Like jsotola mentioned, the way to find information on a thing is to look at its datasheet. From the datasheet (warning: direct download, found on the bottom-right of this page):
So, comparing your picture to the datasheet, it looks like:
Signal | Color
Google research indicates there that seem to be multiple techniques. Some involving multiple looks from different angles (see "High-resolution foliage penetration with gimbaled lidar"). Some involving multiple lasers where the interference between two beams cancels out the foliage return (see "Lidar system able to penetrate foliage and map ...
I saw the article a couple of years ago and thought the same. Looks like filtering is the only option. If they flew the drone near the surface, few of the multi-million beams might have hit the ground and I guess they were extracting the lowest points to generate the ground mesh.
I think that you will find that the laser_assembler in the laser_pipeline does exactly what you are looking for.
The laser_scan_assembler node accepts one topic as input. But you can remap or mux multiple laser scanners into that input topic and the assembler should be able to handle it. The algorithm should be using the actual headers from the incoming ...