PSLE Science Reality Lab Vol No.387 | “Satellite Data: Level 3” — Does Level 3 Mean It Is More Accurate Than Level 2?

PSLE-SCI-REALITY-0387

Wait, what? A science website offers two files from the same satellite mission. One is labelled Level 2. The other is labelled Level 3. A learner points to the larger number and says, “Level 3 must be more accurate. It is one level higher.”

That sounds natural because many everyday systems use levels as ranks: Level 3 can mean harder, stronger or more advanced than Level 2. Satellite and Earth-observation data use the word differently. In NASA Earthdata metadata, a processing level identifies how far the source observations have been processed and organised. It is not automatically an accuracy score, a school grade, a reliability badge or a promise that every Level 3 value is closer to truth than every Level 2 value.

This Reality Lab owns one precise evidence-transfer job: how to evaluate a scientific dataset label such as “Level 3” without turning a processing stage into a quality ranking. The lesson applies PSLE Science inquiry to a real communication object: read what was measured, what transformations were applied, what the product represents, what separate quality information exists, and how far a comparison can travel.

Quick Answer

No. “Level 3” does not by itself mean “more accurate than Level 2”. NASA Earthdata describes processing levels as identifiers for the degree to which source data have been processed. A Level 2 product commonly contains derived geophysical variables at roughly the original observation support, while a Level 3 product commonly places variables onto uniform spatial and/or temporal grids. That extra processing can make a dataset easier to compare across space and time. It can also involve averaging, compositing, interpolation, regridding, filtering or other operations whose effects must be understood.

Accuracy, uncertainty, quality flags and validation maturity are different questions. NASA Earthdata even keeps data maturity as a separate metadata idea, with terms such as provisional or validated. That separation is a useful scientific clue: if processing level already meant quality, a separate maturity field would be unnecessary.

The Exact Learner Job

Owned here: identify what a processing-level label says about the route from observations to product, then refuse to infer an unstated accuracy ranking from the level number.

Not owned here: satellite physics, remote-sensing algorithms, interpolation mathematics, general map reading, spatial resolution, uncertainty theory or model comparison. Those concepts already have owners elsewhere in the eduKateSengkang Science estate. This page applies them to one metadata object that students increasingly meet in scientific maps, environmental dashboards and data portals.

Rebuild the Real-World Object

Imagine a fictional satellite mission called GreenView. Its sensor measures reflected light over Earth. The mission releases several products:

Product	What the file contains	What the label is mainly telling you
Level 1	Calibrated, located sensor observations	The observations have been corrected and organised enough to be physically interpretable
Level 2	A derived vegetation quantity at observation locations	An algorithm has converted sensor observations into a geophysical variable
Level 3	Daily vegetation values on a regular 1 km grid	Level 2-like information has been mapped, combined or summarised onto a standard space-time grid

These are original teaching examples, not copied mission products. The important move is to ask what operation created each product. “3” tells us where the product sits in a processing chain. It does not tell us, on its own, how large its error is.

Observed, Processed, Claimed and Inferred

A strong reader separates four layers:

• Observed: the satellite instrument detected signals at particular times and locations.
• Processed: calibration, retrieval, mapping, filtering, averaging or gridding turned those observations into a usable product.
• Claimed by the metadata: the dataset belongs to a stated processing level and follows that product’s definition.
• Extra inference: “Because Level 3 has a bigger number, it must be more accurate.”

The extra inference needs independent evidence. A label can describe what was done without proving how well every result agrees with reality.

Why More Processing Is Not the Same as More Truth

Processing can improve usefulness. Calibration removes known instrument effects. Geolocation places measurements on Earth. Retrieval algorithms convert signals into physical quantities. Gridding helps compare places. Averaging can reduce random noise. Cloud screening can remove observations known to be unreliable for a particular job.

But every processing step also has assumptions and boundaries. A grid cell may combine several observations. Missing areas may require rules about whether to leave gaps, interpolate or use a model. A daily composite may select the “best” observation from several times. An average can hide short-lived extremes. Reprojection can shift how values are represented near boundaries.

So the scientific question is not “Was it processed more?” but “What processing was done, for what purpose, and what evidence shows the resulting product is fit for this question?”

The Ladder That Is Not a Scoreboard

Think of a bakery rather than a game. Flour, dough and bread are later stages of processing. Bread is not “more accurate” than flour. It is a different product made for a different use. Likewise, a Level 3 gridded environmental product can be more convenient for regional comparison than a Level 2 swath product, while the Level 2 data can preserve observation-scale information that the Level 3 grid no longer shows.

A later processing level may be more suitable for one learner question and less suitable for another. Suitability is not a simple staircase.

Worked Case 1: The Smooth Map

A Level 2 file contains irregular satellite observations with some cloud gaps. A Level 3 map shows a neat daily grid with one value in every valid cell. The Level 3 map looks smoother and more complete, so a student says it must be more accurate.

The neat appearance is a representation property, not accuracy evidence. The Level 3 product may have combined observations within each cell, chosen a representative value or applied screening rules. It can be excellent for comparing broad spatial patterns. But the smooth grid does not prove that each cell is closer to an independent reference measurement than each Level 2 observation.

Better conclusion: “The Level 3 product is more processed and organised for gridded comparison. Its accuracy must be judged using its validation and quality information, not the level number or smooth appearance.”

Worked Case 2: The Average Hides an Extreme

Suppose a Level 2 temperature product contains four good observations for one grid region: 28°C, 29°C, 35°C and 30°C. A Level 3 daily product reports a summary value of 30.5°C.

If the question is “What was the daily regional average represented by this product?”, the Level 3 summary may be useful. If the question is “Did the region ever contain an observation near 35°C?”, the Level 3 average cannot answer by itself. The later processing has not made the 35°C observation false; it has changed the object being represented.

That is an important PSLE Science habit: a summary can be well made and still answer a different question from the individual measurements used to build it.

Worked Case 3: Fewer Gaps, More Assumptions

A cloud-covered region has missing observations. Product A leaves the cells blank. Product B fills some gaps using a stated compositing or model-assisted method. Product B therefore looks more complete.

Does more complete mean more certain? Not necessarily. Filling gaps can be scientifically useful, but the filled values need provenance: were they measured at that exact place and time, interpolated from neighbours, carried from another time, or estimated by a model? The communication should tell the reader which kind of evidence is present.

Worked Case 4: Level 2 Can Beat Level 3 for a Point Event

A learner is investigating a short-lived smoke plume that crossed one small area for twenty minutes. The Level 3 daily grid averages information over a larger cell and longer period. A suitable Level 2 observation happened to pass directly over the plume.

For the narrow question “Was a strong plume visible during that satellite pass?”, the Level 2 observation may retain more relevant detail. For “What was the broad daily pattern across the region?”, Level 3 may be the better tool. Neither label wins all questions.

Worked Case 5: Same Level, Different Quality

Two different missions both publish Level 3 vegetation products. One has extensive validation over the type of forest being studied; the other is new and has limited validation in that environment. A student says the products must have the same reliability because both say Level 3.

Processing level is not enough to support that conclusion. Different sensors, algorithms, calibration histories, clouds, viewing angles, grids and validation evidence can produce different uncertainties. Compare the actual product documentation and quality evidence.

Worked Case 6: Higher Level, Different Variable

A Level 2 product reports aerosol optical depth. A Level 3 product reports a monthly gridded aerosol optical-depth mean. A learner tries to compare the Level 3 number directly with a ground PM2.5 concentration and says the disagreement proves Level 3 is inaccurate.

There are two mismatches: the satellite product and ground instrument measure or represent different quantities, and the time-space supports differ. Before judging accuracy, make sure the objects being compared are commensurate.

Processing Level Versus Data Maturity

NASA Earthdata treats processing level and data maturity as separate metadata. That is a powerful real-world lesson. A dataset can be highly processed yet still be provisional. Another product can be validated while belonging to a lower processing level because the level describes form of processing, not stage of scientific confidence.

So when you see words such as Beta, Provisional or Validated, do not merge them with Level 1, 2, 3 or 4. Ask two separate questions:

1. How was the source data processed into this product?
2. What evidence exists about the product’s maturity, validation and known limitations?

Representation Check

A Level 3 map can look authoritative because it is rectangular, complete and colour-coded. Visual neatness can make uncertainty invisible. Check:

• What does one pixel or grid cell represent?
• Is the value instantaneous, daily, monthly or another summary?
• Were multiple observations combined?
• Are missing cells genuinely missing, screened out or estimated?
• Does the legend show a physical variable or a quality category?
• Is there a separate uncertainty or quality layer?

A map is a representation of a data product. It is not a photograph of every quantity on Earth.

Comparison and Baseline Check

If someone claims “Level 3 is better than Level 2”, ask better for what? A valid comparison needs a criterion. Possible criteria include:

• easier regional comparison;
• preservation of original observation detail;
• fewer gaps;
• lower random noise;
• smaller spatial support;
• closer agreement with independent reference observations;
• faster availability;
• better fit to a particular research question.

Different criteria can point to different products. “Higher level” is not itself a comparison method.

Method and Variable Check

Before using a processing-level label as evidence, identify:

1. Variable: What physical quantity is represented?
2. Source: What sensor or observations feed the product?
3. Processing: What transformations turn source data into the product?
4. Space: What area does a value represent?
5. Time: What time or time window does it represent?
6. Quality control: What observations are filtered or flagged?
7. Validation: What independent evidence checks the product?
8. Uncertainty: What limits are stated for the intended use?

Alternative Explanations for a Difference Between Levels

Suppose a Level 2 observation and a Level 3 grid value differ. Do not jump immediately to “one is wrong”. Differences may arise because:

• the Level 3 cell combines several Level 2 observations;
• the products represent different time windows;
• quality screening removed some source observations;
• the grid value is an average, maximum, median or other summary;
• the source observation falls near a grid boundary;
• the Level 3 product includes modelled or composited information;
• the units or variable definitions differ;
• one value is flagged as lower quality.

A scientific explanation starts by restoring the product definitions.

Evidence That Strengthens a Claim About Product Quality

• clear product documentation explaining each processing step;
• quality flags that can be traced to individual values or cells;
• comparison with independent reference measurements;
• validation results across relevant environments and seasons;
• reported uncertainty or accuracy metrics matched to the variable;
• transparent treatment of missing data and clouds;
• version history showing how known problems were corrected;
• agreement across independent products where the comparison is scientifically valid.

Evidence That Weakens an Overclaim

• “higher level” is the only reason given for “more accurate”;
• the map looks smoother, but no validation evidence is shown;
• a later processing level is treated as a newer software version or stronger quality badge;
• products with different variables or time windows are compared as if identical;
• a provisional product is described as fully validated because it is Level 3;
• quality flags are removed from a screenshot;
• missing or model-filled cells are presented as direct observations;
• the conclusion extends beyond the documented purpose of the product.

How Far Can the Conclusion Travel?

From “this is a Level 3 product”, you can reasonably conclude that the data provider has placed the product at a defined stage in its processing system, commonly involving mapping or summarising derived variables onto regular space-time grids in EOSDIS-style conventions. You cannot conclude from that label alone that it has smaller error than Level 2, better validation, newer data, finer resolution, fewer assumptions or stronger suitability for every question.

A strong conclusion might be: “Level 3 is the gridded product we need for this regional monthly comparison. We still need its quality flags and validation information before making an accuracy claim.”

Tempting but Invalid Reasoning

• “Level 3 is one level higher, so it is one step more accurate.” Processing stage and accuracy are different properties.
• “The Level 3 map has no gaps, so every cell was measured.” Check the product method; some values may be combined or estimated.
• “Level 2 is raw data.” Level 2 is already a derived product in common Earth-observation processing systems.
• “The smoothest map is the best map.” Smoothness can result from averaging or larger support.
• “Validated means Level 4.” Data maturity and processing level are separate metadata dimensions.
• “Level 3 is always better for students.” Suitability depends on the learner question.

A PSLE-Style Transfer Case

A school science team uses a sound sensor. File A stores one reading every second. File B stores one-minute averages calculated from File A. File C places the one-minute averages into hourly summaries.

A learner says File C must be the most accurate because it has gone through the most processing. Evaluate the reasoning.

Model reasoning: more processing changes the representation and can make some patterns easier to see, but it does not automatically increase accuracy. The hourly summary may be useful for comparing long periods but can hide brief loud peaks. Accuracy must be checked against suitable reference evidence and the measurement method, not the number of processing steps.

Delayed Independent Return

1. What does a satellite-data processing level primarily describe?
2. Why can a Level 3 product be more useful without being automatically more accurate?
3. What separate metadata would you seek before making a confidence claim?
4. Why might a Level 2 product be better for a short-lived local event?
5. If Level 2 and Level 3 values differ, name two explanations other than “one is wrong”.
6. What is the danger of judging a product from how smooth its map looks?

Explained Answers

1. The degree and form of processing applied to the source data. 2. Processing can make data gridded, averaged or easier to compare while changing what each value represents; usefulness and accuracy are separate. 3. Quality flags, validation results, uncertainty information and data maturity. 4. It may preserve observation-scale detail that is averaged away in a gridded summary. 5. Different time windows, spatial support, averaging, filtering, compositing or quality screening. 6. Visual smoothness can come from processing and does not by itself show closer agreement with reality.

Route the Core Skills to Their Owners

For the meaning of satellite pixel size and spatial support, use PSLE Science Reality Lab Vol No.062 | “30 m Resolution” — Does One Pixel Describe a Single Point or a Whole Patch?. For the difference between direct observations and a gridded model-assisted historical product, use PSLE Science Reality Lab Vol No.159 | “This Is a Historical Weather Map” — Was Every Grid Cell Directly Measured?. For choosing between models or representations according to a question, use How to Compare Two Scientific Models in PSLE Science and Decide Which One Is More Useful.

Parent and Tutor Teaching Guide

Do not begin by asking a child to memorise a hierarchy. Give three cards: observations, derived values and gridded summary. Ask the learner to arrange how one could be transformed into the next. Then ask a different question for each card: “Which preserves the individual pass?”, “Which is easiest for regional comparison?”, “Which might hide a short extreme?” This makes processing purpose visible.

Next, place a second set of cards beside the first: accuracy, validation, uncertainty, maturity. Ask whether any of these can be inferred merely from the processing-level number. The correct answer is no; each needs its own evidence.

Finally, show a neat fictional Level 3 colour map beside a messy Level 2 swath. Ask the child which one “looks better”, then make them replace that vague word with a criterion: better spatial continuity, better local detail, better time match, or better validated accuracy. Once “better” has a job, the reasoning improves.

Authoritative Sources

• NASA Earthdata Common Metadata Repository — Processing Level, describing processing level as the degree to which source data have been processed and distinguishing EOSDIS levels from raw observations through gridded and model-derived products.
• NASA Earthdata Common Metadata Repository — Data Maturity, showing that maturity and validation status are separate metadata from processing level.
• Singapore Examinations and Assessment Board — 2026 PSLE Science syllabus, including interpretation, analysis, evaluation of information and methods, and communication of reasoning.
• Ministry of Education Singapore — 2023 Primary Science Teaching & Learning Syllabus, including healthy scepticism, objectivity, evidence-based explanation and attention to how Science is communicated.

Quiet Return

A level number can look like a score even when it is not one. When a scientific dataset says Level 3, first ask what happened to the source observations on the way to that product. Then ask the separate quality questions: how was it checked, what uncertainty remains, what does one value represent, and is it fit for the question in front of you? The habit is simple: read the processing label as a route, not a medal.