Hi Gerd One of the reasons I went along the PredictPoints route is that MapArea was calculating the subDivision usage (number of items, amount of data) based on totally unfiltered lines/polygons, regardless of the resolution, so this was forcing splits where no need whatsoever. I didn't want to add additional complexity to it (ie DouglasPeuker). Even without that it makes a big improvement, mainly at lower resolution levels. It must never underestimate the number of points, but slight overestimation is still masses better than previous behavior. Generally subdivisions are much fuller, hence there should be more scope for ShapeMergeFilter. Having backtracking to do the points-limit splitting will be better than doing it earlier based on PredictPoints I don't think this the full reason for the change in images size behaviour that you see; I need to think a bit more about grouping items during splitting. I can't do anything for a couple of days but I'd like to combine the methods. Ticker