PSLE Science Reality Lab Vol No.395 | “30 m Satellite Data Resampled to 10 m” — Did It Gain New 10 m Detail?

PSLE-SCI-REALITY-0395

Wait, What? The Picture Has More Pixels Now

Alicia opens two maps of the same patch of land. The first file is labelled 30 m satellite data. The second looks smoother, has smaller squares and is labelled resampled to 10 m.

“So the second map must contain three times more detail,” she says. “Each pixel is only 10 m wide now.”

That conclusion feels natural. Smaller squares look more detailed. A bigger image file can look sharper on a screen. A program can even produce many more pixel values than were present in the original grid.

But this is exactly where scientific evidence needs careful handling. A new grid can contain more cells without containing new observations of the ground. Resampling can change how existing information is represented. It does not send the satellite back in time to measure extra places.

Quick Answer

No. If 30 m source data are resampled onto a 10 m grid, the output grid has smaller cells, but that alone does not create new native 10 m measurements. Depending on the resampling method, the new cell values may be copied from nearby source cells or calculated from neighbouring source values. The representation has changed; the original observing power has not.

The habit to remember is simple: ask what was measured before asking what is displayed now.

The Exact Learner Job

This Reality Lab owns one narrow transfer job: evaluating a scientific raster, satellite map or gridded image whose output pixel size is smaller than the native source resolution.

It does not replace the general PSLE Science owners for observation versus inference, units and scale, graph reading, measurement resolution or image processing. Those skills remain separate. Here, we apply them to a real communication object that can easily make a processed image look more informative than the measurements that created it.

Build the Evidence Object From Scratch

Imagine an original satellite product covering a square field. The source data use 30 m by 30 m cells. For a simple teaching example, suppose four neighbouring source cells contain the values 40, 50, 60 and 70. Those numbers might represent a measured or derived property such as reflectance, temperature or another mapped quantity.

A mapping program is then asked to create a new grid whose cells are 10 m by 10 m. The program must decide what values to put into all those new smaller cells. It has several possible rules.

Nearest neighbour: give a new cell the value of the nearest source cell.
Bilinear interpolation: calculate a new value from nearby source cells, with closer cells contributing more.
Cubic or other interpolation: calculate values using a larger neighbourhood and a more complex mathematical rule.

All three can be useful. None of them creates a forgotten 10 m satellite measurement that was hiding inside the original 30 m cell.

Observed, Represented, Claimed, Inferred

A strong learner separates four layers.

Observed: what the sensor actually collected, at its stated native sampling or spatial resolution and under its acquisition conditions.
Represented: how software places those source values onto a new grid.
Claimed: what the caption, advertisement, report or social-media post says the new image shows.
Inferred: what we conclude about the real world from that processed display.

The scientific mistake occurs when the represented layer quietly takes over the observed layer. “The file now has 10 m pixels” is a statement about the output grid. “The satellite measured independent 10 m details” is a statement about the observation. They are not automatically the same statement.

Why More Pixels Can Appear Without More Evidence

Take one source cell that covers a 30 m by 30 m area. A 10 m output grid can fit nine 10 m by 10 m cells into the same area. If nearest-neighbour resampling is used and the grids align neatly, several of those smaller output cells may simply receive the same source value.

The file now contains more cells. Yet those nine cells did not come from nine independent measurements. They came from one source measurement being represented on a finer grid.

Interpolation can make the pattern look smoother. Suppose one source cell is 40 and a neighbouring source cell is 60. A program might calculate intermediate output values such as 45, 50 or 55 between them. Those numbers can be useful estimates for display, geometric matching or analysis. But they are calculated from the existing observations. They are not new direct measurements of previously unseen ground detail.

A Smaller Grid Cell Is Not Automatically a Smaller Resolvable Object

This distinction is easy to miss because a computer image has at least two different ideas hiding behind the word resolution. One is the size of the cells in the file. Another is the spatial detail that the observing system can genuinely distinguish.

For this Reality Lab, do not try to memorise a universal formula. Instead, ask two questions:

What was the native source information?
What processing changed the grid after collection?

That protects you from a common communication trick: presenting a fine output grid as though every fine cell were independently observed at that scale.

Worked Case 1: The “10 m Heat Map”

A student is shown an environmental heat map. The caption says: “Original thermal raster: 30 m. Product resampled to 10 m for comparison with a land-use layer.” The map contains very small coloured squares.

A tempting answer is: “The temperature was measured every 10 m because each square is 10 m wide.”

That answer goes too far. The caption itself says the source was 30 m and the product was later resampled. The 10 m grid helps the datasets line up, but it does not prove independent temperature measurements existed at every 10 m location.

A stronger answer is: “The displayed product uses 10 m grid cells, but the source temperature information came from the 30 m raster. The 10 m values depend on the resampling method and should not be treated as new 10 m measurements.”

Worked Case 2: The Sharp Boundary That Was Not Measured

Kai Kai compares two neighbouring source cells. The western cell has value 20 and the eastern cell has value 80. After resampling, the software shows several intermediate colours near the boundary.

Can she say that the real landscape definitely passed through the exact sequence 20, 30, 40, 50, 60, 70, 80 across those locations?

No. The smooth transition may partly reflect the interpolation rule. The real landscape could change gradually, abruptly, irregularly or in some other way. The original source observations do not uniquely reveal every fine-scale value between the cells.

This does not make the resampled map “fake”. It makes it a processed representation whose meaning must be bounded by its source data.

Worked Case 3: Two Datasets Must Share One Grid

Tricia has a vegetation map at 30 m and an elevation map at 10 m. A researcher wants to compare values cell by cell. One sensible processing step is to place both datasets onto a common grid.

Resampling can be necessary here. Without it, the cells may not line up spatially. But the reason for creating a common 10 m grid is compatibility, not magical improvement in the original vegetation sensor.

The correct evidence statement is therefore: “The vegetation values have been represented on the 10 m analysis grid so they can be compared with the elevation layer. Their underlying source detail remains limited by the original observations and processing.”

Representation Check: What Changed?

When you meet a resampled image, compare these features before and after processing:

pixel or cell size;
number of rows and columns;
grid alignment;
map projection;
resampling method;
native source resolution;
whether the values were copied, interpolated or otherwise transformed.

The important idea is not to reject processing. Scientific data often require processing. The important idea is to know which properties belong to the source observation and which belong to the processed product.

Comparison and Baseline Check

Suppose an advertisement says, “Our upgraded map now has 10 m pixels instead of 30 m pixels, giving three times the spatial detail.” What baseline would you need before accepting that?

You would need to know whether the underlying observations also changed. Was a genuinely higher-resolution sensor used? Was a second higher-resolution measurement source fused in? Or was the old 30 m product simply resampled to a 10 m grid?

The same output cell size can arise from very different evidence histories. Therefore, cell size alone is not enough to judge the strength of the observation.

Method and Variable Check

The method matters because different resampling choices answer different practical needs.

Nearest-neighbour resampling tends to preserve original source values but can create blocky-looking output. Interpolation methods can create smoother images by calculating new values from neighbours. That may be useful for continuous surfaces, but it also means some displayed values are mathematical estimates rather than untouched source numbers.

A Primary 5 or 6 learner does not need to become a remote-sensing engineer. The PSLE-level transfer is simpler: if a method calculates or redistributes values, keep the method visible in your conclusion.

Alternative Explanations for a “Sharper” Image

If a processed map looks sharper, several explanations are possible:

the source sensor really had finer native observations;
the image was resampled to a smaller grid;
contrast or sharpening was changed;
higher-resolution information from another band or sensor was fused into the product;
the display software enlarged the image without adding scientific information.

A careful learner does not pick one explanation from appearance alone. Check the metadata or method description.

What Evidence Would Strengthen the Claim of Real 10 m Detail?

A sensor specification showing that relevant source observations were natively collected at about that spatial scale.
A processing description showing that genuinely finer independent measurements were fused into the product.
Validation against higher-resolution reference data showing that fine features are recovered reliably.
Documentation distinguishing output grid spacing from native spatial resolution.
Examples in which fine features seen in the product are confirmed by independent observations.

What Evidence Would Weaken the Claim?

The source is explicitly described as 30 m but the 10 m product is only called “resampled”.
The method states that smaller cells were generated by nearest-neighbour copying or interpolation.
Fine features change noticeably when a different resampling method is used.
The source documentation warns that the cell size should not be interpreted as independent spatial detail.
The claim relies only on how sharp the image looks on a screen.

How Far Can the Conclusion Travel?

If a resampled map is useful for locating a broad hot region, following a large vegetation pattern or aligning several datasets, it may still be excellent evidence for those jobs. The mistake is to make the conclusion travel farther than the observations support.

For example, a 30 m source resampled to 10 m might help a program overlay roads, administrative boundaries and other layers. It does not automatically justify a claim about a tiny 10 m feature that was never independently resolved by the source.

The Tempting but Invalid Reasoning

“There are more pixels, so there must be more measured information.”

The flaw is that file structure and measurement history are being treated as the same thing. A computer can generate additional cells from existing values. The number of stored numbers can increase without the number of independent observations increasing.

“The map looks smoother, so it must be more accurate.”

Smoothness is a display property, not a universal accuracy test. Interpolation can make a surface look smooth even where the true fine-scale pattern is unknown.

“A 10 m output grid means every object larger than 10 m can be reliably detected.”

That is too strong. Detectability depends on the observing system, contrast, processing, object shape, sensor characteristics and other conditions. Output grid spacing by itself cannot guarantee it.

A PSLE-Style Transfer Case

A science club studies a satellite map of a pond. The original dataset contains 20 m pixels. Software resamples it to 5 m pixels before the map is printed. A student says, “Now the map can show four times smaller objects.”

Question: Explain why the student’s conclusion is not supported by the information given.

Reasoned answer: Resampling changes how the original data are placed on the output grid. The new 5 m cells may contain copied or calculated values from the 20 m source data. Since no new measurements are stated to have been collected, the smaller output cells do not by themselves prove that four-times-smaller objects can be resolved.

Notice what this answer does not do. It does not say the processed map is useless. It identifies exactly which extra claim lacks evidence.

Delayed Return: Try This Without Looking Back

Ten minutes later, answer these from memory.

A 100 m climate raster is resampled to 25 m. What changed for certain, and what did not necessarily change?
Why can bilinear interpolation create numbers that were not present in the source file without creating new direct observations?
What extra evidence would you need before claiming that the processed product really contains finer independent detail?

Check: the output grid changed; native observing detail did not necessarily change. Interpolation calculates values from existing neighbours. A real finer-detail claim needs evidence about the source sensor, additional measurements, fusion or validation.

Explained Practice Set

Practice 1. A researcher converts a 1 km rainfall grid to a 250 m grid so it can be used in another model. The new map contains 16 times as many cells per square kilometre. Does that prove 16 independent rainfall measurements existed there?

Answer: No. The increase in output cells comes from the resampling step. The source observation density and measurement method must be checked separately.

Practice 2. A map is resampled with nearest neighbour. Four adjacent output cells have the same value. Is that evidence that four instruments independently recorded exactly the same result?

Answer: No. Multiple output cells can inherit the same source value during resampling.

Practice 3. A 30 m multispectral image is fused with a genuinely measured 15 m panchromatic band. Can you still say “nothing new was added”?

Answer: Not safely. A second source with finer measured spatial detail has now entered the product. You must inspect what that source contributes and what it does not contribute. This is the distinction explored in the neighbouring Reality Lab volume on pan-sharpening.

Route to the Existing PSLE Science Owners

Use this Reality Lab for the real-world resampling claim. For the underlying skills, continue through the existing eduKate Sengkang owners:

Parent and Tutor Teaching Guide

Do not begin by teaching vocabulary such as “bilinear” or “nearest neighbour”. Begin with a concrete evidence question: Where did each new number come from?

Draw one large square on paper and write 40 inside it. Then divide the square into nine smaller squares. Ask the learner whether drawing extra boundaries created nine new thermometer readings. The answer is obviously no. Next, write 40 in all nine cells. That models a simple copy-like case. Then place 40 and 60 in two large neighbouring cells and ask the learner to imagine software calculating 45, 50 and 55 between them. Those values can be useful estimates, but the act of calculation is visible.

Only after the learner understands that distinction should you introduce the word resampling. This keeps the science job ahead of the terminology.

A useful three-question routine is:

Source: What was originally observed?
Processing: What did the software do?
Claim: Does the conclusion stay within what the source plus processing can support?

Stop there. Do not turn this lesson into a remote-sensing degree course. The learner’s transferable gain is knowing that a polished representation can have finer formatting than the underlying evidence.

Authoritative Sources

The Quiet Habit to Keep

A scientific image can be enlarged, smoothed, regridded and beautifully displayed. None of those steps is automatically dishonest. The discipline is simply to keep the chain visible.

What was measured? What was calculated? What was displayed? What can I now conclude?

When those four questions stay separate, a smaller pixel cannot quietly pretend to be a new observation.

Small Group Tutorials