Poster: Comparison of Interpolation Methods for Global Scale Topographic Reconstruction of the Earth Back in Time

Topography modelling relies on interpolating a raster using input vector points containing elevation values. However, due to high-variations of terrain in mountain areas or continental margins,the accurate modelling of these area can become challenging. Geographic information systems (GIS) such as QGIS or ArcGIS offer plenty of interpolation methods, making it a complex task to identify which method and parameters are able to render the terrain with the most fidelity.

Studying the geography of the Earth past (palaeogeography), one cannot rely on spaceborne-derived digital elevation models (DEMs) of the current world, hence the need to use other input data, such as plate tectonics models. DEMs of the Earth past are very useful products that can help estimate sea-level variations and climates of the past (Marcilly et al. 2022).

Generating a palaeo-DEM form a plate tectonic model therefore strongly relies on interpolating a raster from an irregular grid of nodes extracted from the model. However, no studies have been conducted to assess the performances of the available inteprolation methods for a global scale-reconstruction.

We compare the performance of seven interpolation methods with varying parameters (including resolution, power and search radius) using ArcGIS and QGIS and its associated libraries (GDAL, GRASS, SAGA). We use the nodes from the PANALESIS model present-day reconstruction (Vérard et al. 2015), with elevation values resampled from ETOPO 2022 (NOAA 2022).

In total,130 inteprolations are compared with the ETOPO 2022 reference raster in terms of difference of absolute values, and using the Terrain Ruggedness Index (TRI) as an indicator of how much variation in the topogrpahy is captured.

Statistical analysis of the absolute and TRI difference show that best results (with minimal NRMSE for absolute differences and TRI) are obtained using GDAL. In general, open-source solutions available in QGIS tend to perform better than proprietary ones in ArcGIS.