Quantification of Tropical Forest Biomass with Terrestrial LiDAR and 3D Tree Quantitative Structure Modelling

Tropical forest biomass is a key component for the global carbon cycle dynamic. Still large uncertainties persist in its estimation and spatial distribution mapping [1,2]. The ESA Biomass mission will fill the gap of EO systematic coverage over the tropics, non-affected by clouds and allowing for accurate estimations of the very high C content in this region. However, the calibration and validation of such estimates will require accurate and effective methods for in situ quantification of the AGB at plot level. Currently methods used for this are based in very indirect estimations relying in allometric models. These models were reported to be a major source of uncertainty in the biomass estimation in the tropics [3], and with special significant deviations for very large trees[4], which are of high relevance due to their very large contribution to the total biomass of tropical forest [5].

Terrestrial LiDAR Scanner (TLS) data, in contrast, has been demonstrated to be more accurate for inferring forest aboveground biomass (AGB) in a non-destructive and more direct estimation approach [6]. The method consists of measuring tree wood volume by modelling the 3D tree architecture reconstructed with TLS data [7] and converted to biomass using specific wood density. This novel approach was first applied and validated in an open Eucalyptus forest in Australia [6]. However, the application of this method for tropical forest present major challenges, mostly related to the high levels of occlusion due to large vegetation density and structural complexity. This is especially acute for very large trees that tend to be emerging above the main canopy, increasing the level of occlusion. To the best of our knowledge, the current study is the first one validating the accuracy of the use of TLS data and the 3D tree QSM to assess AGB in tropical forest.

Twenty nine plots where stablished in three study sites: in the south-western Amazon, (Peru), north-eastern Amazon (Guyana) and southern Borneo (Indonesia). Forest inventory data was collected in the 29 plots where 38 trees were harvested in total. Plots of 30x50m (9 in Peru) and 30x40m (10 in Indonesia and 10 in Guyana) were installed around a major tree to be harvested. For validation of individual AGB of harvested trees, a reference dataset was produced by estimating the tree volume from detailed geometric measurements of the stem and all branches (until 10 cm diameter), and converted to biomass using its specific wood basic density. TLS data was acquired in the 29 plots before and after harvest using a Riegl VZ 400 scanner. To compare with current methods, AGB estimations were computed using the most local specific allometric models and the most recent pan-tropical models [8].

Preliminary results were calculated for the first site (Peru). The AGB estimates by the TLS-QSM model outperformed the allometric models tested (see Table 1 in supplementary information). The gain in the accuracy achieved by the TLS-QSM method in relation to the pan-tropical allometric model [6] was a 9.8% decrease in CV RMSE. This represents a similar gain in accuracy as the one reported by Calders et al [6]. Moreover, the bias of the TLS-QSM method was lower than the allometric methods, minimizing the underestimation of AGB of large trees.

We conclude that the proposed method (TLS-QSM) can be used to estimate AGB of large trees in tropical forests more accurately than allometric models tested, even under very unfavourable conditions (very large and complex tree crowns, within a great vegetation density, and above the main forest canopy).