Abstract
In addition to timber and non-timber forest products, tropical forests host the largest amount of terrestrial carbon in the world, thus its importance for reducing the effects of climate change. This carbon is generally estimated as aboveground biomass (AGB) because about half of the organic matter is carbon. Therefore, estimating AGB trough differentapproaches and the calibration of relationships between AGB and dendrometric parameters (allometric equations) are of great importance for the estimation carbon stock. From a functional perspective, data from AGB and leaf (e.g. leaf area) ameliorates our understanding of the functioning of tropical forests and their interactions with the atmosphere. However, destructive data sampling commonly used is cumbersome to implement and presents significant uncertainty. This leads to a considerable deficit of accurate data. Studies carried intemperate forests show that using terrestrial LiDAR technologies (TLS) may provide a solution to the limitations of destructive sampling. The TLS produce a tridimensional points cloud which describes with a high precision structures within their natural environment.Therefore, the overall goal of this thesis is to evaluate the potential of using TLS data reducing the lack of quality data observed in the Congo Basin. Here we used TLS data to: i) establish AGB and height-diameter (HD) allometric models ; ii) evaluate the potential impact of vertical variation of wood density (WD) on AGB estimates derived from TLS data ; and finally, iii) ameliorate LA estimation of trees and calibrate allometric model for the prediction of this area using data which derived from LiDAR data.At the local scale, (semi-deciduous forest of south East Cameroon), two datasets were collected. The first has 61 trees and opposes TLS data to destructive data. The second dataset collected in quadrats within plots had 712 trees and opposed TLS data to classic field inventory data. At the regional scale (forest from the Congo Basin), a uniform destructivesampling realized on 821 trees permitted us to obtain WD data, volume data and AGB for each compartment. Thanks to an automatic method of topology reconstruction and the geometry of trees (Quantitative Structure Model QSM) based on the adjustment of cylinders in the points cloud, we compared dendrometric parameters i.e. volumes, AGB and allometricmodels resulting from TLS data to those derived from the destructive and inventory sampling method.It appears that, the TLS estimates wood volume with great precision (88%) and low bias (4.6%). Related to this, the AGB allometric model (R²=95%) established from TLS data is statistically similar to the one obtained with destructive data (R²=98%). Nevertheless, theselast results depend on the manual editing of QSM for some compartments of the tree because obtained AGB with raw QSM (without manual edition) lead to a different allometry (R²=93%). Despite these results, comparing total tree height data collected in plots scalerevealed that there is an average difference of 3 meters between total height derived from TLS and those collected from classical inventory. Using these two datasets separately led to the selection of a similar HD predictive model ; but the AGB estimated with predicted heights byinventory method is systematic underestimation of 10% per ha. The principal component analysis of WD by the different compartments helped in discriminating tree mayor types of WD vertical gradients: constant, decreasing, and increasing. These vertical gradients areconserved within species and are strongly correlated to species guilds, the basal density (WDStu) and the density issued of global databases (GWD). Thus, neglecting the existence of these gradients during the conversion of volumes derived from TLS to AGB leads to an individual bias (rising up to 73%) and an average bias of 8.12%. But using an unbiased estimator of WD defined in this work limits the bias to 0.75% with the WDStu and 1.19% with the GWD. Apart from these results on the AGB, manual segmentation between wood and trees realized before the automatic adjustment of the QSM made it possible to use thevoxelization technique on the leaves points cloud. This is with the aim of subdividing the tree crown in cubic volumes (voxels) to estimate the LA. Comparing LA derived from TLS to those obtained with the destructive method revealed that commonly using the spherical distribution as the “typical” angular distribution of leaves led to a fairly high average bias ofLA (17.28 %) against 6.5% when the distribution is computed per tree. This estimated leaf area (LA) is strongly correlated with DBH (r = 0.88) and AGB (r = 0.97). Linear models established between these variables produced R² ranging from 72 to 95%, illustrating a strong intensity of the link between LA and these two variables.To conclude, using TLS data would facilitate the implementation of international programs whose objective to achieve Tier III levels of accuracy necessary in estimating carbon stocks as recommended by the Intergovernmental Panel on Climate change.