Methods
Within each 50-ha treatment plot in July 2004, we tagged, mapped, and measured height (m) and dbh (at 1.3m) of all trees >40cm dbh (n~930 individuals per plot). Trees and lianas(20–39.9cm dbh) were sampled in belt transects (500×20m) at 0, 30, 100, 250, 500, and 750m from the edge (5.5 ha sampled per treatment plot; n$880 individuals per plot). Nested sub-sampling was conducted within these belt transects to measure trees and lianas (10–19.9 cm; 500×4 m; 1.2 ha sampled per treatment plot;
n~490 individuals per plot). To sample 5–9.9 and 1–4.9cm dbh stems, 2×10m and 1×10m sub-plots, respectively, were also nested within these belt transects (16 sub-plots, 0.032 and 0.016 ha
sampled, and n~30 and 100 individuals per plot, respectively). Over 99% of all sampled stems were identified to species(specimens identified at the Museu Paraense Emílio Goeldi). Using height and dbh measurements, aboveground biomass for trees was
calculated with allometric equations developed for tropical moist forests with a distinct 1–4 month dry season with annual precipitation ranging between 1500 and 3500mmyr−1. For lianas, basal diameter was used to calculate biomass from a diameter-based allometric equation. To determine site-specific wood density, we obtained partial cross-sections (i.e., pith to epidermis) of the bole from the 10 most common tree species (n = 29 individuals). Wood specific gravity was calculated as the oven-dried mass (dried at 65 %C to constant weight) divided by the sampled cross-section volume. From these measurements, a site-derived estimate of the specific gravity of sound wood, 0.59±0.02 gcm−3 (s.e.), was incorporated in these allometric equations to determine aboveground live biomass.
Three annual experimental burns were conducted in either August or September (2004–6), near the end of the dry season, when many escaped wildfires typically occur. During all burns, mean daily temperature ranged 24–29 %C, and relative humidity
from 51% to 57% (measured at the meteorological station). During all years, understory wind speed was low (<0.5ms−1) with no observed effect on fire behavior. Fires were set with kerosene drip torches during 3–4 consecutive days between 9:00 and 16:00;
10km of fire lines were set per 50 ha plot. During all experimental burns, the majority of fires extinguished naturally by nightfall and were then re-lit on subsequent days. Combining both experimental burn plot treatments, initial mean flame heights and fire spread rates (±s.e.) were 31 (±1) cm and 0.21 (±0.01)mmin−1. However, fire intensity and spread declined significantly during the second and third burn. Compared with the
initial burn in B3, mean flame heights declined by ~10cm in subsequent burns, and burned area declined by half in the third fire.
For every sampled stem, the maximum char height and proportion of the stem base that burned were measured within a month following the experimental burns. To distinguish charring events in the B3 plot, char from previous years was marked with
a minimal amount of paint on each stem before each subsequent experimental fire. Char height is an assumed proxy for flame height, which corresponds to fireline intensity. However, the residual mark left by a passing fire line is not independent of individual and species-level variation in diameter, and bark properties, such as thickness. Additional sampling was conducted during a repeat burn in 2007 in the B1 plot. A subset of live trees (>20cm dbh) with visible fire scars (from the initial burn
3 years prior) were observed during the experimental fire (n = 26; sampling conducted within a 20×500m transect, 750-m into the forest). Immediately after the experimental fireline had passed, we recorded whether the fire-damaged wood near the bole base had
ignited.
All inventoried stems were censused annually immediately before the prescribed burns. To determine stem mortality, we applied a 0–5 categorical scale for assessing aboveground tissue:
0 = no visible aboveground live tissue; 5 = 100% of all visible above ground tissue was live. Only a few species were observed to be completely deciduous during the course of the study (e.g.,
Hymenaea courbaril L. (Fabaceae), Terminalia tetraphylla (Combretaceae), and Trattinnickia glaziovii (Burseraceae)), making this method reasonable. Stems were not cut to determine whether
inner wood was alive. A stem was also considered dead if only live basal sprouts were present. Using the visible aboveground tissue to determine stem death was considered to be the best nondamaging
method for multi-year assessments. Thus, we evaluate
aboveground stem death, rather than aboveground and belowground plant mortality.
Annual mortality rate (m) was calculated as:
m = 1 −(1 −D/N)^(1/t)
where N is the initial number of live stems, D= dead stems at the end of the measurement interval, and t = years between surveys. Here, we incorporate 3 years between the first and last census. Confidence limits (95% CI) on mortality rates were calculated based on the inverse F distribution. Mortality rates for individual species with >20 live stems per
plot before any treatment were included in species-level analysis.
