Monitoring began in 1992 using six sites selected at random to characterize the reef community using photoquadrats (0.5×0.5 m). For the present analysis, photoquadrats were pooled among sites to describe the abundance of Millepora spp. on shallow fringing reefs.

At each site, photoquadrats were scattered randomly along a single transect parallel to the 7-9 m depth contour. Prior to 2000, photoquadrats were recorded using Kodachrome 64 film and a Nikonos V camera fitted with a 28-mm lens and strobes. The camera was mounted on a frame that held it perpendicular to the reef and was used to record ~18 photoquadrats along a 20-m transect. Starting in 2000, digital photography (with strobes and a framer) was introduced, first with a 3.3 megapixel camera (Nikon Coolpix 990), and from 2007, a 6.1 megapixel camera (Nikon D70). Digital photography allowed the sample size to increase to 40 photoquadrats site-1, with the additional photoquadrats scattered along a 20 m extension to the original transect. Both photographic techniques produced images in which objects >=10 mm diameter could be resolved, and the annual surveys (pooled among sites) provided ~102 photoquadrats y-1 prior to 2000, with 210-222 photoquadrats y-1 thereafter. Photoquadrats were recorded between July and August in all years except for 1992, 1993 and 1995-1997 when they were recorded between May and June. Slides were digitally scanned (at 3200 dpi) and together with digital images are archived at mcr.lternet.edu/vinp/data/.

Millepora spp. abundance
The population dynamics of Millepora spp. were quantified using three measures of abundance. Percent cover, colony size (planar area of colonies entirely within the photoquadrat), colonies and branches within each quadrat were counted to evaluate population size (number of colonies). Colonies were counted if they were entirely within the photoquadrat, or if present as encrusting bases located partially within the photoquadrat. These criteria overestimated population sizewhen colonies grewwithmultiple encrusting fronts that separately spread into the photoquadrats.

Temperature and storm intensity
To gain insight into the role of environmental factors in mediating changes in Millepora spp. populations, the associations between abundance and seawater temperature were evaluated using Pearson correlations with census years as replicates. Associations between abundance and storm intensity were evaluated using Spearman correlations as the intensity of storms was evaluated on a categorical scale.

Seawater temperature in Great Lameshur Bay was recorded using a Ryan Industries thermistor (±0.3 °C accuracy) at 11-m depth from January 1992 to April 1997, and from November 1997 to August 1999; an Optic Stowaway logger (±0.2 °C accuracy) at 9-m depth from May 1997 to October 1997, and from August 1999 to August 2001; and a Hobo Aquapro logger (±0.2 °C accuracy) at 9-m depth from August 2001 to August 2008. Loggers recorded temperature every 15-30 min. Temperature was averaged by day and used to calculate a mean for the ~12 months between samplings. Daily temperatures were used to categorize days as hot (>29.3 °C) or cold (<=26.0 °C), and the number of hot and cold days in each year was used to evaluate the association between thermally extreme days and Millepora spp. abundance. The temperature defining "hot days" was determined by the coral bleaching threshold for St. John (http://www.coral.noaa.gov/research/climate-change/coral-bleaching. html), and the temperature defining "cold days" was taken as 26.0 °C which marks the lower 12th percentile of all daily temperatures between 1989 and 2005 (Edmunds, 2006).

To analyze the impacts of storms on Millepora spp., storms occurring between sampling intervals were ranked by their potential damaging effects, and the ranks summed over each sampling year to assess the annual impact on benthic taxa. The potential impacts of storms were evaluated from their greatest wind speeds on St. John, which were used as a proxy for the size ofwaves resulting from the closest passage to the south of the island. Wind speeds were used to rank storms on a four-step scale: 1 <= 25 km h-1, 25 kmh-1 < 2 <= 50 km h-1, 50 km h-1 < 3 <= 75 km h-1, and 4 > 75 km h-1.

Wind speeds in St. John were estimated using summaries of Atlantic hurricane seasons (http://www.nhc.noaa.gov/pastall.shtml), which provided the maximum wind speed of each storm at its closest proximity to St. John, and an exponential function to predict the extent to which the maximum wind speed decayed by the time it impacted the island. The exponential function had the form Sd=Sm e^(-lambda*d) where Sd is the local wind speed, Sm is the maximum wind speed at the closest distance (d) to the south coast of St. John, and lambda a constant. lambda was determined empirically for six storms (Hortense, Georges, Lenny, Jose, Debby, and Earl) for which wind speed was recorded at the Cyril E. King Airport, St. Thomas, 25-km west of Lameshur Bay; wind speed was best predicted with lambda=0.016 (r2=0.659, n=6). Wind speeds on the south coast of St. John resulting from the close passage of major storms were therefore predicted using Sd=Sme^(-0.016*d).

coral_temps_storms.csv