Investigate the errors for viewsheds at the extremes of the DEM #8
Description
One of the tradeoffs we make with our algorithm is that, during the main (namely GPU) calculations, all points share the same geometric projection anchored at the centre of the DEM. This is not a problem for viewsheds near that anchor but it could possibly add up to 100s of metres at the edges of the DEM. In the worst case for tiles that expect a longest-line-of-sight-on-the-planet, the distance from the centre could be up to around 1000km. The build up of inaccuracy must then further compound in the visibility calculations, where even millimetres near the beginning of the line of sight can exponentially increase towards the end of the line of sight — think of how "peek-a-booing" a few centimetres from behind a tree dramatically changes what you can see.
So just want to investigate the reality of this? Where do viewsheds problematically get detached from reality? And what can we do to mitigate this?