For LCP analysis, slope is mostly calculated on the basis of a DEM. The precision of the slope values depends on four factors (Huggett and Cheesman 2002, 22):
With respect to archaeological studies, DEM accuracy does not only rely on the modern elevation measurements, but also depends on altitude differences to the surface of the period considered. Substantial modifications of the surface may result from geological processes such as landslides or flooding, or from human impact such as terracing or bulk material extraction.
Errors resulting from the interpolation of the recorded elevation values
In general, the irregularly spaced elevation points are converted to a raster grid by some sort of interpolation (e.g. Conolly and Lake 2006, 94–111; Wheatley and Gillings 2002, 190–9). This is a difficult task if digitised points on contour lines form the basis of this process (Wheatley and Gillings 2002, 113–18; Lock and Pouncett 2010). If third party DEM grids are acquired, information on the accuracy of the data should be part of the data delivered. LCP calculations based on a DEM modified by adding random noise within these error limits is one way of investigating the stability of least-cost analyses (Herzog and Posluschny 2011).
Errors due to spatial sampling effects
Reducing the resolution of the DEM grid always introduces some smoothing; average slope steepness decreases with increasing cell sizes (Herzog and Posluschny 2011). Consequently, grid resolution can have a major influence on the outcome of a cost accumulation algorithm (van Leusen 2002, chapter 6, 9).
Inadequacies of the algorithms used to compute the slope values
Different computational methods have been proposed for slope calculations in a raster grid, and these may produce significantly different slope values from the same DEM (e.g. Lock and Pouncett 2010). Some of these methods are discussed in introductory GIS books for archaeologists (Wheatley and Gillings 2002, 120–1; Conolly and Lake 2006, 191–2). Lock and Pouncett (2010) recommend using the approach for slope computation proposed by Evans, which is based on the altitude data of an n x n neighbourhood (with n ≥ 3). The advantage of this method is that it can be applied at different scales by varying n, moreover the impact of errors in the DEM is reduced. However, it requires some map algebra and is computationally intensive. Lock and Pouncett (2010) argue that by increasing n, the model begins to include global knowledge of the landscape. As increasing n also introduces some smoothing, the costs of traversing abrupt changes in altitude are underestimated. After creating the ACS, Lock and Pouncett apply a drain procedure of steepest descent, which does not necessarily result in the LCP (see Section 3.2), instead of a full implementation of Dijkstra's algorithm.
It seems that the impact of different slope calculation methods on LCP generation remains to be investigated.