Authors: Michael Gebreslasie*, University of KwaZulu-Natal
Topics: Remote Sensing, Environmental Science, Africa
Keywords: LiDAR, Terrain variables, Random forest regression, Forest inventory
Session Type: Paper
Start / End Time: 10:00 AM / 11:40 AM
Room: Maurepas, Sheraton, 3rd Floor
Presentation File: No File Uploaded
Reliable forest inventory measurements can assist commercial forest managers to make informed decisions pertaining to forest volume yields and harvesting planning including, designing of silvicultural protocols within stands of commercial plantations. Thus, the aim of this study is to characterize variation in forest structural attribute measurements of Eucalyptus species based on terrain variability derived from a Light Detection and Ranging (LiDAR) dataset. In this study, 32 terrain variables at five different spatial scales were computed from a LiDAR derived digital terrain model (DTM). Field data were collected for 502 plots within the study area in Richmond, KwaZulu-Natal, South Africa. A Random Forest (RF) regression statistical technique was applied to model the effect of terrain on forest structural variables (volume, heights, and Diameter Breast Heights (DBH)). Variable importance indicated that the incoming solar radiation terrain variable is the most significant variable for modelling forest structural variability. The results from the RF regression showed that HtD is impacted most by the terrain variable, incoming solar and consistently returned the highest coefficient of determination of 70% with RMSE of 1.24 m/ha for young E. dunnii, while the coefficient of determination for mature E.dunni is 74% with a RMSE of 2.33 m/ha. Whereas for a young E. grandis is R2=0.58 with a RMSE of 1.24m/ha and for mature E. grandis is 0.45 with a RMSE of 2.18m/ha. The findings indicate that the terrain variable incoming solar radiation useful for height stratification within plantation forests of South Africa.