An application-driven approach to terrain model construction

Terrain is a surface phenomenon that is measured, modelled, and mapped. However, it is continuously variable and must be simulated by points or mathematical equations that are inherently approximations. The error induced by digitally represented terrain can propagate to surface derivatives and geographical information science (GIS) applications where topography is considered. This can lead to uncertainty in model predictions and the use of data that are unfit for the application to which they are intended. This article outlines the problem of uncertainty in terrain representation and demonstrates the consequences for volcanic mudflow modelling. The response of a simple least-cost single flow algorithm to input parameters was investigated in order to assess output variation from the different sources of input variation. Elevation error was modelled with a probability density function (PDF) and propagated through stochastic simulation (Monte Carlo). Such combined uncertainty and sensitivity analyses enabled a qualitative judgement of the relative significance of elevation error on the flow model prediction. Different methods for terrain model construction were considered and show that supplementing global positioning system (GPS) measurements with information from field notes and reconnaissance photographs greatly improved the model performance and reduced the uncertainty. It is concluded that in terms of validity of model results, there is no substitute for constructing an elevation model that is informed by the terrain.