Abstract
Background and aims: Several studies have shown an increase in tree mortality in intact tropical forests in recent decades. However, most studies are based on networks of field plots whose representativeness is debated. We examine the potential of repeated Airborne LiDAR Scanning data to map forest structure change over large areas with high spatial resolution and to detect tree mortality patterns at landscape level. Methods: The study site is a complex forested landscape in French Guiana with varied topographic positions, vegetation structures and disturbance history. We computed a Gap Dynamics Index from Canopy Height Models derived from successive LiDAR data sets (2009, 2015 and 2019) that we compared to field-measured mortality rates (in stem number and basal area loss) obtained from regular monitoring of 74 1.56-ha permanent plots. Results: At the plot level, the relation between gap dynamics and absolute basal area loss rate (combining fallen and standing dead trees) was overall highly significant (R 2 = 0.60) and especially tight for the 59 ha of unlogged forest (R 2 = 0.72). Basal area loss rate was better predicted from gap dynamics than stem loss rate. In particular, in previously logged plots, intense self-thinning of small stems did not translate into detectable gaps, leading to poor predictability of stem mortality by LiDAR in those forests severely disturbed 30 years before. At the landscape scale, LiDAR data revealed spatial patterns of gap creation that persisted over the successive analysis periods. Those spatial patterns were related to local topography and canopy height. High canopy forests and bottomlands were more dynamic, with a higher fraction of canopy affected by gaps per unit time indicating higher basal area loss rates. Conclusion: Gap detection and mapping via multitemporal LiDAR data is poised to become instrumental in characterizing landscape-scale forest response to current global change. Meaningful comparison of gap dynamics across time and space will, however, depend on consistent LiDAR acquisitions characteristics.