Forest data were gathered from circular 0.05 ha plots, 12.6 m in radius. Plot locations and centers were established randomly and permanently marked with a treated fence post. Every tree stem above breast height (1.4 m) was stem mapped and tagged. Species, diameter at breast height (dbh), crown class, crown ratio, damage type, treefall azimuth, sprouting response, and viability were also recorded. Readings of photosynthetically active radiation (PAR) were taken with a portable LI-COR, LI-191SA quantum sensor above and beneath the canopy at plot center and the periphery at principal compass positions. Light measurements at ground level were taken at a standardized height of 2 m oriented in both east-west and north-south positions. Global positioning receivers were used to georeference plot locations for mapping and hurricane simulation purposes.

Tree damage classes were established to distinguish damage extent between dead and severely damaged trees, moderately damaged, and lightly damaged. Uprooted individuals and boles snapped below the live crown were categorized as severely damaged and unlikely to survive. Individuals which sustained severe crown cropping and major branch loss were classified as moderately damaged. Many of these trees had sufficient bole and branch structure to resprout new leaves and were aptly described in the field as "chia" trees for their spindly epicormic regrowth and form. Lightly damaged trees included any normally foliated and defoliated trees with residual primary and secondary branch structure.

Figure 1. Map of south Florida mangrove distribution, location of permanent forest plots and storm track of Hurricane Andrew. [larger image]

Prehurricane stand conditions were reconstructed from the composite of surviving trees and individuals whose death and damaged condition was directly or indirectly attributed to Hurricane Andrew. In all cases, it was possible to distinguish the timing of death whether prior to the storm, during the storm, or subsequent to the storm. Trees that were dead prior to storm impact were excluded from any data summaries or analyses. Study sites were assigned to distance class and quadrant from storm center to test for differences in site response. All sites designated in the eyepath were less than 10 km from the central storm track. Right and left quadrant sites sampled in this study ranged in distance from 20 to 40 km outside the storm's ring of maximum winds.

Line transects were used to supplement fixed plot data to quantify the degree of canopy disturbance at control sites and additional site locations in the storm's right and left quadrants. PAR readings were taken at 10 m intervals along a 150 m transect and averaged for a mean site condition. Azimuths of hurricane-induced wind-fall of canopy trees intersecting the transect were also recorded.

Hurricane Simulation

A hurricane simulation model, HURACAN, was used to generate a chronology of predicted wind speeds and vectors for each plot location (DOYLE, 1994). This provided important environmental data on hurricane strength and position at each site over the course of Andrew's crossing and exiting the south Florida peninsula. A radius of maximum winds of 18 km (POWELL and HOUSTON, 1993) and an inflow angle of 45 degrees were used to compute surface wind speeds and vectors. National Weather Service tracking data of Andrew's latitude, longitude, date, hour, maximum sustained winds, central pressure, and forward speed were used to parameterize the model. Model output was generated every 15 minutes of Andrew's movement across the peninsula.

Statistical Analysis

Relationships between damage class with site position, tree species, and crown class were analyzed with an analysis of variance procedure for categorical variables. A Chi-square test was used to determine whether a significant relationship existed between damage condition and site, crown class, and species. A categorical modeling approach was used to derive an ordinally scaled response function for gauging the degree of association between response and explanatory variables. To satisfy the asymptotic normality requirements of the procedure, at least 25 to 30 samples were required for each explanatory level; no more than 20% of the categories had a sample size of less than 5. A Chi-square test was used to determine whether treefall azimuths were randomly distributed. Frequency counts of windthrow azimuths by site were assigned to a priori wind angle classes at 0-90, 90-180, 180-270, and 270-360 degrees. Damage response functions were established to compare differences between species, crown class, and site in relation to site-specific wind speeds and vectors projected with the HURACAN model. Correlation analyses were used to associate forest-damage statistics with prevailing wind conditions. Regression analysis was applied to determine whether mean tree fall azimuths and predicted vectors at peak wind intensity were equal to an expected slope of 1.