Average gradient characteristics

We found that average slopes vary more than 3-fold among the world's coastal latitudinal temperature gradients (Fig.3, Table 1). The steepest CLTGs are found along the east coasts of the North-American (NAmAtl  =  −0.91°C lat⁻¹) and Asian continents (As_Pac  =  −0.60°C lat⁻¹), while the weakest CLTGs occur along the South-African Indian Ocean (Af_Ind(1)  =  −0.28°C lat⁻¹) and the South- and North-American Pacific coasts (−0.29°C lat⁻¹ and −0.32°C lat⁻¹, respectively). CLTGs of intermediate strength include the European, South-American and South-African Atlantic gradients (−0.34°C lat⁻¹, −0.37°C lat⁻¹ and −0.42°C lat⁻¹, respectively), as well as the Australian Indian Ocean gradient (−0.46°C lat⁻¹). Some CLTGs could reasonably be described by linear relationships across their entire latitudinal range (NAmPac, EuAtl, AuInd, AuPac), but most contained regions where the latitudinal SST decline differed notably from the overall average (Fig.3). Along the NAmAtl, for example, SST decreases with −2.55°C lat⁻¹ between 33–39°N (by far the steepest CLTG in the world), which is followed to the north by a rate of −0.76°C lat⁻¹ (40–57°N). Similarly, sections of the SAmPac and the SAfAtl gradients have slopes twice as strong as their entire range average (SAmPac(1) 3°N-8°S  =  −0.82°C lat⁻¹; SAfAtl(1) 12-23°S  =  −0.98°C lat⁻¹). Furthermore, the partial SAmPac(1) comprises the world's strongest thermal decline in tropical latitudes (Table 1). In contrast, average SST along the African east coast (AfInd) is almost invariant along a more than 4,000 km long stretch of tropical coastline (SST23_°N-15_°S  = 26.0−27.6°C), followed southward by the weak AfInd(1) gradient (15–34°S: −0.28°C lat⁻¹). Thus, partial CLTGs are even more variable than entire CLTGs and span almost a magnitude of slope differences (Table 1, Fig.3).

Average seasonality (SST_max-min) also varies strongly between CLTGsand latitudes (Fig.3). Lowest values generally occur in tropical and subpolar regions (<8°C), while seasonality maxima are located at subtropical and temperate latitudes (25–55°∶10–26°C), with particularly high values in northern hemisphere gradients (NAm_Atl  = 24.4°C_38.5°N; As_Pac  = 25.6°C_39.5°N). The seasonality peak of the Eu_Atl gradient occurs at comparatively high latitudes (15.9°C_53.5°N), whereas the NAm_Pac gradient shows no peak at all but a depression at mid-latitudes (4.1°C_39.5°N) consistent with the Californian coastal upwelling system. Similar seasonality depressions are also apparent along two other major coastal upwelling areas, i.e., Canary (NAf_Atl) and Benguela coastal upwelling regions (SAf_Atl)(Fig.3).

Spatio-temporal changes in gradient strength & seasonality

Temporal SST trends differed markedly between gradients, latitudes, and seasons of the year, thereby causing complex decadal changes in most CLTGs (Table 1, Fig.3). Out of 11 CLTGs, seven have weakened over the study period (i.e., their absolute latitudinal SST slopes decreased), while four have become stronger (i.e., slopes increased). All northern hemisphere CLTGs except the NAf_Atl, have decreased in strength by 3–10%, in particular the North-American Atlantic and Pacific gradients (NAm_Atl: −9.6%, p<0.001; NAm_Pac: −8.1%, p = 0.041), the European Atlantic gradient (Eu_Atl: −5.3%, p = 0.057), and the Asian Pacific gradient (As_Pac: −3.2%, p = 0.086). In the southern hemisphere, both Australian gradients did not show statistically significant changes (Au_Ind: −5.7%, p = 0.226 Au_Pac: −4.6%, p = 0.173), whereas the South-African Indian Ocean gradient has lost 7% of its strength over the past 31 years (Af_Ind(1) p = 0.022). In contrast, significant increases in gradient strength were detected for the South-African and South-American Atlantic gradients (SAf_Atl: +6.2%, p = 0.030; SAm_Atl: +3.3%, p = 0.007). The steep latitudinal decline in tropical SSTs along the South-American Pacific gradient has strengthened further over the past three decades (SAm_Pac(1): +22%, p = 0.016).

Overall, two thirds of the examined 433 coastal cells showed positive SST trends over time (n = 286), approximately half of which were statistically significant (p<0.05, n = 141), while 147 cells had negative SST trends, 55 of which were significant (Fig.3). Regression slopes ranged from −0.07°C year⁻¹ (NAm_{Atl 31.5°N}) to +0.09°C year⁻¹ (SAm_{Atl 35.5°S}) corresponding to a 31 year change in local SST of −2.21°C to +2.67°C, respectively. Similarly, trends towards more extreme seasons (i.e., increasing annual SST_max-SST_min) were found at more than three quarters of all latitudes (n = 331, p<0.05 for 113 cells).

Patterns & changes of coastal latitudinal temperature gradients

Latitudinal temperature patterns along coastlines are much more heterogeneous than in the open ocean [8], [31], [32], likely because of regional differences in well-understood oceanographic and climatic features such as alongshore surface currents, upwelling, thermocline depth, wind patterns, and cloudiness [7]. The world's strongest coastal gradients (i.e., NAm_Atl and As_Pac), for example, occur along the western boundaries of the Northern Hemisphere Atlantic and Pacific Oceans, where opposing warm and cold alongshore currents transport subarctic waters south- and tropical waters northward (NAm_Atl: Florida/Labrador Currents, As_Pac: Kuroshio/Oyashio Currents). The European gradient's (Eu_Atl) elevated thermal profile [7], [33], with coastal latitudes being on average 5–10°C warmer than the same latitudes along the NAm_Atl or As_Pac, is consistent with the influence of the North-Atlantic Drift, i.e., the warm continuation of the Gulf Stream past its main branching point. Similarly, the world's weakest gradient was found along the South-African east coast (Af_Ind(1) −0.28°C lat⁻¹), where another western boundary current - the warm Agulhas Current - transports heat southward [34]. Coastal upwelling, on the other hand, appears to be recognizable by distinct temperature depressions (‘troughs’) that cause substantially different regional gradient slopes. This signature pattern appeared in all four major coastal upwelling regions (Fig.3), i.e., (i) the California Current upwelling system (NAm_Pac), (ii) the Benguela Current upwelling system (SAf_Atl), (iii) the Canary Current upwelling region (NAf_Atl), and (iv) Humboldt Current upwelling system off Peru and Chile (SAm_Pac). In the latter, the observed depression in coastal temperatures (5–17°S) result in a uniquely steep tropical SST decline with latitude (SAm_{Pac(1) 3°N-8°S}  =  −0.82°C lat⁻¹), which sets a potentially great natural contrast for large-scale ecological comparisons.

Temporal trends in CLTG patterns likely represent the complex net result of changes in atmospheric and oceanographic features [1], [35], both of which are expected to differ between regions. However, we found that the NAm_Atl, Eu_Atl and As_Pac gradients share a common Northern Hemisphere warming pattern [21], [30] that involves intensified warming at high compared to lower latitudes, generally as a consequence of precipitous SST increases during fall and winter (wks 35–11) that are partially offset by cooling trends in spring and summer (wks 12–34) [6]. This has caused the weakening in overall gradient strengths while increasing seasonality across latitudes. The latter is consistent with Lima & Wethey's [6] finding that the incidence of extreme SSTs (warm and cold) has increased in most coastal waters. However, common patterns may be evoked by different mechanisms [21], such as increased heat transport at high latitudes, stronger stratification due to warmer freshwater run-off [21] or changing ocean currents [36], [37]. For example, the NAm_Atl weakened not only because of high latitude warming, but also because of strong cooling trends in coastal waters between 26–35°N [6]. SSTs in coastal waters between the tip of Florida (25°N) and Cape Hatteras (35°N) are influenced by the warm Florida Current, i.e., the beginning of the Gulf Stream [38]. Cooling of this coastal section is intriguing [39], given that it might signal an overall decrease in the Florida Current heat transport or a progressive diversion of warm Florida Current waters from the coast, both of which have been suggested by Ezer et al. [37] in a recent analysis. These SST changes have been substantial and therefore likely had profound ecological consequences for this region. Another example can be found along the Australian East coast (Au_Pac), where the warm East Australian Current has been shown to extend further and further southward [40], which is consistent with the significant warming trends observed between 36–39°S along the Au_Pac (mainly an austral fall and winter phenomenon) [6].

In general, SST trends at individual latitudes were consistent with those reported in studies of regional coastal temperature change [21]. For example, Nixon et al.'s [41] analysis of a 117-year SST record from the pier at Woods Hole, Cape Cod (located at ∼41.5°N on the North-American Atlantic coast) showed average warming rates of 0.04°C year⁻¹ between 1970–2002, which is comparable to rates estimated here (NAm_{Atl, 41.5°N}  = 0.02°C year⁻¹ and NAm_{Atl, 42.5°N}  = 0.03°C year⁻¹). Similarly, Gomez-Gesteira et al. [7] found that SST in coastal European waters between 37–48°N increased at annual rates of 0.012 to 0.035°C year⁻¹ (1985–2005), which is consistent with the present estimates of 0.014 (Eu_{Atl, 37.5°N}) to 0.034°C year⁻¹ (Eu_{Atl, 49.5°N}).

Coastal upwelling has been predicted to intensify as a result of increasing wind-stress linked to global climate change . [1], [35], [42], [ but see 43]. Our analyses support this prediction by identifying a second remarkable pattern shared by some coastal latitudinal SST gradients: cooling [6]. This has been most evident along the South-American Pacific coast, where all latitudes between 6–47°S but particularly those between 6–14°S have consistently cooled (average of −1.58°C in 31 years), which is spatially consistent with the Humboldt Current upwelling system [44]. In addition, cooling affected all seasons of the year, but was most intense during austral spring and summer (wks 35–5). Hence, the substantial strengthening (+22%) of the partial SAm_Pac(1) gradient, i.e., the unique decline in tropical SSTs along the South-American Pacific coast, may have largely been a consequence of upwelling intensification [21]. Similar patterns, albeit weaker, were also detected for the region of the NAm_Pac gradient corresponding to the California Current upwelling system, for which upwelling has been examined in more detail [21], [45], [46]. Intensification of the South African Benguela upwelling system, as suggested by Santos et al. [32] is consistent with the cooling patterns we detected at the southern edge of the SAf_Atl gradient. It may also explain why coastal latitudes of the Canary Upwelling system (22–30°N) showed no significant SST trends, contrary to substantial warming observed at all other coastal cells of the NAf_Atl gradient [6], [31]. Conversely, several studies have detected a weakening of upwelling intensity along the western Iberian coast [47], which corresponds to small but significant warming trends along the lower latitudes of the Eu_Atl gradient in this study. In general, decade-scale cooling had likely as severe, if opposite, ecological consequences for marine life as decadal warming.

We conclude that the complex and regionally heterogeneous patterns imply that direct heat transfer from the atmosphere to ocean (i.e., global warming) cannot by itself explain the observed changes in CLTGs. Regional scale, wind-driven, indirect effects, via heat fluxes and upwelling, must drive changes in both the magnitude and seasonality of coastal SST gradients.

Ecological implications

Given temperature's fundamental role in governing marine life, long-term changes in CLTGs will likely precipitate shifts in ecological gradients. Adaptive latitudinal variations in intra-specific morphological, physiological and life history traits are well documented in marine taxa [reviewed in 12] and encompass traits such as body size [10], [11], growth and development rates [48]–[50], predator vulnerability [51], life span [52], or vertebral number [16], [17]. For example, adapting to decreasing temperatures with latitude, many estuarine fish populations from high latitudes have evolved higher genetic growth capacities than their lower latitude conspecifics, a form of genotype × environment interaction known as countergradient variation [53]–[55]. Baumann and Conover [56] found that the strength of countergradient growth variation in silverside fishes from the North-American Pacific versus Atlantic coasts mimicked the strength of their contrasting latitudinal SST gradients. Similarly, increases in vertebral number with latitude, a form of co-gradient variation [12], is twice as strong in silverside fishes along the North-American Atlantic (strong CLTG) versus the Pacific coast [weak CLTG, 57]. Hence, latitudinal temperature gradients likely determine genotype*environment interactions in coastal systems, and shifts e.g., towards weaker gradients may precipitate weaker genetic gradients by reducing the amount of genetic variance between populations.

These adaptive trait gradients along CLTGs may comprise useful spatial analogies for studying climate change adaptation [12], [56]. How organisms will adapt to global climate change is uncertain, partly because of difficulties besetting the methods to study it (e.g., duration, replication, plastic vs. genetic responses). In contrast, adaptation principles can be rigorously examined across spatial climate gradients of different strength, such as the world's CLTGs. If countergradient variation, for example, is as ubiquitous an adaptation principle in time as it is in space, global warming may not lead to oft-anticipated increases in observed growth rates among ectotherms [1], [12]. Instead, genetic decreases in growth capacity could offset the plastic, temperature-driven growth increases, analogous to the local growth adaptation in high versus low latitude fish populations [56].

CLTGs may also underlie the ubiquitous decline in species diversity from low to high latitudes [9], [58], [59]. Whether coastal diversity actually scales with the strength of the underlying temperature gradient has received less attention, but could be effectively tested using the full and partial thermal slopes described here. Paleo-biological analyses indicated that diversity-temperature relationships in planktic foraminifers have been surprisingly stable over the past three million years, suggesting that gradients in species richness rapidly reorganize in response to climatic changes [58]. Hence, weakening thermal gradients due to large-scale latitudinal temperature shifts could precipitate weaker diversity gradients, by increasing diversity at high latitudes and/or decreasing diversity at low latitudes [e.g., 60].

One already widely observed biotic reaction to global climate change [60]–[64], is the shift in geographic ranges in marine species, with the vast majority comprising poleward movements [e.g., 61: 75% of 129 marine species]. This is consistent with our observation of weakening coastal latitudinal temperature gradients, not only because high latitude warming increases potential habitat for lower latitude organisms, but also because decreases in thermal gradient strength would make habitats more similar across latitudes and thus even more suitable for some intermediately located genotypes. In the North Atlantic Ocean, for example, where both coastal temperature gradients (Eu_Atl, NAm_Atl) have weakened substantially, range shifts over the past decades are particularly evident, e.g., in the fish assemblages in the North Sea [65] and along the North-American continental shelf [66]. On the other hand, the considerable variability in geographic trends among regions of the world is consistent with the substantial differences in coastal thermal gradients we have described here. While focusing on temperature alone is likely too simplistic to predict shifting species distributions, ‘bioclimate-envelope’ approaches are still useful on large spatial and temporal scales [67], [68] and may benefit from explicitly incorporating coastal latitudinal temperature gradients.

Lastly, shifts in coastal temperature patterns like the observed trends towards more extreme seasons may affect the connectivity between coastal marine populations. More extreme winters could increase overwinter mortalities in some organisms, while more extreme summers could both extend or curtail reproductive and growing seasons in others, thereby potentially turning continuous species distributions into more mosaic ones. Along the Australian Indian Ocean coast (Au_Ind), for example, seasonality increased significantly along the entire gradient, a trend that may challenge stenothermal tropical organisms (e.g., some corals or reef fish). Reduced connectivity may also result from opposing temperature trends between adjacent coastal regions. At the southern tip of the African continent, for example, the warm African Indian Ocean gradient and the bordering cold South-African Atlantic gradient have exhibited divergent trends over the past decades. Our analyses indicated significant warming on the Indian Ocean side (Af_Ind: SST_32.5°S  = 22.3°C₁₉₈₂→23.3°C₂₀₁₂) but cooling on the Atlantic Ocean side (SAf_Atl: SST_32.5°S  = 17.0°C₁₉₈₂→16.3°C₂₀₁₂), thereby exacerbating the temperature differential that migrating organisms must cross by more than 25% (5.3 to 7.0°C).

Conclusions

We found that the rate of latitudinal temperature change along the world's major coastlines varies more than threefold and has – in seven of 11 cases – decreased over the past three decades. SST and seasonality have increased in the majority of coastal cells, with the greatest changes observed in Northern Hemisphere gradients, but opposite trends, i.e., coastal cooling, have occurred particularly at upwelling influenced coastlines. As expected, decadal changes in gradient strength, latitude-specific SST and seasonality were found to be highly heterogeneous between regions, yet consistent with regionally differing consequences of global climate change. We suggest that past and future ecological changes in coastal marine ecosystems are linked to these regionally diverse CLTG patterns and that quantifying them is useful for future meta-analyses of large-scale ecological and thermal gradients. This would improve the current understanding about adaptive responses of marine life to climate gradients in space and, by inference, time. Valuable next steps include the identification of non-linear and cyclic SST trends, backward extension of the SST dataset, and forecasts of thermal gradient evolution under different IPCC scenarios [3] in general circulation climate models.