Figure 2. Mean warm-season MHW properties over the 1985-2014 period. (a-d) the heating rate (HRc, °C·day-1), measured from the onset of a continuous event to annual peak HS; (e-h) the priming period (Dp, day), measured as the number of days between the first HS of a year and the start of the continuous MHW. The mean states from the first to last rows correspond to the properties observed and simulated by GFDL-ESM4, CESM2-WACCM and MRI-ESM2, respectively.
The mean warm-season MHW properties, including the duration (Dc), accumulated heat stress (Ac), annual peak HS (HSpeak), heating rate (HRc) and “priming” period (Dp), simulated by GFDL-ESM4 generally show smaller magnitudes of model bias versus observations than CESM2-WACCM and MRI-ESM2 (Figure 1, 2, S1, 2). The spatial pattern of the biases in these thermal properties is similar between CESM2-WACCM and MRI-ESM2, and different from that of GFDL-ESM4 (Figure S1, 2).
In CESM2-WACCM and MRI-ESM2, Dc and Acare overestimated by higher magnitudes (i.e., up to 60 day and 60 °C·day in the tropical Pacific) across more of the ocean than those in GFDL-ESM4 (Figure 1a, 1c, 1d, S1a-c). The global pattern of biases is similar for the accumulated heat stress (Ac) to that for Dc in the three models (Figure 1i-l, S1g-i), which suggests a larger contribution of Dc than HSpeak to Ac. The longer duration and the associated greater accumulated heat stress are related to damped daily SST variability in the warm-season in CESM2-WACCM and MRI-ESM2 that could result in smoother HS time series (Figure S3). The annual total heat stress days (Dtot) and accumulated total heat stress (Atot), versions of the duration and accumulated heat stress metrics used in previous studies (e.g., Frölicher et al. 2018; Oliver et al. 2018), show similar spatial distribution of model biases to those of Dc and Ac in each of the three models (Figure S4).
Unlike the duration and accumulated heat stress, the maximum intensity (HSpeak) and heating rate (HRc) are overestimated by GFDL-ESM4, and underestimated by CESM2-WACCM and MRI-ESM2 across most of the ocean (Figure 1, 2, S1, S2). GFDL-ESM4 overestimates HSpeak up to 0.2-0.8 °C across most of the low- to mid- latitudes of the ocean, particularly in the eastern tropical Pacific (Figure 2b, S2a). In contrast, HSpeakis mostly underestimated in CESM2-WACCM and MRI-ESM2, with the largest underestimates up to 1.2 °C in the western boundary current and extension regions at high latitudes (Figure 1g-h, S1e-f), where model underestimates of interannual variability of the warmest month SSTs contributes to the negative biases in HSpeak (Figure S5). Smaller underestimates of up to 0.6 °C also show in these regions with GFDL-ESM4 output (Figure S1d).
For heating rate (HRc), the spatial bias patterns in the three models appear driven by that of HSpeak. In GFDL-ESM4, a large fraction of the ocean shows great overestimates in the eastern equatorial Pacific (up to 0.18 ºC·day-1, i.e., simulated HRc up to double its observed historical level), as the positive biases in HSpeak and the negative biases in Dc both contribute to the overestimates (Figure 1a-b, 1e-f, 2a-b, S1a, 1d, 2a). In contrast, large underestimates of HRc, above 0.12 °C·day-1, occur across most of the ocean in ESM2-WACCM and MRI-ESM2, due to the negative biases in HSpeak and the positive biases in Dc (Figure 1e-h, 2a-d, S1e-f, S2b-c).
The priming period (Dp) is mostly underestimated, particularly in the western to central tropics with the largest underestimates up to 90 day (Figure 2e-h, S2d-f). The negative biases in Dp are largely driven by the positive biases in Dc, as short periods of continuous heat stress combined into one longer continuous period. This is also reflected in the larger negative biases for Dp in CESM2-WACCM and MRI-ESM2, as the models show larger positive biases for Dc. (Figure S1b-c, S2e-f). Conversely, there are large overestimates, up to 90 day in GFDL-ESM4 and 60 day in MRI-ESM2, over the eastern equatorial Pacific where the model overestimated variability of daily warm-season SSTs contribute to a higher chance of occurrence of longer Dp(Figure 2e-h, S2d-f, S3).
4.2 Evaluation of future projected changes in warm-season MHW properties