I am looking for the methods on assessment of understorey grass biomass using Lidar technology. I have ground Lidar point cloud from forest plot and it looks challenging to detect grass as there are other features such as trees, shrubs and downed woods as well. Does anyone have any experience?
Thank very much in advance!
The Lidar point cloud you have was it from airborne or terrestrial? Your terrestrial scans in general should have a higher point density. If you have an airborne scan it will be much harder to work with. It all depends on your point density, laser spot size and the number of passes you took or the number of scans taken of your target research area. It also helps if your Lidar scanner can process multiple multiple returns or better yet waveform data. Here is a video from Linkedin of what you can see even with trees or how to remove unwanted vegetation : https://dms.licdn.com/playback/C5605AQHq6aGpgGWlWw/a3efe709ad8d4e1f98bca9844e2c218a/feedshare-mp4_3300-captions-thumbnails/1507940147251-drlcss?e=1564671600&v=beta&t=pt5rffqcSbbB9rMzdx23SchuG-VdENBl04yInYTvTgI
Dear Reuben Reyes, thank you very much for your response. I have single return terrestrial lidar point cloud. I took 7 scans in 1 hectare (50mX200m) plot, with point density of 6 mm at 10 m.
Single return will limit your penetration and definition into your understory structure. Not sure if you have enough data for what you are trying to accomplish. If you took a physical count of small sectioned areas within your scan areas to identify bio density types both live and dead this would be a good start. As a first step you could compare physical count to a point cloud filter at different ground height (Z) densities to see if you have any correlation. You could also use spherical point cloud filters or cylindrical point cloud filters to find if plant leaf/stem fit better to your physical count and actual physical count areas. This would be much harder to do as i dont know of any software that could accomplish spherical or cylindrical filters on point clouds without writing your own code to do the geometric filtering. I figure your point density will vary greatly at the median distance from scan origins and this may skew some of your results but you could add a correction factor based from a histogram on a coarse grid to identify thinning or higher concentrations of points within the grid. You have an interesting project and problem to solve, i hope some of the suggestions may be helpful.
Even with a point density of 6mm (GSD of 6mm?) in this case because the target blade of grass is so small and the possible gap between the grass bade and the terrain is so small you in addition need to have a point size on target of possibly just 5mm or better and multiple returns per point, but a larger point size can be used if in addition you have a full waveform processing unlimited return type scanner like the Riegl instruments, as long as sufficient point power falls on the terrain a return for that will be valid and form the terrain reference model. Also the scan position point of origin needs to be quite high in relation to the area being scanned to avoid low look angles through many blades of grass.
In most cases classification of this kind of biomass over this area is better achieved with multispectral imagery which can separate different types of grasses or shrubbery, where Lidar can only classify on shapes and elevations.
In either case, as Rueben points out, ground truthing by direct physical samples for validation will add precision to the outcome.