The Coral Reef Temperature Anomaly Database (CoRTAD) - Global, 4 km, Sea Surface Temperature and Related Thermal Stress Metrics for 1985-2005 (NCEI Accession 0044419)

[Text below adapted from: Selig, E. R., Casey, K. S., & Bruno, J. F. (2010). New insights into global patterns of ocean temperature anomalies: implications for coral reef health and management. Global Ecology and Biogeography, 19(3), 397–411. https://doi.org/10.1111/j.1466-8238.2009.00522.x. Hereafter referred to as "SCB2010".]

The CoRTAD was developed using data from the Pathfinder Version 5.0 collection produced by the National Oceanic and Atmospheric Administration's (NOAA) National Oceanographic Data Center (NODC) and the University of Miami's Rosenstiel School of Marine and Atmospheric Science. These sea surface temperature data are derived from the Advanced Very High Resolution Radiometer (AVHRR) sensor and are processed to a resolution of approximately 4.6 km at the equator. These data have the highest resolution covering the longest time period of any satellite-based ocean temperature dataset (see Figure 1 of SCB2010). Weekly averages of day and night data with a quality flag of 4 or better were used, which is a commonly accepted cutoff for "good" data (Kilpatrick et al., 2001, Casey and Cornillon, 1999). By using a day-night average, the number of missing pixels was reduced by 25% with virtually no loss in accuracy (see Table 1 of SCB2010).

The Pathfinder algorithm eliminates any observation with a Sea Surface Temperature (SST) more than 2 degrees C different from a relatively coarse resolution SST value based on the Reynolds Optimum Interpolation Sea Surface Temperature (OISST version 2.0) value, a long-term, in situ-based data set (Kilpatrick et al., 2001, Reynolds et al., 2002). Observations were added back into the analysis if the SST was greater than the OISST-5 degrees C, but less than the OISST+5 degrees C. The 5 degrees C threshold is a reasonable selection that allows diurnal warming events (Kawai and Wada, 2007) or other spatially limited warm spots back into the dataset without including unrealistic and erroneously warm values. Values less than the OISST were not included because they may have been biased by cloud contamination and other satellite errors, which tend to result in cooler SST estimates. These processes resulted in a dataset with only 21.2 percent missing data. To create a gap-free dataset for analysis, 3 x 3 pixel median spatial fill was used. A temporal fill was performed using the Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) function in Matlab (The Mathworks Inc., 2006) to fill the remaining gaps. This conservative approach was chosen because it provided interpolated SSTs that are bounded
by the nearest available values in time. It also used data from only a very limited spatial domain, which is an important consideration given the variability of coral reef environments.

Using these gap-filled data, we then created site-specific climatologies for each reef grid cell to describe long-term temperature patterns over the 21-year dataset (Eqn. 1). The climatology was generated using a harmonic analysis procedure that fits annual and semi-annual signals to the time series of weekly SSTs at each grid cell:

climSST(t) = A*cos(2pi*t + B) + C*cos(4pi*t + D) + E (1)

where t is time, A and B are coefficients representing the annual phase and amplitude, C and D are the semi-annual phase and amplitude, and E is the long-term temperature mean.

Similar approaches have been used for generating climatologies because they are more robust than simple averaging techniques, which can be more susceptible to data gaps from periods of cloudiness (Podesta et al., 1991, Mesias et al., 2007).

Sea surface temperatures from AVHRR quantify only the temperature of the 'skin' of the ocean, roughly the first 10 micrometers of the ocean surface (Donlon et al., 2007). Most field surveys of coral cover occur between 1 and 15 m depth. To be useful for coupling with coral reef biological data, these temperature data must be relatively accurate beyond the 'skin' of the ocean. Linear regression was used to examine how data from in situ reef temperature loggers compared with data from the CoRTAD to demonstrate the good accuracy of the CoRTAD temperature data compared to in situ data at a variety of depths and locations around the world (see Table 1 of SCB2010 for details).

Temperature anomaly metrics:

Several metrics could be used to link coral reef ecosystem health with temperature including trophic structure, diversity or percent coral cover (Newman et al., 2006, Roberts et al., 2002, Bruno and Selig, 2007). However, this analysis focused on coral bleaching and disease because they are key drivers of coral decline and their relationships with temperature patterns are better understood (Aronson and Precht, 2001, Bruno et al., 2007, Glynn, 1993). Analyses were performed on two metrics (see Table 2 of SCB2010): one that is commonly known to lead to bleaching (Liu et al., 2003, Strong et al., 2004, Glynn, 1993), and one that is correlated with increased disease severity (Selig et al., 2006, Bruno et al., 2007). Coral bleaching results when corals lose their symbiotic zooxanthellae (Glynn, 1993, Glynn, 1996). Bleaching is a natural stress response not only to warm temperatures, but also to cool temperatures (Hoegh-Guldberg and Fine, 2004) as well as light and salinity values different from the normal range (Glynn, 1993). Corals can recover from bleaching, but their ability to do so is dependent on the magnitude and duration of the anomaly event (Glynn, 1993). he temperature thresholds that result in coral bleaching vary by location and species (Berkelmans and Willis, 1999). Bleaching is often connected to Thermal Stress Anomalies (TSAs), which are defined as areas where temperatures exceed by 1 degree C or more the climatologically warmest week of the year (Table 2, Glynn, 1993). The temperature anomaly thresholds relevant to disease have been studied in only one pathogen-host system (Selig et al., 2006, Bruno et al., 2007). In that system, changes in disease cases were correlated with Weekly Sea Surface Temperature Anomalies (WSSTAs), temperatures that were 1 degree C greater than the weekly average for that location. The best metric for predicting bleaching or disease may vary according to location, species, and pathogen (Selig et al., 2006, Bruno et al., 2007, Berkelmans, 2002). For example, bleaching on the Great Barrier Reef was best predicted by the maximum anomaly over a 3 day period (Berkelmans et al., 2004), rather than an anomaly metric like the TSA. Although the 7-day averaging approach in the CoRTAD may be too temporally coarse to capture all bleaching events, it is necessary to maintain consistency and minimize gaps in the dataset across broad spatial scales. In addition, the data are less likely to yield false positives for TSAs and will likely capture most WSSTA events, which have a lower temperature threshold.

References: (see SCB2010 for complete list)