Decadal trends in Red Sea maximum surface temperature

Ocean warming is a major consequence of climate change, with the surface of the ocean having warmed by 0.11 °C decade−1 over the last 50 years and is estimated to continue to warm by an additional 0.6 – 2.0 °C before the end of the century1. However, there is considerable variability in the rates experienced by different ocean regions, so understanding regional trends is important to inform on possible stresses for marine organisms, particularly in warm seas where organisms may be already operating in the high end of their thermal tolerance. Although the Red Sea is one of the warmest ecosystems on earth, its historical warming trends and thermal evolution remain largely understudied. We characterized the Red Sea’s thermal regimes at the basin scale, with a focus on the spatial distribution and changes over time of sea surface temperature maxima, using remotely sensed sea surface temperature data from 1982 – 2015. The overall rate of warming for the Red Sea is 0.17 ± 0.07 °C decade−1, while the northern Red Sea is warming between 0.40 and 0.45 °C decade−1, all exceeding the global rate. Our findings show that the Red Sea is fast warming, which may in the future challenge its organisms and communities.

seawater temperatures are already extremely high. Yet, available analyses of thermal regimes in the Red Sea focus on annual mean values 18,19,24,25 , rather than the dynamics of maximum temperature. Here we characterize the variability in temperature maxima across the Red Sea and over time (1982 to 2015), based on daily values, identifying rates of change in annual maximum sea surface temperature, hereafter T max , as well as the distribution of anomalies, relative to T max over time.

Results
Warming rates and timing. The Red Sea displays a latitudinal gradient of increasing T max from north to south with the southern Red Sea exhibiting the highest T max (33 °C) until the southernmost Bab-el-Mandeb Strait (Fig. 1). The Gulf of Suez and the Gulf of Aqaba both exhibit lower temperatures than the open Red Sea (Fig. 1).
The northern Red Sea experiences T max throughout July while T max is reached between late July and mid-August in the southern Red Sea (Fig. 2). The area off of Al Lith, Saudi Arabia, prominently exhibits delayed T max from approximately mid August to early September (red area in Fig. 2).
We assessed the rate of change in the magnitude and timing of T max across the Red Sea. We observed a significant trend toward increased T max across the Red Sea, at an average rate of 0.17 ± 0.07 °C decade −1 (p = 0.02, Insert shows the latitudinal changes in mean (from 1982 to 2015) T max . Values based on daily temperature data. Image created using R (v3.3.1, www.R-project.org) 45 including packages: ggplot2 46 and rasterVis 47 , RStudio (v1.0.143, www.rstudio.com), and InkScape (v0.91, www.inkscape.org). df = 32, t = 2.437). Rates of change in T max varied across the Red Sea, with highest rates found in the colder areas of the Red Sea, including the northern Red Sea with rates for the Gulf of Suez and Gulf of Aqaba at 0.40 -0.45 °C decade −1 (Fig. 3a). The region experiencing the lowest rate of warming is, again, that exhibiting a delayed T max off the coast of Al Lith, Saudi Arabia (blue area in Fig. 3a).
In addition to a general pattern toward increasing T max , maximum temperatures in the Red Sea are also being reached earlier, with an average rate of change in the timing of T max of 0.19 ± 0.30 days earlier decade −1 (Fig. 3b). Most of the Red Sea experienced progressively earlier T max by 0.1 to 2 days earlier decade −1 , but a region in the southern Red Sea showed a delay in T max by 1 to 2 days decade −1 . This is the same region that exhibits anomalous trends in the annual timing of T max (Fig. 2).
Heat anomalies. Heat waves representing anomalies of 1.0 °C above the average T max were observed more frequently in the northern half of the Red Sea over the last 34 years. The majority of the basin experienced such anomalies during at least one year and up to 6 years (which may or may not have been successive years). Some areas in the northern Red Sea, including the Gulf of Aqaba, experienced 1.0 °C magnitude heat waves as often as 5 or 6 years over the 34 year period examined here (Fig. 4).
T max values 0.5 °C above the mean (1982 -2015) values occurred 15 to 24% of the years, whereas thermal anomalies involving T max values 0.75 °C above the mean values occurred 6 to 12% of the years, and years with T max values of 1.0 °C above the mean values occurred with a probability <6% (Fig. 5). The decline in the frequency of T max anomalies with increasing magnitude of anomalies was significant (Kruskal-Wallis, p < 2.2 e −16 , chi-squared = 2674, df = 4, Fig. 5) and significant differences were found among all groups (Dunn's, p < 0.05, Z range = [4:44]).

Discussion
The latitudinal gradient of increasing T max from north to south in the Red Sea is largely a consequence of the variation in solar radiation associated with these latitudinal differences, and is consistent with previous studies reporting the same trend based on mean temperatures, with the warmest thermal regime in the southern region 19 . The Gulf of Suez and the Gulf of Aqaba have colder thermal regimes. Previous studies reported that, in the summer, the surface water entering the Gulf of Aqaba from the Red Sea is about 2 °C warmer than the water inside the Gulf 26 .
The Red Sea basin presents a discontinuity in terms of the timing of T max , associated with an abrupt transition between 20 and 22 °N. The timing of T max occurs two months earlier south of this boundary compared to the timing north of this boundary. The distinct break between North and South ( Fig. 2), may be evidence for the strong coupling of wind and sea surface temperatures over the basin as in other ocean systems [27][28][29] . During winter (October-April), the basin experiences opposing southward and northward winds, converging at about the same belt between 19 -20 °N 19 where the divide in timing of T max is observed. From May to September, the major wind vector is from north to south 19 .
The warming rate of the Red Sea, 0.17 ± 0.07 °C decade −1 , is higher than the global ocean rate of 0.11 °C decade −1 1 . The northern Red Sea is warming faster with the Gulf of Suez and Gulf of Aqaba (0.40 -0.45 °C decade −1 ) (Fig. 3a) warming four times faster than the mean global ocean warming rate. The semi-enclosed nature of the two gulfs as well as that of the Red Sea as a whole may account for the intense warming 17,30,31 , while the slower rate of increase in the southern Red Sea may be buffered by its closer connection to the Indian Ocean. Although the northern Red Sea is warming faster, it remains the coolest region in the basin throughout the year.
Increased T max will have effects on marine biota, which are particularly vulnerable to heat waves, when their thermal limits may be approached or exceeded 23,32 . The occurrence of heat anomalies, which are also likely to increase in the future 1 , are greatly relevant to the physiology of organisms, particularly for those inhabiting already warm environments, like the Red Sea, where temperature anomalies may lead to thermal collapse 24, 32-34 . The years 1999 and 2001 experienced the largest anomalies across the basin (Fig. 6). During the years 1997 -1998, one of the strongest El Niño events occurred, while 2000 -2001 was considered a weak La Niña event 35 . The years 2003 and 2015, also El Niño years, showed the second greatest percentage of area covered by T max anomalies, although of a relatively small, 0.5 °C, magnitude (Fig. 6).
Systematic monitoring efforts are required to detect the effect of heat anomalies on marine organisms, such as bleaching and mass mortality events 36 . Unfortunately, there is no systematic monitoring of biological events in the Red Sea, such as bleaching events, which may be affected by thermal anomalies such as those reported here. Extensive bleaching was reported in the southern half of the Red Sea in 2015, one of the years with extensive, but relatively moderate, thermal anomalies in our analysis (Fig. 6). Whether bleaching events also occurred in other years with extensive T max anomalies is unknown due to lack of long-term monitoring.
The distribution of T max in the Red Sea conforms to the four provinces, described by Raitsos et al. 19 based on phytoplankton biomass. The warmer T max regime in the South is associated with higher phytoplankton biomass, while the lowest T max in the northern Red Sea is associated with the lowest phytoplankton biomass. However, this pattern may be a result of the decrease in nutrient concentrations from south to north along the Red Sea 37 , rather than its thermal regime. A region in the central Red Sea emerges as deviating from the general pattern with a slower rate of warming and T max reached later in the year over time.
That T max is rapidly increasing in the Red Sea, which is already one of the warmest seas, anticipates challenges to biota. Whereas T max is increasing more rapidly in the North than in the South, the warmer thermal regime in the South may already be near the thermal limits of organisms and, therefore, even a modest increase in T max may suffice to exceed their thermal tolerance, although experimental work is necessary to test this suggestion. Unfortunately, although the Red Sea ranks as the warmest sea on the planet, aside from one study examining the effect of temperature on grazing rates of Red Sea parrotfish 38 , there is, at present, no quantitative information on the thermal limits of Red Sea biota. However, reports of a decline in coral growth and calcification across the thermal range of Red Sea corals 39 , together with widespread bleaching in the southern half of the Red Sea during 2015, as well as lower growth rates reported for brown macroalgae 40 , suggests that warm Red Sea temperatures already challenge the capacities of organisms. In addition to increasing T max , the general tendency towards an earlier occurrence indicates that phenology patterns of organisms might need to adjust to this shift. Marine organisms generally cope with warming by shifting their biogeographical range poleward tracking the migration of isotherms 2, 14 . However, this strategy is not possible in semi-enclosed seas, such as the Red Sea 14, 15 , rendering its large pool of endemic species at risk of extinction unless they become Lessepsian migrants and colonize the Mediterranean Sea as a hundred Red Sea species have done 41 . Altogether, higher and earlier T max may challenge the capacities of Red Sea biota to cope.
Results presented here provide a context for experimental analyses examining thermal limits, by defining the regimes and trends in T max across the Red Sea, as well as the likelihood of observing anomalies of different magnitudes. In addition, these results may help understand biodiversity patterns and losses across natural gradients in the Red Sea by matching the distribution of communities and habitats with the distribution of T max . This will provide an underpinning to the assessment thermal maxima play in explaining patterns of biodiversity across the Red Sea.
In conclusion, Red Sea biota are exposed to increased ocean warming, particularly in the northern Red Sea, which may affect their future persistence, especially if unable to migrate into the Mediterranean. The results on Red Sea warming presented here, coupled with experimental evidence on the thermal limits of Red Sea organisms, yet to be resolved, would provide a powerful tool to predict the future of marine biodiversity in this biodiversity hotspot containing a high degree of endemism.

Methods
The dataset. We used remotely sensed sea surface temperature (SST, °C) data to examine maximum temperatures on a basin-wide scale across the Red Sea. The AVHRR-OI (Advanced Very High Resolution Radiometer-Optimum Interpolation) Pathfinder sensor currently provides the longest continuous daily dataset of infrared SST from 1981 to present 42 , allowing the assessment of decadal trends of temperatures. Whereas other sensors provide higher resolution, in terms of pixel size, they encompass a period too short to be climatically-relevant as yet (ERS-1/ATSR-1 and Acqua/AMSR-E) 43 and do not allow us to identify, with confidence, the maximum temperature achieved over time. A daily Level-4, gap-free dataset merging day and night analysis AVHRR SST was obtained from NASA's (National Aeronautics and Space Administration) National Climatic Data Center 44 at podaac.jpl.nasa.gov accessed on January 5, 2016 encompassing 34 years over the period 1982 to 2015. This dataset has been optimally interpolated and mapped on a 0.25° × 0.25° grid. The values in the dataset were corrected with in situ measurements from buoys and ships 42 . Daily fluctuations in daily SST time series may significantly affect the measurement of maximum SST phenology and magnitude, because the recurrence of the passage of AVHRR Pathfinder is 2 to 3 days and, the time of passage may not match the time of T max , typically found in the late afternoon with a daily range in T max , derived from moorings in the central Red Sea, of up to 3 °C. Moreover, the individual estimates may be affected by dust, which is prevalent in the region at the time of T max , and cloud cover. Whereas the data we used is interpolated, the individual daily values may be affected by the sources of error above, leading to underestimates of the actual T max . To attenuate this source of error, we extracted the maximum daily T value within sets of interpolated daily values over 8-day periods, and then selected, for each of the 669 pixels, the highest T observed in any one year as that providing the best estimate of T max for that pixel and year. The dataset can be downloaded from the Pangea open-access data repository (Chaidez et al. 2017) 48 .
Calculating decadal trends. The decadal trends of maximum temperatures and time of occurrence were estimated by fitting a linear regression relating T max to year for each of the pixel's yearly time series. The slopes of the fitted linear regressions provide an estimate of the rates of change for each pixel in the Red Sea (units: °C decade −1 , and days decade −1 , respectively). We tested the possible occurrence of autocorrelation in T max among years, and found, for a sample of pixels, no evidence of autocorrelation, i.e. the T max in any one year is independent of T max in preceding year(s).

Calculating heat anomalies.
For each pixel, a reference maximum temperature was computed by taking the mean of the highest temperatures per year over the study period. A heat wave event was defined as a yearly maximum temperature above the reference maximum temperature by a given threshold chosen at 0.5 °C intervals between 0.5 and 1.5 °C. The number of heat wave events over the 34 years were counted for each pixel, as well as the area of the Red Sea experiencing heat waves of various magnitudes in a given year. A Kruskal-Wallis test followed by Dunn's test for multiple comparisons, was used to compare the frequencies of occurrence for all magnitudes of heat anomalies in Fig. 5. The percentage of area in Fig. 6 was calculated as the percentage of pixels. We are aware that the area of each pixel depends on latitude, as the length of a degree longitude varies with latitude. However, for the narrow range of latitude covered by the Red Sea, the difference is minimal, so percent of pixels and area are essentially equivalent.
Data Availability. The data set supporting the analysis presented here can be found in the Pangaea open data repository: (Chaidez et al. 2017, http://www.pangaea.de) 48 .