Historical marine heatwaves over 1982–2018
We consider two metrics for MHWs: annual MHW days (the number of MHW days per year) and mean MHW intensity (the mean temperature anomaly during all MHWs in each year relative to the seasonal climatology). For globally important habitat-forming organisms, annual MHW days alone was strongly and significantly correlated with increased coral bleaching, decreased seagrass density, and decreased kelp biomass in an observational study9. Furthermore, the number of annual MHW days showed a more robust correlative relationship than other common measures of ocean temperature such as mean and maximum SST9, and is, therefore, a key metric for assessing these kinds of ecological impacts.
To evaluate model performance, simulated historical MHW metrics are compared with observations (MGD). We focus on the 60°S–60°N spatial comparison of the climatological state during the overlap period among OFAM3, CMIP5, and MGD (1982–2018; Figs. 1 and 2). Simulated and observed annual MHW days exhibit low spatial variability (mostly around 30 days) with relatively high values (>35 days) in the equatorial Pacific (Fig. 1a–c). CMIP5 shows much smoother spatial distribution than OFAM3 and MGD, with small inter-model spread (mostly <3 days; Fig. 1d). Both OFAM3 and CMIP5 show <10 fewer MHW days than MGD in western boundary currents (Fig. 1e, f).
The spatial patterns of simulated and observed historical mean MHW intensity are characterised by more intense MHWs in western boundary currents and lower intensity in the tropics and subtropical gyres except for the equatorial Pacific (Fig. 2a–c). Although these spatial patterns are in close agreement between the models and observations, the magnitude of mean MHW intensity differs considerably in western boundary currents (Fig. 2e, f). OFAM3 simulates >1 °C more intense MHWs in western boundary currents, whereas CMIP5 simulates <1 °C less intense MHWs.
The spatial patterns of mean MHW intensity resemble those of the standard deviation of de-seasonalised daily SST (Fig. 3), which is indicative of temporal variability at the scale relevant for mesoscale processes and MHWs. This high correlation in the spatial patterns between these two variables is consistent with the findings of previous studies using different observational and model data products9. Similar to the mean MHW intensity, OFAM3 exhibits higher SST variability relative to MGD primarily around western boundary currents, whereas CMIP5 shows lower variability. While the higher values in OFAM3 may suggest an overestimated variability in these eddy-rich regions, we believe that the actual biases are smaller than indicated here. Previous studies suggest systematic negative biases of SST variability represented in gridded observation-based SST analysis products13,23,24. Specifically, these studies report a systematic underestimate of eddy kinetic energy in the Southern Ocean by as much as 60–70% when calculated from gridded altimetry data because of interpolation, which smooths variability, compared to along-track data. Given that the SST analysis products such as MGD are also optimally interpolated, the SST variability, and hence the mean MHW intensity calculated from these gridded products may likewise be underestimated. This implies that the positive biases of OFAM3 are likely smaller and the negative biases of CMIP5 are larger than the ground truth. Another factor in these differences is the spatial resolution, which determines the variability due to fine-scale features that can be captured in each product. In this aspect, the higher variability of OFAM3 and the lower variability of CMIP5 relative to MGD are expected results.
Projected changes over the next three decades
Projected changes in mean SST, annual MHW days, and mean MHW intensity under the highest-emission Representative Concentration Pathway 8.5 (RCP8.5) scenario25 are compared between OFAM3 and CMIP5 (Fig. 4). These changes represent the difference between the simulated climatological state over the recent decades (1982–2018) and the next 30 years (2021–2050). In terms of the 60°S–60°N global averages, the projected changes agree well between OFAM3 and CMIP5: an increase of 0.75 °C in mean SST (for both OFAM3 and CMIP5), an increase of 149 days (OFAM3) vs. 144 days (CMIP5) in annual MHW days, and an increase of 0.17 °C (OFAM3) vs. 0.16 °C (CMIP5) in mean MHW intensity (Fig. 4a-f). Therefore, OFAM3 and CMIP5 experience the same level of SST warming and nearly the same level of MHW intensification at the global scale.
The spatial patterns of projected mean SST increase are consistent between OFAM3 and CMIP5, depicted by >1 °C warming in the subpolar North Pacific, equatorial Pacific, and parts of the northern North Atlantic (Fig. 4a, d). The rate of warming is lower in the subpolar North Atlantic and to a lesser extent in the Southern Ocean. The difference in the rate of warming between OFAM3 and CMIP5 is generally larger in western boundary currents (Fig. 4g). Notably, the OFAM3-projected warming is about 0.5 °C lower in the Kuroshio Current and Gulf Stream than the CMIP5-projected warming. The inter-model spread of the CMIP5 SST projections is highest in the subpolar North Atlantic (Fig. 4j).
Both OFAM3 and CMIP5 project a greater increase in annual MHW days in many parts of the tropics and subtropical gyres (except for the equatorial Pacific) and subpolar North Pacific (Fig. 4b, e). In contrast, the rate of increase is much less in the subpolar North Atlantic and near 60 °S. In addition, OFAM3 projects similarly low rates of increase in the western boundary currents, central equatorial Pacific, Leeuwin Current, and Antarctic Circumpolar Current, but these features are absent in CMIP5. Consequently, the projected increase in annual MHW days in these regions is substantially (>120 days in some cases) lower in OFAM3 than CMIP5 (Fig. 4h). Conversely, OFAM3 projects notably higher increase in annual MHW days in the subpolar North Pacific, northeast North Atlantic, subtropical gyres, and the south of the Indian sector of the Antarctic Circumpolar Current. The inter-model spread of CMIP5 is relatively high in some parts of the Southern Ocean, subpolar North Pacific, and North Atlantic (Fig. 4k).
Projected changes in mean MHW intensity common to both OFAM3 and CMIP5 include greater intensification in the subtropical North Atlantic and lesser intensification and weakening in some cases in the subpolar North Atlantic and Southern Ocean (Fig. 4c, f).
There are several regions of large differences in the projected changes in mean MHW intensity between OFAM3 and CMIP5 (Fig. 4i). OFAM3 projects greater intensification in some parts of the North Pacific, East Australian Current (including the South of Tasmania), and Brazil Current. Furthermore, the OFAM3-projected mean MHW intensity exhibits a decrease in the central and eastern equatorial Pacific, Kuroshio Current, and Gulf Stream. The inter-model spread of CMIP5 is relatively large in the subpolar North Atlantic (Fig. 4l), consistent with the spread for SST and annual MHW days discussed above (Fig. 4j, k).
Spatial details in western boundary currents
As demonstrated by the global map comparison (Fig. 4h, i), western boundary currents stand out as the regions of disagreement between OFAM3 and CMIP5 in MHW projections. We now focus on the spatial patterns of the projected changes in each of the five western boundary currents in the 0.1° OFAM3 output, and compare them with those of the 1° CMIP5 output (Figs. 5 and 6).
Both OFAM3 and CMIP5 generally agree in the spatial patterns of projected changes in annual MHW days and mean MHW intensity, but the former provides much more spatial details, including elevated changes along the tracks of boundary currents. For example, in OFAM3, the pathway of the Kuroshio Current along the southern coast of Japan is depicted by a more pronounced increase in annual MHW days than its surroundings (Fig. 5a). Resolving such spatial variability will be essential for predicting bluefin tuna recruitment26. Another notable example is in the projected mean MHW intensity change along the track of the Brazil Current (Fig. 6i, j). While both OFAM3 and CMIP5 show an increase in the northern part and a decrease in the southern part of the domain, the former exhibits much finer spatial structures with higher variability.
Importantly, the 0.1° product can provide information on the projected changes along coastal areas, which is impossible with the 1° product as indicated by missing values in white. Dynamics in these narrow regions can be quite different from those of the offshore areas, and so the projected changes can likewise be different. For example, the southeast coast of Tasmania experiences a negligible increase in mean MHW intensity (<0.1 °C), whereas its offshore counterpart shows much greater increase (Fig. 6g). Hence, the higher-resolution model output such as ours can be used for risk assessment of temperature-sensitive marine aquaculture, such as Pacific oyster farming27.
Relationship with sea surface temperature warming
The relationship between projected MHW intensification and SST warming is examined both at the global and regional scales (Fig. 7). Both OFAM3 and CMIP5 show that annual MHW days and mean MHW intensity increase linearly with mean SST warming in the next three decades globally as well as in the western boundary current regions. This finding is expected because we use a fixed baseline period (1982–2018) to define MHWs, but is useful to quantify such a relationship for assessing ecological impacts9.
Both the OFAM3 and CMIP5 projections indicate that the global- and regional-mean SSTs in western boundary currents will be at least 1 °C warmer in the next three decades under the RCP8.5 scenario than their averages over 1982–2018. At this level of warming, the number of simulated annual MHW days increases by ~200 days and the mean MHW intensity increases by nearly 0.3 °C on global average (Fig. 7a, b). Regionally, the same linear relationship holds, but reveals different slopes (Fig. 7c–l). Notably, the rates of increases in annual MHW days and mean MHW intensity are generally lower (roughly 150 days and 0.15 °C increases at 1 °C warming) in the western boundary current regions than those of the global averages. Among these western boundary currents, the Agulhas Current is where OFAM3 and CMIP5 differs the most in terms of the rate of increase in annual MHW days (roughly 100 vs. 200 days per 1 °C warming; Fig. 7g). The linear relationship with mean SST warming is noisier for mean MHW intensity than annual MHW days, in which OFAM3 reveals more variability than CMIP5, presumably due to the multi-model averaging of CMIP5.