Decline of sea-ice in the Greenland Sea intensifies extreme precipitation over Svalbard

Extreme precipitation over the Svalbard Archipelago in the Arctic can have severe consequences for the ecosystem and society. In recent years several extreme precipitation events have been observed at Ny Ålesund, a weather station in the north-western part of the Svalbard Archipelago. The most recent observed events in the years 2012, 2016, and 2018 were the highest events in the entire precipitation record from 1974 till today. The key question of our study is whether those recently observed extremes are part of a climate change signal or are a random accumulation of extremes. With a novel approach based on a large ensemble of model simulations, we show that the likelihood of occurrence for extreme precipitation over Svalbard has increased over the last four decades. We find that the likelihood of occurrence is connected to the sea ice extent east of Greenland because the presence of sea ice shields the west coast of Svalbard from the incoming southerly moist air. Our analysis suggests, that in the future with a further decline of the sea ice coverage east of Greenland, the recently observed precipitation extremes will become even more frequent.


79
The re-forecast and forecast (simply referred to as forecasts in the follow-

88
We generally assume that precipitation events are not predictable more 89 than about two weeks in advance because of the chaotic nature of the atmo-90 spheric system (Lorenz, 1969 where values exceed 15mm/day is defined as the study area over which pre-107 cipiation values are average for the following extreme precipitation analysis 108 (red contour line in Fig. 2b).

109
The ERA5 data set provides 38 seasonal (NDJ) precipitation maxima 110 from 1981 till 2018. The data set of SEAS5 covers the same years but with 111 100 times more realizations. We find that most of the ERA5 extremes fall 112 within the range of SEAS5 simulated extremes (Fig.1b). In order to further 113 compare the statistical characteristics between the two data sets, we sample 114 38 data points selected from all 3800 data points of SEAS5 and repeat this  We perform this test on the raw SEAS5 extreme precipitation and on a 122 bias-corrected SEAS5 extreme precipitation data set. The bias correction is 123 a simple scaling of SEAS5 to ERA5, where we use a constant ratio between 124 the mean of ERA5 and SEAS5 NDJ precipitation maxima, i.e. a bias correc-125 tion factor of 1.34 for the northwest Svalbard region. After bias correction, 126 we find that the statistical characteristics, i.e. the mean, standard deviation, 127 skewness and kurtosis, fall within the 95% confidence intervals of SEAS5 (not 128 shown). We apply this simple scaling in order to keep the sea-ice to precipita-129 tion relationship the same, which would be distorted with commonly applied 130 (uni-variate) quantile-mapping approaches (Cannon et al., 2020). The ratio 131 between ERA5 and SEAS5 mean and median extreme precipitation events 132 does not influence the scaling by more than 5%.
where µ is the location, σ the scale, and ξ the shape parameter(Coles et al.,

143
The GEV analysis is performed for either all the data from SEAS5 and 144 ERA5 for a respective region, or the data is subdivided into events which 145 occurred during a specific sea-ice coverage. This sea-ice coverage is computed 146 for the region R1 ( Fig.2b) and defined as the area (in %) covered by sea-ice ice concentration larger than 15% at the same day of the event.

172
Concluding from these comparisons, we consider ERA5 useful for further 173 study of these type of precipitation extreme events, but the total precipita- towards the west coast of Svalbard, with subsequent precipitation due to 181 the orographic uplift (Fig.2). In the following, we are using the area at the 182 northwest coast of Svalbard which is most affected by those extremes (red contour line in Fig.2b).

184
Frequency of extreme precipitation and its relation to sea-ice extent 185 We utilize an ensemble hindcast of ECMWF  100 years for sea-ice extents of more than 50%.

225
In the Greenland Sea the sea-ice coverage in the NDJ-season decreased 226 over the last decade according to both ERA5 and SEAS5 (Fig. 6a). The sea-  The combined precipitation-evaporation fields reveal that during the ex-256 treme events a considerable amount of atmospheric moisture precipitates not 257 only at the west coast of Svalbard but also in coastal areas in the south of Ice-258 land and in the south-east of Greenland ( Fig. 8a and b). In addition, over the and reflects the net loss of atmospheric moisture due to increased precipita-280 tion and decreased evaporation over the sea-ice. 281 We performed an additional assessment of the composite fields divided 282 into events occurring between 1981 to 1990 and in the more recent period 283 from 2010 to 2018 (Fig. 9). As expected from the analysis in Fig. 6a, it 284 reveals that extreme events in the recent period have lower sea-ice than past 285 ones (Fig. 9). This enhances the likelihood of more intense extreme events 286 through the previously discussed changes in precipitation-evaporation and 287 the IWV (Fig. 8c and d). However, the changes in precipitation-evaporation 288 and IWV are not as clear over time as they are with changes in sea-ice extent 289 ( Fig. 8 and 9). This result is consistent with the extreme value analysis 290 in the previous section. There the partitioning into extreme precipitation 291 events during high and low sea-ice resulted in a clearer separation than the partitioning into past and present time periods (Fig. 5). Hence we conclude 293 that the sea-ice extent east of Greenland describes more robustly than a 294 trend over time the changes in the frequency of extreme precipitation over 295 Svalbard. The SEAS5 ensemble is shown by black boxplots representing the median, inter quartile range, whiskers (1.5 times the inter quartile range) and outliers (events outside the inter quartile range). The ERA5 precipitation maxima is denoted by blue x.  Return value (mm/day) c SEAS5 ice 50% SEAS5 15% < ice < 50% SEAS5 ice 15% Figure 5: Return values and periods for extreme precipitation at the West Coast of Svalbard. (a) Obtained by the General Extreme Value (GEV) statistics the return values versus return periods for extreme precipitation in the northwest Svalbard region obtained from ERA5 (blue line) and SEAS5 realizations (orange line) are shown. The shaded areas indicating the 95% confidence intervals. The dots are indicating the actual data points following an empirical distribution. (b) Same as in (a) but for the SEAS5 realizations and subdivided into events between 1981-1990 and 2009-2018. (c) Same as in (a) but for the SEAS5 realizations and subdivided into events occurring for different states of the sea-ice east of Greenland. Figure 6: The sensitivity of extreme precipitation to the recent decline of precipitation. In (a) the mean sea-ice coverage of the Greenland Sea (region R1, blue crosses) and the 3σ standard-deviation (blue shading) obtained from ERA5 are shown. Orange points indicate sea-ice coverage in SEAS5 during each of the precipitation extreme events (maximum precipitation in each NDJ season) and the blue circle indicates the sea-ice during the event in ERA5. In (b) precipitation return values (10, 20, and 50 years) for different percentages of sea-ice cover are shown.  Figure 8: Atmospheric conditions of events with low and high sea-ice. All events with a return period higher than 20 years are used in this analysis. The analysis for precipitation extremes in the northwest region are shown in (a) for events with a sea-ice coverage of the Greenland Sea region larger than 50% and in (b) less than 20%. The number of events used for the composite analysis are indicated in the upper left corner of each sub-figure. The color contours represent the composite of 1-day precipitation (positive) + evaporation (negative). The thick yellow line indicates the 15% sea-ice concentration contour line. In (c) and (d) the same decomposition into high and low sea ice events as in (a) and (b), respectively, but the color-contoured field shows the vertically integrated water vapour. Grey dots indicate areas where the g500 anomalies are significant (student t-test 98% confidence, tested with False Discovery Rate approach.