the public graphic hides known warnings that matter to the claim;
- flag meanings are assumed rather than checked;
- the difference being claimed is tiny compared with measurement limitations.
Practice
1. Pixel X = 0.51 with high quality. Pixel Y = 0.52 with low quality because of cloud. Which is definitely higher in reality?
Answer: The dataset reports Y as slightly higher, but the evidence does not justify “definitely” without considering measurement quality and the small difference.
2. A map legend explains colours but says nothing about data quality. What should you look for next?
Answer: Product documentation, metadata, QA layers, cloud masks, observation counts or other source information describing reliability.
3. Removing all low-quality pixels leaves the same broad pattern. What does that do?
Answer: It strengthens the case that the pattern is not mainly caused by low-quality observations.
Delayed independent return
The next time you see a scientific number, ask a second question before interpreting it: What does the source say about the quality of this number? If the answer is hidden in metadata, you have discovered why scientific evidence often has more than one layer.
Where to go next
- Reality Lab Vol No.039 — Were Some Values Filled In?
- Decide Whether Two Measurements Are Different Enough
- Handle an Anomalous Result Without Deleting It
- Use Healthy Scepticism Without Distrusting Every Result
Teaching guide for parents and tutors
Give the learner two columns labelled result and quality. Present four measurements with the same numerical values but different conditions: clear view, fogged lens, low battery, repeated reading. Ask the learner to rank confidence without changing the numbers.
The core repair is to separate “what value was recorded?” from “how strong is the evidence supporting that value?” This is a foundation for later work with uncertainty, reliability and scientific data quality without requiring advanced mathematics.
Authoritative sources
- SEAB — 2026 PSLE Science Syllabus
- MOE — 2023 Primary Science Teaching and Learning Syllabus
- NASA Earthdata — MODIS/Aqua Surface Reflectance Quality Daily product
- NASA Earthdata — Pixel quality filtering and MODIS QA information
The quiet return
Science is not only a collection of values. It is also a record of how those values were obtained, checked and limited. A learner becomes harder to fool—and better at real inquiry—when every number quietly brings a second question with it: how good is the evidence behind this?
