Lengthening of Warm Periods Increased the Intensity of Warm-season Marine Heatwaves Over the Past Four Decades

6 Marine heatwaves (MHWs), periods of anomalously warm sea surface temperature (SST) which can 7 have significant impacts on marine ecosystems, have increased in frequency and severity over recent 8 decades. Many coastal systems (e.g. coral reefs) are particularly vulnerable to warm-season heat stress 9 when temperature can exceed organisms’ thermal thresholds and lead to mass mortality. While many 10 studies have examined the change of the warm-season heat stress occurrence over time, e.g., for coral 11 reefs, there has been less analysis of the thermal structure of heat stress events. Here we examined the 12 trend in the characteristics of warm-season heat stress (referred to as warm-season MHWs) at the 13 global-scale from 1985 to 2019, using multiple metrics for each of duration, peak intensity, 14 accumulated heat stress, heating rates and level of intensity. The results showed that warm-season 15 MHWs have become more frequent, longer-lasting, featured higher peak intensity and accumulated 16 heat stress across most of the ocean over the past four decades. Furthermore, decomposition of the 17 trends in warm-season MHWs structure showed that the increased accumulated heat stress was 18 predominantly driven by the increased duration rather than the increased intensity, especially in the 19 western and central equatorial Pacific. The results contribute to improving the understanding of warm- 20 season MHWs, which may help inform the prediction of their impacts on marine ecosystems as well as 21 marine conservation and management under climate change.


59
For example, a long period of moderate heat stress (e.g., four weeks of 1°C anomaly) may imply less 60 severe biological impacts than a numerically equivalent short period of high SST anomalies (e.g., one

61
week of 4°C anomalies). Thus, it is important to decompose the trend of increasing heat stress intensity 62 via a comprehensive examination of the structure of heat stress events.

63
In this study, we examined changes in the thermal structure of warm-season MHWs globally over 64 the past four decades. We developed 13 metrics that describe the thermal structure of warm-season complete SSTs gridded at 0.05º (i.e. about 5km, a resolution comparable with the scale of many marine 82 ecosystems, e.g. coral reefs; Skirving et al. 2020). The gap-free SST dataset with high spatial and 83 temporal resolutions provides an unprecedented opportunity to develop the fine-scale metrics of warm-84 season MHWs and evaluate them at small regional scales worldwide. Here, the data analysis excluded 85 the regions poleward of 60ºN and that of 60ºS to avoid the influence of sea ice in calculating trend in 86 warm-season MHW characteristics.

98
We conducted all analyses twice, using two different methods to define the thermal threshold in    152 the p90 threshold are generally lower than those computed using the MMM threshold (Fig. 2, 3; Fig.  S4). Therefore, we present the results using the MMM threshold here, with equivalent results for the 155 p90 threshold available in the supplementary material (Fig. S3, S4).

156
The duration of continuous positive HS period (Dc) significantly increased in 93% of the ocean 157 from 1985 to 2018 at a rate of 0.2 to 3 day·year -1 (Fig. 2a, S3a). The western equatorial Pacific, Bering

165
The trends in the number of total days with HS (Dtot) and the warm season length between the first 166 and last HS (Dws) in a year, show broadly similar spatial patterns to that of Dc, but with a greater extent 167 of decreasing values (Fig. 2a-c). These two metrics increased across 90% and 72% of the ocean over

173
The combination of negative trends in Dtot and Dws but positive trends in Dc, as occurred in parts 174 of the central and eastern tropical Pacific, eastern Indian Ocean, and the south of 50°S, suggests that

197
Sea and the ocean of south of Africa (Fig. 3a).

198
Unlike the duration and accumulated heat stress metrics, the heating rate metrics, HRc and HR, 199 significantly decreased in 60% and 84% of the ocean, respectively, with exceptions in some western 2a) and small or no trends in HSpeak (Fig. 3a, S5a). The trend in heating rate measured from the date of 205 first HS (HR) showed similar global patterns but lower magnitudes, primarily due to higher magnitudes 206 of increase in the corresponding duration, Dtot (Fig. 3b, S5b).    (Fig. 4). The coefficients for Dc are significantly larger (i.e. p< 223 0.05 and r 2 >0.8) than that for HSpeak over the central and eastern equatorial Pacific, as well as parts of 224 the mid-and high-latitude oceans (Fig. 4). The greater contribution of duration to the increasing trend

272
The loss of such conditioning or priming periods may be a threat to coral reefs and other coastal

275
The large positive trends in the periods of warm days (i.e. Dc, Dtot and Dws) and small or even proposed different processes that theoretically restrain the maximum SST, including the feedback 288 effects of evaporation and wind (Newell 1979;Wallace 1992), clouds (Ramanathan and Collins 1991) 289 as well as ocean dynamics (i.e. advective and vertical heat transport; Sun and Liu 1996). Although the 290 significant global ocean warming drove the increase in peak SST intensity, these processes might

342
The use of multiple data sources in generating the CoralTemp (v3.1) SST dataset might create 343 inconsistencies that potentially affect our results. The dataset bias correction reference derived using 344 NRT OSTIA is consistent with those used in the input data sources (i.e. OSTIA reanalysis SST and 345 NESDIS Geo-Polar SST products). Although the consistency of the bias correction data and the use of 346 a linear transition approach in combining data sources reduce the biases when data sources switch, the 347 differences in data densities among different data sources result in spatial and temporal inconsistencies.

348
Compared with the open ocean and the area around Europe/Africa, the rest of the ocean has less dense 349 SST data before 2002, mainly because of the regional variations of the frequency of ship collected in-350 situ data from ICOADS that are blended for generating the SST dataset. The higher data density from