ELEVATION ACCURACY
How many centimetres off is the surface? STREAM's Elevation Accuracy panel compares points of known elevation — ground control points, levelling, GNSS — with the surface you choose and reports, point by point, difference = surface elevation − point elevation. The summary gives RMSE, mean difference, standard deviation, robust deviation and 95% accuracy; an interpretation card says in plain sentences what the numbers mean; and a PDF or Excel report documents it. Nothing is assumed for you: no surface is preselected, and no pass / fail verdict is given until you enter an acceptance criterion.
A DEM accuracy assessment measures how closely an elevation surface matches independent field measurements. Each point has an elevation measured on the ground; the surface is read at the same position; the differences are summarised as a root mean square error (RMSE) and related statistics. It is the usual way to document a drone or LiDAR surface before volumes are computed from it.
The answer is only as good as the points. Points used to build the model (GCPs) give an optimistic figure; independent check points give the honest one. STREAM lets you label each point as GCP or Check and count only one kind. A handful of points clustered in one corner of the site does not represent the whole surface.
Read a CSV / TXT with name, X, Y and measured elevation — the preview says how many points fall on the surface and swaps X ↔ Y in one click; click the map and type the measured elevation; or take the points already selected.
Mean difference (systematic shift), standard deviation, RMSE, mean absolute difference, median, robust deviation (NMAD), minimum / maximum, 95% accuracy (1.96 × RMSE) and the 95th percentile, which is not given for fewer than 20 points.
A suspected blunder is flagged, never removed for you, with the RMSE you would get without it. A systematic shift is detected and the surface can be offset by it, undoably. A spatial trend, steep ground and a vertical-reference mismatch are called out.
Give each point a kind and use the Scope card to count only control points or only check points. A result from GCPs alone can be optimistic, and the panel says so.
While the panel is open the points are marked in the colour of their difference — surface high or low — with a legend; excluded points, suspected blunders and points off the surface have their own marks. Clicking a table row selects the point on the map.
The report carries the point location map, deviation measures, point list, difference chart, excluded and non-comparable points with their reasons, and method notes. It is written from the panel's tables; nothing is computed a second time.
Difference = surface elevation − point elevation. A positive difference means the surface is higher than the measured point; a negative one means it is lower.
A result is given from two points, but the statistics only carry weight with more: the 95th percentile, for example, is not reported for fewer than 20 points. Spread the points over the site rather than in one corner.
Only against your own criterion. The acceptance criterion starts empty; until you enter a value there is no pass / fail verdict.
Yes. If the mean difference cannot be explained by chance, the panel offers to offset the surface by that amount as an undoable raster offset record. A result from the same points is then no longer independent accuracy, and the report says so.
The one you choose in the Surface box, which opens unselected. With several terrain layers visible, Visible terrain (all layers) compares the composite surface on screen, raster offsets and terrain edits included.
Free beta for Windows x64. Works fully offline with GeoTIFF, LAS/LAZ and DXF.
Download STREAM free