Since you're already binning your data points, an alternative final step could be to create a contour plot (this would typically require a bit of smoothing over the binned data). Have you tried exploring that direction alongside the density plots?

Hi Job, I haven't tried that. It's an interesting thought – are contour plot methods significantly slower than binning methods on raw data?

I'm not the right person to ask, tbh - in practice I just use https://github.com/d3/d3-contour. But you already got the binned values anyway, you might as well try plugging them into that library and see what the results are!