The recent normalization of historical marine heat extremes

Climate change exposes marine ecosystems to extreme conditions with increasing frequency. Capitalizing on the global reconstruction of sea surface temperature (SST) records from 1870-present, we present a centennial-scale index of extreme marine heat within a coherent and comparable statistical framework. A spatially (1 ̊ × 1 ̊) and temporally (monthly) resolved index of the normalized historical extreme marine heat events was expressed as a fraction of a year that exceeds a locally determined, monthly varying 98 percentile of SST gradients derived from the first 50 years of climatological records (1870– 1919). For the year 2019, our index reports that 57% of the global ocean surface recorded extreme heat, which was comparatively rare (approximately 2%) during the period of the second industrial revolution. Significant increases in the extent of extreme marine events over the past century resulted in many local climates to have shifted out of their historical SST bounds across many economically and ecologically important marine regions. For the global ocean, 2014 was the first year to exceed the 50% threshold of extreme heat thereby becoming “normal”, with the South Atlantic (1998) and Indian (2007) basins crossing this barrier earlier. By focusing on heat extremes, we provide an alternative framework that may help better contextualize the dramatic changes currently occurring in marine systems.


Introduction
The United States (US) National Oceanic and Atmospheric Administration (NOAA) generates the US Climate Normals [1] based on decadal averages of temperature and precipitation captured in the most recent climatological period. While the US Climate Normals provide advice for what may occur locally in the near future, fixed historical benchmarks may be more useful for describing contemporary climate departures and ecosystem disruptions. Such reference points are timely, as the persistent warming of sea surface temperatures (SST) and the increased frequency and intensity of extreme climatic events has fueled the perception that unprecedented climate normals are emerging in the global ocean [2,3]. When it comes to measuring extremes, a large-scale shift in the mean and amplitude of ocean warming variability can affect the frequency of extreme temperature events through changes in the standard deviations and skewness [4,5]. Nonetheless, increases in the frequency, magnitude, and a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 persistence of extreme marine heat events (variously defined) and climate disruptions have been recently described across spatiotemporal scales [6,7].
The statistical analysis of past, present and future extreme events has primarily highlighted how background warming increases the intensity, frequency, and duration of extreme events. However, by contrast to these characterizations, we know little of how extreme events, especially those associated with warming, have evolved historically to become a new normal under global warming [11,24]. This knowledge gap is problematic in part as it may lead to the false perception that marine ecosystems are primarily at risk in future time horizons, with less attention to the impacts of unprecedented heat extremes already being experienced on a global scale [11,25,26]. Furthermore, while these projection-based analyses are essential to climate adaptation efforts, substantial uncertainties and biases can arise due to the stochastic nature of global climate systems [27][28][29][30], further contributing to the lack of public consensus and action on climate change [31,32].
Rigorous and transparent analyses of the dynamic marine environment with long-term records, however, can reliably characterize the emerging new normal marine climate in an appropriate historical context. Here, we use reconstructed climatological datasets providing a unique opportunity to explore the historical change in the occurrence of extreme marine heat events. Our analysis focused on sea surface temperature (SST), a key driver of marine ecosystems. We used 150 years of reconstructed SST datasets and analyzed changes in the frequency of historical marine heat extremes across all Exclusive Economic Zones, Biogeochemical Provinces, and Large Marine Ecosystems. With this analysis, we aim to describe the progression and expansion of extreme heat in global ocean and identify regions that have observed the most and least frequency of extreme heat. In doing so, our goal is to contextualize the present status of marine regions, indicating current climate disruptions and ecosystem risk, and to provide a timeline for the global ocean when historically benchmarked extreme events occurred more than 50%, thus becoming "normal."

Methods
We used 150 years (1870-2019) of gridded, monthly reconstructed historical SST data sources to evaluate centennial changes in global occurrences of extreme marine heat events: the Hadley Centre Sea Ice and SST dataset (HadISSTv1.1) [33] and the Characteristics of the Global SST data (COBESSTv2) [34] (Table A in S1 File). These independent and complementary global SST products were reconstructed from instrument records and the historical network of in situ measurements and have been widely deployed as ground-truthed SST fields [35,36].
Our statistical definitions frame how we characterized the frequency and extent of extreme heat events across space and time. For each month, in each 1˚× 1˚grid, we defined the extreme marine heat as a monthly average SST value that exceeds the 98 th percentile SST value observed over 1870-1919 (corresponds to the period of second industrial revolution), or hottest temperature observed in the earliest 50-year period of record (Fig A in S1 File) [37 -39]. Such percentile based thresholds can be derived from climatological data and are robust to drivers or variabilities associated with individual extreme events [10]. Our particular percentile based threshold also easily relates to the standard deviation (σ) which offers an alternative expression of dataset anomaly-the 98 th percentile is when σ = 2.05. Though they are limited in describing daily and hourly extremes, we selected monthly SST products in order to evaluate the properties of extreme marine heat events at a 1˚× 1˚scale within the longest historical context possible-150 years. At daily scales, species may respond to stressful abnormal temperatures by changing distributions but could suffer greater thermal-induced stress if an extreme heat event persists beyond one month. In addition, extreme heat events at shorter time scales are more likely to be smaller in scale than our analyzed spatial units (e.g., EEZs, large marine ecosystems). Furthermore, statistical analysis of monthly-resolved temperature variations can offer centennial-scale proxies of the frequency of extreme heat event properties [11].
A metric of the normalization of historical marine heat extreme event can be expressed as a fraction of a year that exceeds a locally identified threshold relative to the 1870-1919 climatology (hereafter referred to as a "local extreme heat index: LEHI"). For each grid cell, the LEHI is the proportion of each year (0-1) that exceeds the monthly extreme SST values. Thresholdbased LEHI can quantify the magnitude and frequency of extreme events by summarizing the number of time units exceeding a fixed threshold at a given location. The LEHI can be further aggregated across space and time, offering a simple summary of heat extremes, or exceedance rates above the fixed historical benchmark for a season, period, or a region of interest. This proxy-based inference allowed us to generate global and centennial summaries of normalized extreme index from 1870 to 2019. If the climatology remains stationary throughout the full series, this value should approximate 0.02 independent of time or spatial scaling.
Having derived historical benchmarks of extreme heat, we generated summaries of a normalized LEHI for the global ocean from 1870-2019. For the global ocean, we mapped seasonal (Jan-Mar, Jul-Sep) LEHI summaries in recent decades (1980-2019) and calculated regional LEHIs for 2010-2019. We selected the regional frameworks of Large Marine Ecosystems [LMEs; 40], Exclusive Economic Zones [EEZs; 41], and biogeochemical provinces [BGCPs; 42] as they describe ecologically bounded transnational waters, political boundaries, and subbasin scale perspectives on the distribution of marine heat extremes, respectively. To add to our understanding of centennial trends, we plotted the time series of changes in the fraction of surface area exceeding the historical 98 th percentile threshold based on all calendar months for each of the seven major ocean basins (Table B in S1 File). Listed by descending area these are the: South Pacific (84.8 × 10 6 km 2 ), North Pacific (77.0 × 10 6 km 2 ), Indian (70.6 × 10 6 km 2 ), North Atlantic (41.5 × 10 6 km 2 ), South Atlantic (40.3 × 10 6 km 2 ), Southern (22.0 × 10 6 km 2 ), and Arctic (15.6 × 10 6 km 2 ). For each basin, we calculated the year (if applicable) when 50% of the total area fraction exceeded and remained above the 1870-1919 climatology (i.e., the historical extreme heat benchmark, and referred to simply as climatology hereafter). The timing of such a thermal regime shift can be used to approximate the emergence of a new normal.
Further, our index ranks economically and ecologically important marine regions, identifying regions having the most and least risk of extreme heat.
Next, to highlight any different information presented in our approach, we compared the global variability of the LEHI to a more traditional SST anomaly metric. For the year 2019, we computed both spatial outputs from the same 1870-1919 climatology and the same spatial scale (1˚× 1˚) to ensure that any differences are from the methodology alone. SST anomalies are widely used and an important impact parameter in climate extreme studies [10,11,21]. The difference in distributions of two climate indices derived from the same baseline period offers an alternative assessment of climate stress from the conventional anomaly-from-mean signals.
We conducted all data wrangling and analyses in the R programming environment [43], and provided our data and scripts in open access repositories (https://bit.ly/2QjhYld) and through the Open Science Framework (https://osf.io/mj8u7/).  Contemporary summaries of LEHI from 2010-2019 highlight EEZs, LMEs, and BGCPs where current ocean climate stress is most pronounced (Fig 2). Regions that experienced a significant departure from climatology are not distributed evenly across the global ocean. Of the 142 EEZs, 66 LME, and 55 BGCPs examined, LEHI scores were highest between 35˚N and 35˚S, in the Tropic and Subtropic regions. Regions from the eastern Pacific Ocean with higher SST variability experienced relatively low LEHI values. The Maldives (median LEHI = 0.929) and Tanzania   We find that from 1900 to 2019, the area fraction of global ocean surface exceeding the historical threshold (98 th percentile during 1870-1919) exceeded 50% in 2014, while several major ocean basins outpaced the global average. Fig 3 shows an increase in the fraction of global and regional ocean surfaces exceeding the 1870-1919 extreme event threshold from less than 20% in the early 1900s to more than 50% in the 2010s. The period of 1980-2018 experienced a significant increase in global and regional area fractional exceedance, with strong El Niño events contributing to pronounced increases in 1998 and 2016. We define the point of no return ("PoNR") as the year when more than 50% of surface areas exceeded and remained above the 1870-1919 threshold. This is first achieved in the South Atlantic (1998) and Indian  Table C in S1 File).

Results
The spatial distribution of normalized LEHI can be compared to the conventional climate indices derived from the IPCC-relevant climatology  in order to provide an alternative climate stress assessment at any point in time (Fig 4). Not unexpectedly, distributions of normalized LEHI and the SST anomalies have similarities and differences. According to the 2019 SST anomalies map, SST anomalies above 99 percentile (1.37˚C) were limited to the Chukchi Sea, Bering Sea, Gulf of Alaska Davis Strait, and southern portions of the North Sea and Baltic Sea, which a priori occurs in 1% of global ocean. However, the area of the global ocean with a normalized LEHI value above the 99th percentile (LEHI = 1) is substantially larger, covering 6.7% of global ocean. High LEHI values in the Northern Hemisphere also correspond to the Arctic-sea-ice boundary, where some of the strongest warming trends have been observed [46]. On the other hand, low LEHI values (and low S/N ratios) are found in the equatorial Pacific region with high SST fluctuations dominated by ENSO processes [44, 46]. Regional differences in LEHI values reflects (1) the strength of measured climate signals relative to the amplitude of local and regional variabilities and (2) show that the emergence of significantly different climates (i.e., normalization of extreme events) has already happened in many regions, particularly in the low latitudes. These findings are consistent with recent studies that demonstrated the emergence of climate signals that would be unknown by past baselines were detectable at regional scales [44,47,48].

Added value from historically informed benchmarks
Knowledge of how local climate has varied in the past, relative to any historical baselines, is fundamental for understanding many aspects of the marine system in a changing climate. Studies on climate extremes has often addressed relatively recent trends or events in geographically restricted regions [49] or addressed present-future time horizons [9]. In the immediate aftermath of extreme events, there is often great public interest in attributing causes to extreme events soon after they occurred [50,51]. However, researchers are increasingly recognizing the need to incorporate larger external forcings (e.g., atmospheric or oceanic circulation) for extreme event attributions at broader temporal scales. Detection of climate change impacts within a historical context offers clear advantages as reconstructed data are generally more reliable, especially during the period when global or near-global in situ data became available [52].
Using a statistical framework and globally reconstructed SST datasets, we demonstrate that many local climates have shifted out of their historical bounds. Our approach followed the statistical framework that is commonly used to examine the changes in the occurrence of univariate climate extreme events across time and space. This method that incorporates seasonally varying percentile thresholds with fixed or sliding baselines offers a suite of metrics that defines the characteristics of each event. For example, Oliver et al. [11] provided a global summary of the increase in annual MHW days, defined by a seasonally varying 90th percentile threshold based on a 1983-2012 climatology, along with other statistics including the duration and maximum intensity of each event. An advantage of our method is that it follows similar statistical definitions but provides a new metric that can be easily translated to the magnitude of climate normals, which can be calculated at any spatial and temporal scale.
Historical records can also provide new insights and empirical support to understand the role of non-equilibria dynamics and cumulative impacts of novel thermal disturbances that many regions have already incurred [53]. Understanding the full consequences of an individual extreme event is also best interpreted with substantial information from past disturbances [54]. While data from the longer past (e.g., before 1980) can be subject to larger uncertainties (e.g., sparser sampling and different measurement practices [12,35], historical data can provide a wealth of information to understand interactions among sequences of climate-driven events if used with caution.

Extremes accounting provides an ecological reality check
Normalizing extreme marine heat events in a historical ecological context can be applied to evaluate the shifting baseline of ocean health and identify climate vulnerable regions. Large parts of the tropics are exhibiting clear emergence of warming signals [44]. Upwelling areas may moderate the severity and occurrence of extreme marine heat and therefore could function as ecological refugia for adjacent regions. Amplified normalization of extreme heat events in the Tropics suggests that biological hotspots in these regions have experienced unprecedented warming over the past decades. Major biodiversity hotspots in the Tropics include coral reef ecosystems that contain coral species that are close to their upper thermal tolerance. For habitat-forming organisms, relatively small increases of SST above the average summer maximum can lead to mass coral bleaching events, reduce seagrass density, and diminish kelp canopy coverage [26,55]. The maintenance of this state or, worse, further increases in heat extremes could potentially push many ecosystems beyond their thermal tolerance and towards permanent shifts.
While marine biota can typically adapt to gradual changes in environmental conditions [56-59], abrupt changes in the frequency and extent of extreme events experienced by local biodiversity can fundamentally alter the ecosystem structure, functions, and services [60]. Our results showed some of the largest LEHI values are found in the tropical Atlantic and tropical Indian Oceans (Fig 1). These changes in LEHI values may indicate the changes in the distribution of highly migratory species that are highly conditioned by water temperature. For example, our findings align with Monllor-Hurtado et al. [61], which showed the shift of tuna catches away from the tropics to the subtropics, likely reflecting the movement of tropical tuna populations to avoid these warmer thermal regimes in these regions.
Negative impacts of extreme marine heat events have been documented worldwide through extensive coral bleaching [62], mass mortality events [63], and toxic algal blooms [64]. Furthermore, a change in the frequency of extreme heat events in coastal waters, could be more harmful to sessile benthic species in the shallow water ecosystems [65] such as corals, kelps, and seagrasses. Further normalization of extremes could substantially impact the growth and reproduction of many commercial fisheries within LMEs and EEZs, with dramatic socioeconomic implications [21]. Improving our ability to connect past-present-future climate-driven ecological risks will allow us to develop a different set of management and conservation measures of living marine resources that better reflect spatially varying ocean health baselines [8,9,21].

Implications for climate science communication
Recent increases in extreme climatic events have heightened public discourse and concern over climate change impacts. At the same time, participants in climate change dialogues often express a strong interest in reliable information on historical and future changes as a basis for policies aimed at adaptation and planning. Simultaneously, characterization and assessment of extreme climatic events have become critically important criteria for a wide range of policy decisions. This study provides a robust historical framework to characterize extreme marine heat in order to inform climate change impacts at various spatial and temporal scales. Our globally resolved, centennial-scale extreme reanalysis can be used as a flexible and effective climate description and communication tool for the public and policymakers, and may help advance further science communication efforts to gain public understanding and confidence in extreme climate events and their attribution to anthropogenic climate change.
While the prediction of future climate change impacts remains challenging, facilitating constructive climate change dialogue may face fewer barriers when drawing from historical climate records. Using the methods applied here, we find that extreme climate change is not a hypothetical future possibility, but a past historical event that has already occurred in the global ocean. Though this occurred earlier in some regions, 50% of the ocean's surface experienced extreme heat in 2014, and this has steadily increased thereafter.