PSLE-SCI-REALITY-0039
Wait, What? A map with no blank spaces can contain places the satellite did not successfully observe.
A satellite map looks complete. Every square has a colour. Every date in the animation has a value. There are no holes, no blank pixels and no obvious warning signs.
It is natural to think: the satellite measured every one of those places on every one of those dates.
But clouds can block observations. Sensors can produce lower-quality data. Some scientific products are deliberately made gap-filled: missing or poor-quality values are replaced using a stated algorithm, often drawing on nearby times, neighbouring data, or a model. NASA publishes products that say this explicitly. For example, some MODIS vegetation products are described as gap-filled and smoothed because clouds and other problems create missing or low-quality observations that many modelling systems cannot use directly.
The map can still be useful. The scientific job is to stop treating every displayed value as the same kind of evidence.
Quick Answer
When a scientific map has no gaps, ask whether every value was directly observed or whether some were filled in. If values were gap-filled, identify which data were original observations and which were estimates, why gaps occurred, what algorithm was used, and whether your conclusion depends heavily on the filled values. A seamless map can be scientifically valuable without every cell being an original measurement.
Reality Lab rule: No blank space does not mean no missing observation.
The learner job this page owns
This page owns one applied question: how should a Primary 5/6 learner evaluate a scientific map or dataset when missing observations have been filled in? It does not replace eduKate’s existing owners for missing values, interpolation, smoothing or model limits. It applies those skills to a real scientific communication object: a display that looks complete because gaps have been repaired by a processing method.
Original case: the green-island vegetation map
Imagine a fictional satellite vegetation map of Green Island. The map reports a vegetation index every eight days. For one month the values at one location are shown as:
| Date | Displayed value | Data status |
|---|---|---|
| 1 June | 0.61 | Good satellite observation |
| 9 June | 0.63 | Cloud-covered; filled value |
| 17 June | 0.64 | Cloud-covered; filled value |
| 25 June | 0.66 | Good satellite observation |
The public chart draws one smooth green line through all four values. A student writes: “The satellite observed the vegetation increasing steadily every eight days.”
That overstates the evidence. The satellite directly supplied good observations on 1 June and 25 June. The middle values were estimated by a gap-filling process because clouds prevented suitable observations. The better statement is:
“The processed dataset shows an increasing pattern across June, but the 9 June and 17 June values were filled estimates rather than good direct observations.”
Four evidence states that should not be collapsed
| Evidence state | What it means |
|---|---|
| Good observation | The instrument produced a usable measurement under the product’s quality rules. |
| Poor-quality observation | A measurement exists but clouds, geometry, contamination or another issue reduces confidence. |
| Missing observation | No usable value is available for that place and time. |
| Gap-filled value | A processing method estimates a replacement so the dataset can remain continuous. |
The displayed number alone may not tell you which state you are looking at. That information can live in metadata, a quality layer or the product documentation.
Why scientists fill gaps
Some analyses need complete time series or maps. If a model requires one value for every eight-day period, leaving cloud-covered dates blank may make later processing difficult. Gap filling can therefore be a sensible engineering and scientific choice.
The important point is not that gap filling is suspicious. The important point is that filled values and directly observed values have different evidence histories.
Gap filling is not the same as smoothing
Reality Lab Vol No.031 asks what averaging or smoothing can hide in a trend. Gap filling is a different job. Smoothing changes how a sequence of existing values is represented. Gap filling creates an estimated value where a usable observation is missing.
A product can do both. NASA’s descriptions of some MODIS vegetation products explicitly say they use one stage for smoothing and another for gap filling. A learner should therefore ask two separate questions:
- Were existing observations smoothed?
- Were missing or poor-quality observations replaced?
What did the algorithm know?
A gap-filling method might use nearby dates, nearby locations, seasonal patterns, another sensor, a physical model or a combination of inputs. Each method carries assumptions.
Imagine a simple method that fills a missing middle value halfway between two surrounding observations. If 1 June is 0.60 and 17 June is 0.70, it may estimate 9 June as 0.65. That is reasonable only if a smooth change is plausible. A storm, fire, sudden harvest or flood between the observation dates could make the real path very different.
The estimate may still be the best available value. But the algorithm did not travel back in time and observe the missing day.
Source and provenance check
Before using a seamless map as evidence, look for its product name and processing level. Ask:
- Which satellite or instrument supplied the underlying observations?
- What conditions create missing or poor-quality data?
- Does the product include quality flags?
- Which values are filled?
- How are filled values calculated?
- Does the product distinguish original and filled observations?
- What question is the gap-filled product designed to answer?
Representation check: the picture can hide the evidence state
A visualisation may colour a directly observed value and a filled value exactly the same way. That makes sense if the goal is to show a continuous field. But it can also hide the difference between measured and estimated evidence unless the legend or metadata explains it.
This is why scientific reading sometimes requires two layers at once: the value layer and the quality/provenance layer.
Worked reasoning case 1: one filled day
A 30-day dataset has 29 good observations and one filled value. The broad monthly conclusion is supported by many surrounding measurements and hardly changes if the filled value is slightly different.
Here, the gap matters, but it may not dominate the conclusion.
Worked reasoning case 2: the key event occurs inside the gap
A scientist wants to know the exact day vegetation dropped sharply after a storm. Unfortunately, the satellite had clouds for the four days around the storm and those dates were gap-filled.
Now the missingness is central. A gap-filled trend may suggest a gradual transition even if the true change was sudden. The product may still support the broad before-versus-after difference while being weak evidence for the exact timing.
Worked reasoning case 3: a whole region is cloudy
If one small patch is missing, nearby spatial information may help. If an entire region is cloud-covered for a long period, the algorithm has less direct local evidence available. Confidence in detailed claims should fall accordingly.
Tempting reasoning that fails
- “Every coloured cell was observed.” Some may be gap-filled.
- “Filled data are fake.” They are estimates produced for a stated purpose and can be scientifically useful.
- “If the algorithm is sophisticated, the filled value becomes an observation.” It remains a derived estimate.
- “A complete-looking map contains complete direct evidence.” Visual completeness and observational completeness are different.
- “One filled point invalidates the whole dataset.” The effect depends on how much is filled and whether the claim relies on it.
How far can the conclusion travel?
A gap-filled product may be excellent for broad seasonal patterns but unsuitable for proving an exact local event at a missing time. It may be useful for modelling but weaker for a claim that requires direct observation. Scientific conclusions should match the information pathway that produced the data.
What evidence would strengthen the claim?
- a clear mask showing which values were observed and which were filled;
- a documented gap-filling method;
- nearby independent observations;
- agreement with another sensor or ground measurement;
- few filled values relative to the full record;
- a conclusion that remains similar when filled values are removed or changed within plausible limits.
PSLE-style transfer case
A student records plant height on Monday, Wednesday and Friday. Tuesday and Thursday are missing. A spreadsheet automatically fills Tuesday and Thursday by drawing a straight line between the measured days.
Can the student write, “The plant was measured at 12.4 cm on Tuesday”?
No. The Tuesday value is an estimate derived from the surrounding measurements. It may be useful for a simple graph, but it must not be reported as a direct observation. This connects to How to Tell When a Missing Value in a PSLE Science Table Can Be Inferred—and When It Must Stay Unknown.
Practice
1. A satellite dataset contains 100 dates. Five dates were gap-filled because of clouds. What should a careful learner say?
Answer: Most dates are based on usable observations, while five dates contain estimated replacement values. The importance depends on whether the conclusion relies on those dates.
2. A gap-filled value sits exactly at the date when a sudden flood occurred. Why does that matter?
Answer: A smooth estimate may not capture an abrupt real change, so the dataset is weaker evidence for the exact timing or size of the flood-related change.
3. A map has no blanks. What is the next question?
Answer: Ask whether all values are direct observations or whether some were filled, modelled or otherwise derived.
Delayed independent return
The next time you see a perfectly complete scientific map, imagine temporarily removing every value that was not directly observed. Would the picture suddenly contain holes? If so, ask whether the published visualisation tells you how those holes were repaired. That habit turns a pretty surface back into an evidence chain.
Where to go next
- Read Data With Gaps Without Inventing What Happened
- Tell When a Missing Value Can Be Inferred
- Reality Lab Vol No.029 — What Happened to the Missing Results?
- Use a Scientific Model Without Mistaking It for Reality
Teaching guide for parents and tutors
Use four sticky notes: observed, poor quality, missing and filled. Give the learner a short sequence with one example of each. Ask them to build a graph, then ask which points they would describe as observations and which as estimates.
The important repair is not “never use filled data”. It is “never lose provenance”. A child who can say where a value came from can reason more carefully about how much weight it deserves.
Authoritative sources
- SEAB — 2026 PSLE Science Syllabus
- MOE — 2023 Primary Science Teaching and Learning Syllabus
- NASA Open Data — MODIS/Terra Gap-Filled, Smoothed NDVI 8-Day product
- NASA Open Data — MODIS/Aqua Gap-Filled Evapotranspiration product
The quiet return
Science often repairs incomplete observations so larger questions can still be studied. The repair becomes trustworthy when it is visible, documented and kept distinct from the original measurement. A seamless map is not a promise that nature was seen without interruption. It is a prompt to ask how the gaps disappeared.
