A simulation method to infer tree allometry and forest structure from airborne laser scanning and forest inventories

International audience

إنجليزي. Tropical forests are characterized by large carbon stocks and high biodiversity, but they are increasingly threatened by human activities. Since structure strongly influences the functioning and resilience of forest communities and ecosystems, it is important to quantify it at fine spatial scales. Here, we propose a new simulation-based approach, the "Canopy Constructor", with which we quantified forest structure and biomass at two tropical forest sites, one in French Guiana, the other in Gabon. In a first step, the Canopy Constructor combines field inventories and airborne lidar scans to create virtual 3D representations of forest canopies that best fit the data. From those, it infers the forests' structure, including crown packing densities and allometric scaling relationships between tree dimensions. In a second step, the results of the first step are extrapolated to create virtual tree inventories over the whole lidar-scanned area. Across the French Guiana and Gabon plots, we reconstructed empirical canopies with a mean absolute error of 3.98 m [95% credibility interval: 3.02, 4.98], or 14.4%, and a small upwards bias of 0.66 m [−0.41, 1.8], or 2.7%. Height-stem diameter allometries were inferred with more precision than crown-stem diameter allome-tries, with generally larger heights at the Amazonian than the African site, but similar crown-stem diameter allometries. Plot-based aboveground biomass was inferred to be larger in French Guiana with 400.8 t ha −1 [366.2-437.9], compared to 302.2 t ha −1 in Gabon [267.8-336.8] and decreased to 299.8 t ha −1 [275.9-333.9] and 251.8 t ha −1 [206.7-291.7] at the landscape scale, respectively. Predictive accuracy of the extrapolation procedure had an RMSE of 53.7 t ha −1 (14.9%) at the 1 ha scale and 87.6 t ha −1 (24.2%) at the 0.25 ha scale, with a bias of −17.1 t ha −1 (−4.7%). This accuracy was similar to regression-based approaches, but the Canopy Constructor improved the representation of natural heterogeneity considerably, with its range of biomass estimates larger by 54% than regression-based estimates. The Canopy Constructor is a comprehensive inference procedure that provides fine-scale and individual-based reconstructions even in dense tropical forests. It may thus prove vital in the assessment and monitoring of those forests, and has the potential for a wider applicability, for example in the exploration of ecological and physiological relationships in space or the initialisation and calibration of forest growth models.