Is my request too fine?
In many cases, yes. Requesting a finer grid than the underlying terrain does not create detail; it resamples the same elevation posts into more pixels. The cost rises with the square of the grid, while the information does not increase at all.
The one line that decides it
useful grid (px per side) ≈ width of the map (m) ÷ terrain post spacing (m)
Width of the map is the full span edge to edge, so for a radius-based request it is twice the range. Terrain post spacing is 1 m where LiDAR has been flown, and about 10 m (1/3 arc-second) everywhere else in CONUS.
Beyond that number, every additional pixel is interpolated from elevation data that contains no further detail. The result looks smoother, but it gives no different answer about where your signal reaches, and each doubling costs four times as much.
Worked numbers
| Map width | Terrain tier | Terrain-matched grid | Credits at that grid |
|---|---|---|---|
| 10 km | 1/3 arc-sec (~10 m) | 1 000 px | 1 |
| 20 km | 1/3 arc-sec (~10 m) | 2 000 px | 4 |
| 50 km | 1/3 arc-sec (~10 m) | 5 000 px | 25 |
| 100 km | 1/3 arc-sec (~10 m) | 10 000 px | 100 |
| 2 km | 1 m LiDAR | 2 000 px | 4 |
| 8 km | 1 m LiDAR | 8 000 px | 64 |
Credits here are the square of the grid divided by a million, rounded up: one transmitter at supersample ×1. Both of those multiply on top — see Pricing for the full formula. Nothing in this table is a price.
What an over-fine request costs
Consider a 50 km-wide study on 1/3 arc-second terrain. The terrain-matched grid is 5 000 px, which is 25 credits. Compare that with the same 50 km requested at 16384 px:
| Request | Ground sample | Credits | New terrain detail |
|---|---|---|---|
| 50 km at 5 000 px | 10 m/px | 25 | all of it |
| 50 km at 16 384 px | 3.1 m/px | 269 | none |
That is 10.7× the cost for zero additional elevation information. 16384² is also the largest size we have measured, taking a mean of 152 seconds across 29 such renders, and here that time produces what is effectively a 5 000 px render spread across more pixels.
Range is not currently a factor
At a fixed output size, asking for a wider area does not change your credit cost: range is not in the formula. Treat that as NOT FINAL rather than a promise — it is listed among the undecided pricing parameters. What widening the area definitely does is coarsen your ground sample, which is the decision this page is about.
Supersampling is not free: its cost is squared
Supersample ×2 is four times the credits and ×3 is nine times, because supersampling is passes: the scene is rendered that many times and resolved into an output image of unchanged size. It does not show up in a megapixel count, which is why supersample² is in the formula.
It is still the right setting when you want a smoother image rather than more detail, since a finer grid costs four times per doubling and adds nothing the terrain does not contain. Note, however, that ×2 supersampling and a doubled grid cost the same: four times the credits.
Both directions cost money
Requesting too fine a grid is the common and expensive mistake. Very small renders are also proportionally more expensive per megapixel, because every render is rounded up to a whole credit. The most economical choice is a grid that matches the terrain.
When a finer grid is worth it
There are three cases in which a finer grid is justified:
- You are inside the 1 m LiDAR footprint. Where LiDAR has been flown the terrain is ten times finer, and the matched grid increases accordingly. A 2 km-wide site study over 1 m data supports 2 000 px.
- Buildings matter more than terrain. Building footprints enter the model as knife edges with their own outlines, so their positions are not limited by the elevation post spacing. In dense urban work a somewhat finer grid resolves building edges that the elevation data alone could not show.
- The output is a deliverable, not an analysis. If the map is going into a document at a fixed print size, the pixel count you need is set by the page, not by the terrain. That is a legitimate reason to pay for pixels, provided it is understood that presentation, not additional detail, is what is being purchased.
A short checklist before you spend
- Work out the map width in metres (twice your range, if range is a radius).
- Divide by 10 — or by 1 if you know you are inside flown LiDAR. That is your terrain-matched grid.
- If the grid you intended to request is more than about twice that, the additional pixels are interpolation.
- Square your grid and divide by a million, then multiply by the square of your supersample setting and by the number of transmitters, and round up. That is your credit cost, and it is the number to sanity-check, not the pixel count.
- If the result still looks coarse, check whether the cause is the terrain rather than the grid. 10 m elevation data looks coarse at any pixel count, and more pixels will not change that.
See also: Pricing for how credits are counted, and the FAQ on full versus small renders.