Forest inventories are the backbone of forest management in Germany. In most federal forestry administrations in Germany, they are performed every ten years in order to assure that logging activities are sustainable. The process involves trained foresters who visit each stand (i.e. an area where the forest is similar in terms of age structure and tree species) and perform angle count sampling as developed by Walter Bitterlich in 1984. In a second step the annual growth is calculated using yield tables and finally a harvest volume is derived. There are three particular reasons to investigate how remote sensing can be integrated in the current inventory system:
- The current process does not involve random sampling of the sampling points and thus does not offer any measure of the accuracy of the data.
- Forest engineers hardly ever rely on the inventory data as a stand-alone basis for logging planning. Most often they rely on intuition alone and on the total volume count that they have to deliver for a wider area every year.
- In the last ten years, the collection of high-resolution LiDAR data has become more cost-effective and most federal agencies in Germany have access to it.
In order to be able to integrate the available remote-sensing data for forest inventories in Germany, it is important to tell apart different tree species as well as estimate their volumes.
The goal of this project is to perform an object-based classification of conifer trees in Northern Hesse based on high-resolution LiDAR and multi-spectral orthophotos. The first step is to delineate the tree crowns. The second step is to perform a semi-automated classification using the spectral signature of the different conifer species.
+ DSM (1m), DTM (1m), DSM (0.2 m) of the study area
+ Stereo images with 0.2 m resolution
+ high-resolution LiDAR data (average 10 points/m²)
+ forest inventory data
+ vector files of the individual forest stands
+ ground control points (field data)
All of this data is provided by the Hessian Forest Agency (HessenForst).
1) merge and clip the LAZ files [las2las]
2) classify ground and non-ground points [lasground]
3) remove low and high outliers [lasheight, lasnoise]
4) identify buildings within the study area [lasclassify]
5) create a normalized point cloud [lasheight]
6) create a highest-return canopy height model (CHM) [lasthin, las2dem]
7) create a pit-free (CHM) with the spike-free algorithm [las2dem]