PSLE Science Reality Lab Vol No.520 | “Weather Model Grid Spacing = 1 km” — Is the Forecast Accurate to Within 1 km?

PSLE Science Reality Lab Vol No.520

Wait, What? A 1 km Weather Grid Is Not a 1 km Accuracy Promise

A weather-model page says horizontal grid spacing: 1 km. A student points at a forecast map and says, “Great. If the storm is shown here, it should really be within 1 km of this spot.” The number looks like a location tolerance. It is not.

Grid spacing describes the computational mesh on which a numerical weather model represents and calculates the atmosphere. It tells us something important about the model’s spatial framework, but it does not by itself tell us how close a predicted rain band, wind maximum or thunderstorm will be to its real location.

This Reality Lab owns one learner job: when a weather forecast or model description gives a grid spacing such as 1 km, separate the size of the computational mesh from a guarantee of forecast location accuracy.

Quick Answer

No. “1 km grid spacing” means neighbouring model grid points or cells are separated by about that spatial scale in the horizontal model mesh. It does not mean every weather feature is represented perfectly at 1 km scale, and it does not mean a forecast feature will be located within 1 km of where it later occurs.

Forecast accuracy also depends on the quality of the starting atmospheric state, observations used in data assimilation, model equations and approximations, how clouds and turbulence are represented, terrain and surface information, forecast lead time, chaotic growth of small errors, and the weather feature being predicted.

The useful learner sentence is: grid spacing tells me how the model divides space; forecast accuracy must be checked against observations.

Owned Learner Job — Not a New Owner of Weather Models

This page does not try to teach numerical weather prediction as a complete topic. It does not own atmospheric equations, cloud physics, generic model limitations, graph reading, variables or probability. It applies those existing skills to one communication object: a model page, app, infographic or news sentence that turns grid spacing into an unsupported claim about forecast accuracy.

The Evidence Object: Two Forecast Cards

Imagine two fictional forecast-system cards:

Property	Model A	Model B
Horizontal grid spacing	1 km	3 km
Forecast lead time	30 hours	6 hours
Starting observations	Older analysis cycle	Newer analysis cycle
Rain band location error in one verified case	18 km	5 km

A student who uses grid spacing as an accuracy score would declare Model A automatically more accurate. The verification data in this original example show why that conclusion is invalid. The finer grid gives Model A a finer computational mesh, but in this particular forecast the 3 km model placed the rain band closer to the observed position.

That does not prove 3 km grids are always better. It proves something more important: mesh spacing and forecast error are different quantities.

The Four-Layer Test: Mesh, Resolved Feature, Forecast, Verification

When you see a weather-model resolution claim, separate four layers.

• Mesh: how the model divides the atmosphere horizontally and vertically for computation.
• Resolved feature: how small a weather structure the model can meaningfully represent. This is usually larger than one grid interval.
• Forecast: what the model predicts at a future time.
• Verification: how the forecast compares with observations after the event.

Confusion happens when the first layer is treated as if it directly supplies the fourth.

Why One Grid Cell Is Not One Perfectly Known Square of Atmosphere

A numerical weather model cannot follow every air molecule. It represents large-scale physical quantities such as temperature, pressure, wind and humidity on a mathematical grid or mesh. Equations are solved step by step to move the represented atmospheric state forward in time.

A 1 km grid therefore means the model’s horizontal calculation points are closely spaced compared with a 10 km or 25 km grid. That can help the model describe smaller spatial variations and more detailed terrain. But the value at a grid point is not a tiny weather station that has perfectly measured the real atmosphere at that exact spot.

UCAR’s meteorology teaching material explicitly warns that the word “resolution” is often used loosely for grid spacing even though the two are not identical. Multiple grid boxes are needed to represent a feature meaningfully. ECMWF likewise explains that effective resolution is larger than raw grid spacing because the smallest grid-scale structures cannot be simulated perfectly.

Representation Check: What Does the “1 km” Number Attach To?

Before interpreting the number, identify its noun.

• 1 km grid spacing — distance between neighbouring points or characteristic cell spacing in the model mesh.
• 1 km map pixel — display or product sampling scale; this may be different from the model’s native mesh.
• 1 km observation spacing — distance between instruments; again a different property.
• 1 km location error — a verification result comparing forecast and observation.
• 1 km uncertainty — an uncertainty statement with its own definition.

All contain the same unit. They do not measure the same thing. Unit matching is not meaning matching.

Starting Conditions Matter: The Model Cannot Correct What It Never Knew Perfectly

Weather forecasting starts from an estimate of the atmosphere now. That estimate combines many observations with a model through data assimilation. Observations are numerous but not everywhere, not continuous and not error-free. The starting state is therefore extraordinarily good but still an estimate.

A finer mesh cannot magically recover every missing detail in the starting atmosphere. If the initial position, moisture structure or wind pattern of an important feature is imperfect, that error can influence the future forecast. As lead time increases, small differences can grow.

This is why operational centres do not judge models only by their grid spacing. They verify forecasts against observations and often run ensembles to explore uncertainty.

Model Physics Matters: Some Processes Are Smaller Than the Mesh

Cloud droplets, turbulence, surface exchange and many other processes happen at scales smaller than even a fine weather grid. Models therefore use physical approximations and parameterisations to represent effects that cannot be explicitly calculated in complete detail.

Making the grid finer can improve representation of some structures, but it does not remove every approximation. The Met Office’s work on the “grey zone” shows why kilometre-scale modelling still involves difficult choices about which processes are explicitly resolved and which remain parameterised.

For a Primary 5/6 learner, the transfer is straightforward: smaller boxes can show more detail, but more detail is not the same as perfect truth.

Worked Case 1: The Thunderstorm Is Two Cells Away

A 1 km-grid model predicts the centre of a thunderstorm over Location P at 4 pm. Radar later shows the storm centre 12 km east at 4 pm. A student says, “That cannot happen because the model spacing was only 1 km.”

It can happen. Grid spacing describes how the model represents space, not a maximum storm-position error. Convection is sensitive to moisture, instability, boundaries, terrain and the evolving atmospheric state. The storm can develop with a substantial location error even in a fine-grid model.

What evidence evaluates accuracy? Compare the forecast storm location, timing and intensity against radar, rain gauges or other appropriate observations over many cases.

Worked Case 2: Finer Detail Can Reveal a Hill Without Fixing the Rainfall

Model A uses a 1 km grid and represents a narrow ridge. Model B uses a 10 km grid and smooths much of the ridge into a broader hill. In a rain event, Model A places heavy rain on the wrong side of the ridge while Model B gives a closer regional rainfall total.

Model A’s finer topography is valuable. Yet rainfall accuracy also depends on wind, moisture and cloud processes. A finer mesh can improve the ingredients available to the model without guaranteeing the final forecast will be closer in every variable and every case.

Worked Case 3: Same Grid Spacing, Different Forecast Skill

Two models both advertise 2 km grid spacing. Model X starts from newer observations and uses one set of cloud and surface schemes. Model Y uses a different assimilation system and different physics. Their forecasts disagree.

The shared grid spacing does not force equal accuracy. Grid spacing is one design property among many. To compare the forecasts scientifically, use verification data for the variable, place, season and lead time that matter.

Worked Case 4: A 1 km Forecast Map Does Not Mean Street-Level Certainty

A weather app displays coloured rain probabilities in 1 km squares. A family wants to know whether one particular playground will be dry at 3:15 pm. The fine-looking map encourages a false sense of exactness.

The display cell may be derived from model information with fine spatial sampling, but the atmosphere does not obey cell boundaries. A shower can form, move, weaken or miss a point. The appropriate conclusion may be that the area has a certain forecast risk, not that one exact playground boundary has a guaranteed outcome.

Comparison Check: Is a Finer Grid Automatically a Better Model?

Often, a finer grid can provide real benefits: more detailed terrain, more explicit representation of smaller weather systems and less spatial averaging. But “finer” is not a universal score.

• Are the two models verified for the same variable?
• Are you comparing the same forecast lead time?
• Do both use observations of similar freshness and coverage?
• Are their physics and data-assimilation systems comparable?
• Is the weather feature large enough to be meaningfully represented?
• Are you judging one dramatic event or many independent cases?
• Is the fine output merely interpolated from a coarser native model?
• Does the verification metric match the question?

The strongest comparison is not “which grid number is smaller?” It is “which forecast system performs better for the job, when checked fairly against observations?”

Baseline Check: Accuracy Compared With What?

A claim such as “our 1 km model is more accurate” needs a baseline. More accurate than an older version? A 3 km model? A simple persistence forecast? For temperature, rain amount, wind, storm tracks or something else? Over what period and region?

Without those comparison details, “more accurate” is too vague to evaluate. Grid spacing may explain one model change, but verified performance supplies the evidence.

Alternative Explanations When a Fine-Grid Forecast Misses

If a fine-resolution forecast gets a storm wrong, do not leap to one cause. Plausible explanations include:

• the starting atmospheric analysis had an important error;
• small initial differences grew during the forecast;
• the feature was too small or short-lived to predict deterministically;
• cloud, turbulence or surface processes were represented imperfectly;
• terrain or land-surface data were incomplete;
• the forecast time was correct but the location was displaced;
• the observation used for verification also had limitations;
• the display was post-processed or interpolated and did not equal the native grid.

Science gets stronger when a miss starts an investigation instead of producing the one-line explanation “the grid was too coarse” or “the model was bad.”

Evidence That Strengthens a Forecast-Accuracy Claim

• Accuracy is defined with a clear verification metric.
• The model is compared with appropriate observations.
• Many independent forecast cases are evaluated, not one success.
• The same region, season, variable and lead time are used in the comparison.
• The improvement remains after accounting for observation uncertainty.
• The test includes difficult as well as easy weather situations.
• Results are reported separately for location, timing and magnitude where relevant.

Evidence That Weakens the Claim “1 km Grid = 1 km Accurate”

• The claim uses only the grid-spacing specification and no verification.
• It treats one grid cell as a maximum possible forecast error.
• It assumes every feature as small as one cell is fully resolved.
• It ignores lead time and initial-condition uncertainty.
• It compares screenshots rather than quantified performance.
• It confuses native grid spacing with display-pixel size.
• It generalises from one successful forecast to all weather.

Tempting but Invalid Reasoning

“The model uses 1 km squares, so each square’s weather is known exactly.” A grid cell is part of a numerical representation, not a perfectly observed box of air.

“A 1 km model cannot be wrong by 10 km.” It can. Forecast-position error is not capped by grid spacing.

“A smaller grid always gives a better forecast.” A finer mesh can represent more detail, but total forecast skill depends on the whole system and the event.

“The screen has 1 km pixels, so the model itself must be 1 km.” Display resolution, interpolation and native model resolution can differ.

“A detailed-looking forecast is a certain forecast.” Visual detail and evidential certainty are separate properties.

How Far Can the Conclusion Travel?

From a trustworthy statement that a weather model has approximately 1 km horizontal grid spacing, you may conclude that its computational mesh is much finer than that of a 10 km or 25 km model, and that it has the potential to represent smaller spatial structures.

You may not automatically conclude that every feature smaller than 1 km is resolved, that every predicted feature will be located within 1 km, that a 1 km model is always more accurate than a coarser model, or that a map value is a direct observation at that point.

To make an accuracy claim, add verification evidence.

PSLE-Style Transfer Case

Original case: Weather Model P has horizontal grid spacing of 2 km. It forecasts the centre of a heavy-rain area over Town A at 5 pm. Radar observations later show the centre 14 km east of Town A. A student says, “The radar must be wrong because a 2 km model cannot have a 14 km location error.”

Evaluate the student’s reasoning.

Explained answer: The reasoning is incorrect. The 2 km value describes the model’s horizontal grid spacing, not the maximum error in the predicted location of a rain area. Forecast position depends on the model’s starting conditions and representation of atmospheric processes. The forecast should be evaluated by comparing it with appropriate observations such as radar and rain gauges.

Practice: Sort the Number Into the Right Box

1. “Grid spacing = 3 km.” What category is this?
A model-design and spatial-representation property.

2. “Mean storm-track error = 35 km at 24 hours.” What category is this?
A verification result describing forecast performance under the stated evaluation.

3. Can a 3 km model have a 50 km storm-position error?
Yes. Grid spacing does not set a maximum forecast-location error.

4. Why can a finer grid still be scientifically useful?
It can represent terrain and atmospheric structures at finer scales and may improve forecasts when the rest of the model system supports that detail.

5. What evidence is missing from the slogan “1 km model = hyper-accurate”?
Independent forecast verification for the relevant variables, places, lead times and weather situations.

Delayed Independent Return: The Graph-Paper Challenge

Draw a 10 × 10 grid on graph paper and sketch a curved rain band crossing it. Now shift the whole rain band three squares east while keeping the grid unchanged. Ask yourself: did the grid spacing change? No. Did the forecast-location error change? Yes.

That simple drawing separates the quantities. Repeat it with a finer grid. The curve can be drawn with more detail, yet it can still be displaced from the observed curve. Detail and accuracy can work together, but one does not define the other.

Route to Existing Canonical PSLE Science Owners

For the underlying job of judging simplified representations, continue with How to Compare Two Scientific Models in PSLE Science and Decide Which One Is More Useful. For controlling what changes in a fair comparison, use How to Decode Variables and Fair Tests in PSLE Science Questions. For selecting the evidence that actually decides a claim, use How to Choose the Decisive Evidence for a PSLE Science Answer Without Copying the Whole Table. For moving carefully between diagrams, tables and graphs, use How to Translate the Same PSLE Science Relationship Between Words, Diagrams, Tables and Graphs.

Parent and Tutor Teaching Guide: The Chessboard Is Not the Prediction

Use a chessboard and a coin. Tell the learner each square represents 1 km. Place the coin on one square as the forecast storm centre. Then place a second coin four squares away as the observed centre. Ask, “Did the square size stop the forecast being four kilometres wrong?” The answer is immediately visible.

Next, draw a second board with smaller squares. A curved shape can now be represented with more detail. Ask whether the finer board guarantees that you place the shape in the correct position. Again, no. The board controls representational detail; the placement depends on the information and rules used to make the prediction.

Finish by asking the learner to design a fair model comparison. Require the same event, same observation dataset, same lead time and a clear error measure. This turns the lesson back toward PSLE Science inquiry rather than weather trivia.

Authoritative Sources

• Met Office — MOGREPS ensemble system: describes forecast variables calculated at grid points and explains multiple sources of weather-forecast uncertainty.
• ECMWF — Current forecasting equations are still valid near km-scale resolution: distinguishes horizontal grid spacing from effective resolution and explains that the smallest meaningfully resolved scale is larger than raw grid spacing.
• UCAR MetEd — Observation, Analysis and Prediction: explains that grid spacing and resolution are not identical and that multiple grid boxes are needed to resolve a feature.
• NOAA National Centers for Environmental Information — Global Forecast System: provides an operational example of model grid spacing and forecast configuration.
• Ministry of Education, Singapore — 2023 Primary Science syllabus: frames science learning around evidence, models, uncertainty, objectivity, open-mindedness and healthy scepticism.
• Singapore Examinations and Assessment Board — 2026 PSLE Science: assessment objectives include interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning.

The Core Habit

Weather models divide an impossibly complicated atmosphere into a form computers can calculate. Finer grids are an extraordinary engineering achievement, and they can make forecasts more detailed and often more useful. But a mesh is not a promise. When a page says “1 km grid spacing”, read it as a statement about the model’s spatial machinery. Then ask a separate scientific question: when this model predicts the real atmosphere, how well do its forecasts verify against observations? Keeping those two questions apart is the evidence habit that makes the number honest.