Overstory-understory classification from discrete return aerial LiDAR

I have an opportunity to collaborate with bat biologists who are interested in predicting probability of roost occurence in seasonally flooded ponderosa woodland. We suspect that both properties of canopy and understory are relevant, and I hoped to extract metrics from each, eg, rumpleness and density, from a discrete return aerial LiDAR point cloud. My impression of the system is that there are single overstory trees with a predominantly herbaceous understory, and I’m wondering if there are useful tools for binarizing the cloud into overstory and understory in such a system. I’ve found recent methods that look for dips in histograms of Z values in a moving (XY) window, and make a cut at the dip. I’m wondering if there is a good candidate algorithm (currently published and implemented) that could perform this task?