Supplementary material for
“Temporal clustering of tropical cyclones on the Great Barrier Reef and its ecological importance”
Model reef locations
The Great Barrier Reef Marine Park Authority (GBRMPA) reef polygon layer (Fig. S1a) was intersected with a 4 x 4 km polygon grid in GIS. The centroid of all grid polygons that had reef habitat within was used to run reef-scale analysis of cyclone impacts in this study (Fig. S1b). At each location, the ecological model was used to determine reef trajectories which incorporated both cyclone disturbances and subsequent recovery.
Figure S1. Comparison of the Great Barrier Reef Marine Park Authority (GBRMPA) reef polygon layer (a) and the 1,312 representative reef locations used to model reef-scale cyclone impacts (b) in this study. The GBRMPA reef polygon layer was provided by the Spatial Data Centre, Great Barrier Reef Marine Park Authority, 2012.
Synthetic cyclone tracks
Synthetic cyclone tracks generated using process-based models driven by general circulation models (Emanuel et al. 2008) we used to assess reef-scale vulnerability (Fig. S2). Cyclone tracks were generated for 20th century environmental conditions using four IPCC AR5 general circulation models (HADGEM, GFDL, MIROC, ECHAM) downscaled to drive a tropical cyclone model (Emanuel et al. 2008).
Fig. S2 7,092 synthetic cyclone tracks (Emanuel et al. 2008) used to examine reef-scale vulnerability across the Great Barrier Reef. The Queensland, Australia coastline (black) and the boundary for the Great Barrier Reef Marine Park (grey) are also shown. Each portion of a track is coloured according to the Australian Bureau of Meteorology (BoM) cyclone category
Dispersion effects by cyclone category
Fig. S3 Differences in mean Acropora cover (%) from model simulations for different cyclone rates (x-axis) and dispersion characteristics (legend) for all five storm categories. Dispersion statistics represent values found across the Great Barrier Reef. Standard deviations for the mean values across 100 simulations are also shown. Each simulation was captured as a mean across a 100 year time-series.
Gap-filling of missing central pressure data for observed cyclone positions
Cyclone tracks for the Great Barrier Reef were acquired from the BoM for the period 1906 to 2011 (Table S1). Cyclone positions were recorded by BoM every 6–24 hr along with maximum wind speed, central pressure, location (latitude and longitude) and time attributes. Pre-processing included some gap-filling of data (Table S1). Where only central pressure was missing, the central pressure of the previous entry from the same cyclone was used. If a pressure was missing between two known pressures, the pressure was interpolated. Given a strong linear relationship between observed central pressure (P, hPa) and maximum wind speed (MWS, m s-1), missing wind data was estimated using the following relationship:
R2 = 0.93; MWS = -0.579P +593.63
The maximum wind speed was used to determine the category of the cyclone at every time step (generally every 6 hr) and cyclone tracks were further subdivided by these categories. Position data were converted to polylines in GIS for use in this analysis.
Although pressure data was frequently missing from the decades up to the 1940s, the proportion of those positions determined to be category 1 or greater (56%) was on average similar to the proportion found after the 1940s (53%) when missing data was far less of an issue. This suggests our gap-filling method was a reasonable solution. We used this gap-filling solution because we were concerned that the assumption that all positions with missing pressure data represented at least category 1 storms would result in an overestimate of storm frequency.
Although we present rate and dispersion results for both the 1906–2011 and 1970–2011 time periods, only the post-1970s results were used for our reef-scale analysis.
Table S1 Summary of the gap-filling performed on the BoM tropical cyclone database (http://www.bom.gov.au/cyclone/history/ downloaded October, 2013). Shown per decade are the numbers of storm positions, the percent of pressure data missing for those positions (null pressure), the percent of wind data missing (null wind) and the percent of positions that reached category 1 winds (≥18 m/s) or greater. Years included are 1906–2011.
Decade / Positions / Null pressure / Null wind / Cyclones1900s / 18 / 0 / 100 / 33
1910s / 163 / 85 / 100 / 73
1920s / 222 / 81 / 100 / 68
1930s / 151 / 87 / 100 / 78
1940s / 275 / 30 / 100 / 28
1950s / 444 / 13 / 100 / 50
1960s / 474 / 1 / 100 / 20
1970s / 1423 / 4 / 100 / 57
1980s / 1076 / 0 / 83 / 60
1990s / 754 / 1 / 4 / 85
2000s / 702 / 0 / 7 / 66
2010s / 291 / 4 / 19 / 32
Cyclone track buffering
A buffer was applied to each category-track using wind speed extents defined by Keim et al. (2007) following a method described in detail in Edwards et al. (2011). After adjusting for differences between category classification systems (BoM versus Saffir–Simpson), asymmetric buffers were applied to individual cyclone tracks according to the maximum sustained wind (Table S2). Buffers were applied to both the observed BoM tracks for the 1970–2011 period and the synthetic tracks. These buffered polygons were intersected with model reef locations for both data sets. The results for both methods were strikingly similar: BoM data revealed that 64% of the reef impacts were category 1; 32% were category 2 and 3; 4% were category 4 and 5. Synthetic tracks demonstrated 63% of the reef impacts were category 1; 33% were category 2 and 3; 4% were category 4 and 5. This gave us confidence that the synthetic tracks captured observed category distributions and needed no adjusting for our reef-scale vulnerability analysis.
For the reef-scale modelling of cyclone impacts, a probability distribution of the synthetic tracks was used to assign whether a category 4 and 5 buffer was a category 5 or 4 and whether a category 2 and 3 was a category 2 or 3. In both cases, synthetic tracks revealed that the less severe category in each grouping (2 and 4) occured approximately twice as often as the more severe (3 and 5). Thus, when a reef was impacted by a category 4 and 5 buffer, there was 66.6% chance in our model that it would experience category 4-related coral damage and a 33.3% chance it would experience category 5 damage. This same distribution was applied to the category 2 and 3 buffers. In summary, although the buffer approach allowed us to capture extents for only three cyclone category groupings (Table S2), we modelled the impacts of all five categories on the trajectories of coral cover.
Table S2 Extent (km) of storm force winds applied to cyclone tracks. Shown are the total extent of the asymmetric buffers (greater extent to the left of the storm) from the cyclone eye, and the distance range of each category of winds.
Cyclone category / Total extent / Category 4 and 5 / Category 2 and 3 / Category 1Left / Right / Left / Right / Left / Right / Left / Right
Category 4 and 5 / 250 / 125 / 0–90 / 0–45 / 90–170 / 45–85 / 170–250 / 85–125
Category 2 and 3 / 160 / 80 / --- / --- / 0–80 / 0–40 / 80–160 / 40–80
Category 1 / 80 / 40 / --- / --- / --- / --- / 0–80 / 0–40
References
Edwards HJ, Elliott IA, Eakin CM, Irikawa A, Madin JS, McField M, Morgan JA, Van Woesik R, Mumby PJ (2011) How much time can herbivore protection buy for coral reefs under realistic regimes of hurricanes and coral bleaching? Glob Chang Biol 17:2033–2048
Emanuel K, Sundararajan R, Williams J (2008) Hurricanes and global warming: results from downscaling IPCC AR4 simulations. B Am Meteorol Soc 89:347–367
Keim BD, Muller RA, Stone GW (2007) Spatiotemporal patterns and return periods of tropical storm and hurricane strikes from Texas to Maine. J Clim 20:3498–3509