Run the tree point segmentation by the watershed segmentation algorithm of Vincent and Soille (1991) to identify the individual trees

The processed CHM will be treated as a topographic surface and identify the ridges and valleys that separate the watersheds of each tree

Flood the surface from the lowest points and mark each pixel (volex) with the label of the corresponding watershed

In the watershed segmentation algorithm, 

The gradient of the CHM will be defined as G(x,y) = ||∇CHM(x,y)||, where ∇ is the gradient operator, the set of markers as M = {x | G(x) = 0} and the set of catchment basins as B = {B_i | i = 1,...,N}, where N is the number of catchment basins and B_i is the set of all points that flow to marker i.

The set of watersheds as W = {W_i | i = 1,...,N}, where W_i is the set of all points that flow to catchment basin i.

For each of the segmented trees, computed the height, crown spread, inclination and extract the mid-point coordinates.

             Height is determined by the maximum z point of the segmented tree minus the averaged height of the interpolated digital terrain model from classification code 2.

             Crown spread is estimated by the Minimum bounding rectangle (MBR), the maximum and minimum x and y coordinate of the segmented crown, it is used to define the corners of the MBR, apply the distance formula to the resulted length and width of the bounding box will be the estimated crown spread.

             Tree inclination is computated by the Principal component analysis (PCA), by computing the principal components of the tree points, this code is essentially finding the direction of maximum variance in the x and y coordinates of the tree points. Then, apply the arctan to calculate the correct quadrant and inclination of the tree angle.

Canopy-slope intergrate simulation 

Digital Terrain model and surface model was created and result by the Airborne LiDAR survey, the DTM shows the original terrain profile (included) the cutted and filled slopes and the DSM shows the canopy layers among the man-made slopes. Therefore, we can adopt both of them do eveulated the flow direction and simulate the flow accumulation areas. 

Compute the CHM model and the flow directions from DTM and DSM. 

Then the DSM and result from the Auto-tree inventory can help us to plot the trees along the slope for referencing. 

Calculate the flow direction and accumulation raster by the DTM, create a fishnet plotting for the data visulisation, the raster computation is avalibale online in https://huggingface.co/spaces/OttoYu/FlowDynamics. 

Then, we can make use of the flow raster to eveulate the protential root decays and damage areas that will higher flow accumulation rate.

NDVI from Tucker, C. J. (1979) is computed by subtracting the reflectance of near-infrared (NIR) radiation from the reflectance of visible red light and dividing the result by the total of the two. This indicator is sensitive to variations in plant cover and productivity since it primarily monitors the quantity of chlorophyll in vegetation. NDVI levels are often greater in healthy vegetation than in stressed or barren regions.

EVI from Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002) is a modified form of NDVI that considers atmospheric impacts as well as other elements that might influence plant reflectance. It is more sensitive to changes in plant cover and productivity than NDVI and is frequently employed in locations where there are significant aerosol concentrations or other atmospheric disturbances.

GCI from Gitelson, A. A., & Merzlyak, M. N. (1994) is especially intended to assess the quantity of green chlorophyll in plants. It is computed by dividing the reflectance of green light by the reflectance of red light. 

In the Wang, M., & Wong, M. S. (2023) proposed algorithm is evaluated using two types of tropical trees with two different branch structures, i.e. linear branch structure and complex branch structure of large and crown-heavy tropical trees and general tropical trees, respectively. The separation results demonstrate that the proposed method can obtain promising wood-leaf separation accuracy for both general and large trees.

Still, there will be very different case in the MMS (close-ranging mobile mapping system) and ALS, as the point cloud LoD (Level of detials) are very different and coarse, spreaded compare to the TLS and handheld scanning, also the scanning angles is limited for the MMS and ALS due to they are embed into the vehicles.

On the other hand, Forestree is still developing a solution to deal with MMS and ALS data for tree species identification, we have already explore 350+ tree samples and 33 common species in Hong Kong. There are some key findings though our explorations, such as the intensity can help use to derive the co-dominate trucks, point spacing can shows the tree leaf style (needle-like or board leaf), geometries and point alignment can shows the type of tree by it shapes.