Selective logging is a dominant form of land use in the Amazon basin and throughout the humid tropics, yet little is known about the spatial variability of forest canopy gap formation and closure following timber harvests. We established chronosequences of large-area (14-158 ha) selective logging sites spanning a 3.5-year period of forest regeneration and two distinct harvest methods: conventional logging (CL) and reduced-impact logging (RIL). Our goals were to: (1) determine the spatial characteristics of canopy gap fraction immediately following selective logging in the eastern Amazon; (2) determine the degree and rate of canopy closure in early years following harvest among the major landscape features associated with logging - tree falls, roads, skid trails and log decks; and (3) quantify spatial and temporal differences in canopy opening and closure in high- and low-damage harvests (CL vs. RIL). Across a wide range of harvest intensities (2.6-6.4 felled trees ha(-1)), the majority of ground damage occurred as skid trails (4-12%), whereas log decks and roads were only a small contributor to the total ground damage (<2%). Despite similar timber harvest intensities, CL resulted in more ground damage than RIL. Neither the number of log decks nor their individual or total area was correlated with the number of trees removed or intensity of tree harvesting (trees ha(-1)). The area of skids was well correlated with the ground area damaged (m(2)) per tree felled. In recently logged forest (0.5 years postharvest), gap fractions were highest in log decks (mean RIL=0.83, CL=0.99) and lowest in tree-fall areas (RIL: 0.26, CL: 0.41). However, the small surface area of log decks made their contribution to the total area-integrated forest gap fraction minor. In contrast, tree falls accounted for more than two-thirds of the area disturbed, but the canopy gaps associated with felled trees were much smaller than for log decks, roads and skids. Canopy openings decreased in size with distance from each felled tree crown. At 0.5 years postharvest, the area initially affected by the felling of each tree was approximately 100 m in radius for CL and 50 m for RIL. Initial decreases in gap fraction during the first 1.5 years of regrowth diminished in subsequent years. Throughout the 3.5-year period of forest recovery, tree-fall gap fractions remained higher in CL than in RIL treatments, but canopy gap closure rates were higher in CL than in RIL areas. During the observed recovery period, the canopy gap area affected by harvesting decreased in radius around each felled tree from 100 to 40 m in CL, and from 50 to 10 m in RIL. The results suggest that the full spatial and temporal dynamics of canopy gap fraction must be understood and monitored to predict the effects of selective logging on regional energy balance and climate regimes, biogeochemical processes including carbon cycling, and plant and faunal population dynamics. This paper also shows that remote sensing of log decks alone will not provide an accurate assessment of total forest area impacted by selective logging, nor will it be closely correlated to damage levels and canopy gap closure rates.
