A Digital Elevation Model (DEM) is a digital 3D representation of the bare earth's topography, or "ground surface," used in geographic information systems (GIS). It is created from elevation data and can be represented as a raster (a grid of squares) or a triangular irregular network (TIN). Unlike a Digital Surface Model (DSM), a DEM excludes features like buildings and trees to provide a "bare-earth" model for applications such as flood modeling, land-use studies, and geological applications. What is the best interpolation method to calculate the elevation from a DEM?

Nearest Neighbour

simply takes the nearest pixel + Advantage: extremely fast, no smoothing, values are exactly those from the original grid - Disadvantage: staircase artefacts, no subpixel information, often visually ‘blocky’ Accuracy: o correct for categorical data (e.g. land use classes) o tends to be most inaccurate for elevation data because jumps occur

Bilinear Interpolation

takes the 4 nearest pixels and calculates a weighted average • A+ Advantage: smooth transitions, fewer artefacts • A- Disadvantage: smooths out small details → • Extreme elevations (e.g. mountain peaks) are slightly softened • Accuracy: o Typical standard for DEM resampling o Good balance between computational effort and smoothing

Cubic interpolation (cubic / bicubic)

uses the 16 nearest pixels, fits a cubic function • + Advantage: very smooth surfaces, attractive • visualisation • - Disadvantage: can produce overshoots (e.g. • heights that do not occur in the original → ‘fake peaks’ or depressions) • Accuracy: o oftenvisually the most attractive o numerically not always ‘realistic’ because it can generate values that exceed the original min/max

Aster 30m

With Aster we get a precision of ±7m

Swisstopo 0.5m

With Sisstopo 0.5m we get a precision of ±0.3m

Conclusion for DEMs (elevation models)

Accurate in the sense of faithful to the original -> nearest (no new values, but blocky) Accurate in the sense of best approximation of true terrain continuity -> bilinear (standard in GIS, often best compromise) Smoothest representation for visualisation -> cubic (nice, but can be exaggerated) Many GIS programmes (QGIS, GDAL itself) use bilinear for elevation data by default. If you want the raw values for scientific analysis -> nearest is better. You can find more information for Python an GDAL on my github repository.