Ocean Fronts and Eddies Remotely Forcing Atmospheric Rivers and Heavy Precipitation

approximately a 40% increase in landfalling ARs and up to a 30% increase in heavy precipitation in mountainous regions. The modeling results further show that this remote impact occurs within two weeks, implying the potential influence of the dynamical processes on AR predictability at subseasonal-to-seasonal time scales. A 30 proposed mechanism for the influence of mesoscale SST forcing on ARs is the 31 asymmetrical response of the atmosphere to warm vs . cold mesoscale SSTs over the 32 eddy-rich Kuroshio Extension region, which results in a net increase of moisture flux 33 above the planetary boundary layer, prompting AR genesis via enhancing moisture transport into extratropical cyclones in the presence of mesoscale SST forcing.


approximately a 40% increase in landfalling ARs and up to a 30% increase in heavy 27
precipitation in mountainous regions. The modeling results further show that this 28 remote impact occurs within two weeks, implying the potential influence of the 29 dynamical processes on AR predictability at subseasonal-to-seasonal time scales. A 30 proposed mechanism for the influence of mesoscale SST forcing on ARs is the 31 asymmetrical response of the atmosphere to warm vs. cold mesoscale SSTs over the 32 eddy-rich Kuroshio Extension region, which results in a net increase of moisture flux 33 above the planetary boundary layer, prompting AR genesis via enhancing moisture 34 transport into extratropical cyclones in the presence of mesoscale SST forcing. 35 Since the term was coined nearly three decades ago 1 , atmospheric rivers (ARs), which 36 are plumes of intense water vapor transport emanating from the atmospheric moisture pool, 37 have been recognized as one of the most important sources of extreme hydroclimate events 38 in the global extratropics, capable of producing torrential rains and floods when making 39 landfall over regions of elevated orography, such as the West Coast of North America 1,6,,7 . 40 Some of the most severe river floods in California were associated with ARs 8 . A recent 41 global analysis of ARs' role in driving hydrological extremes found that ARs can 42 contribute not only to extreme floods in many major drainage basins, but also to drought 43 occurrence when ARs are inactive 9 . Therefore, developing and improving the capability of 44 predicting ARs at subseasonal-to-seasonal (S2S) time scales, especially landfalling ARs, 45 can have important implications for water resource management, flood control, and 46 drought relief. Though forecasts of overall occurrence and intensity of ARs have been 47 improved in current weather forecast and climate models, the timing and location of 48 landfalling ARs, as well as their precipitation impact, are notoriously difficult to 49 predict 3,4,5,10,11 , underscoring the importance of a better understanding of ARs' dynamics 50 and their predictability sources. A community-driven research effort is currently ongoing 51 to understand and quantify uncertainties in AR science 12,13 . 52 Previous studies of ARs' predictability sources have focused on tropical modes of 53 variability at S2S time scales, such as the Madden-Julian Oscillation (MJO) 14,15 and El 54 Niño-Southern Oscillation (ENSO) [16][17][18] . To the best of our knowledge, none of the 55 published studies explore the potential influence of midlatitude mesoscale SSTs induced 56 by fronts and ocean eddies on ARs, despite the fact that these features along major ocean 57 fronts, such as the Kuroshio Extension (KE) and Gulf Stream (GS), are well known for 58 their impact on the overlying atmosphere, as revealed by high-resolution satellite 59 observations and climate model simulations [19][20][21][22][23][24] . Evidence is also mounting that the 60 influence of mesoscale SSTs can extend beyond the atmospheric boundary layer, affecting 61 extratropical cyclones and midlatitude storm tracks at far distance [25][26][27][28][29][30][31][32][33][34][35][36] . Given the close 62 association between the occurrence of ARs and extratropical cyclones 37-42 , a natural 63 question to ask is: can mesoscale SSTs influence ARs on S2S time scales, particularly 64 landfalling ARs and associated heavy precipitation events? 65 To investigate this question, we conducted two ensembles of twin simulations in the 66 North Pacific sector using a regional climate model -the Weather Research and 67 Forecasting (WRF) model with a 27 km horizontal resolution (Methods). Each of the twin-68 simulations consists of two nearly identical runs with only difference between them being 69 the SST forcing: the control run (hereafter CTRL) was forced with the high-resolution 70 (0.09°) satellite-based MicroWave InfradRed Optimal Interpolated (MW-IR) 43,44 daily 71 SST, whereas the filtered run (hereafter FILT) was forced with the same SST but subject 72 to a lowpass spatial filter to suppress mesoscale features (Methods). The first ensemble 73 was based on a set of boreal-winter (6-month) twin simulations from 2002 to 2014 74 (hereafter seasonal-ensemble or SE), and the second ensemble was based on a set of two-75 week twin experiments for a selection of winter cyclone cases (hereafter cyclone-ensemble 76 or CE). The SE experiment was designed to examine the overall impact of mesoscale SST 77 forcing on ARs over the winter season, whereas the CE experiment was designed to further 78 investigate the impact of mesoscale SSTs on ARs during cyclogenesis and development in 79 the KE region. We chose the North Pacific because 1) the largest population exposed to 80 ARs related flood risk is along the West Coast of North America 9 and 2) the KE front and 81 eddies generate the strongest mesoscale SST variability in North Pacific (Extended Data 82 Fig. 1) and substantially influence the North Pacific storm track 45-47 . 83 Figure 1a&b shows the integrated water vapor transport (IVT) averaged over the AR 84 landfalling days in the SE CTRL compared to that derived from the latest European Centre 85 for Medium-Range Weather Forecast (ECMWF) reanalysis (ERA5) 48 . WRF faithfully 86 reproduces the observed landfalling ARs but with a slightly underestimated IVT intensity, 87 due to a higher frequency of landfalling ARs simulated in WRF (Extended Data Fig. 2). 88 Landfalling ARs can produce high precipitation over elevated terrain through orographic 89 lift. This orography-locked precipitation feature appears in the high-resolution dataset 90 based on the Parameter-elevation Regression on Independent Slopes Model (PRISM) 49 , 91 and is reproduced remarkably well by WRF (Extended Data Fig. 3). However, because the 92 PRISM dataset covers only the continental U.S., the following observational analysis uses 93 the satellite-based Global Precipitation Measurement (GPM) dataset 50 for heavy 94 precipitation events and the ERA5 reanalysis for landfalling ARs. Although the GPM 95 dataset does not have sufficient resolution to resolve the orography-locked precipitation 96 feature shown in PRISM, it does show a high precipitation concentration along the West 97 Coast of North America (Fig. 1f) that corresponds well to the landfalling ARs (Fig. 1b). 98 In fact, the landfalling AR-induced precipitation in GPM shows comparable values to those 99 derived from the SE CTRL (Fig. 1c&f), both of which are considerably greater than the 100 corresponding winter-mean (Extended Data Fig. 3), indicating that WRF is skillful in 101 simulating AR-related precipitation. However, the probability density function (PDF) of 102 daily precipitation rate averaged over the West Coast of North America (the magenta box 103 in Fig. 1c) shows an overestimation of accumulated precipitation by WRF (Fig. 1d&g). 104 Nevertheless, the fractional contribution (%) of AR-related precipitation to total 105 precipitation exhibits similar distributions between GPM and WRF (Fig. 1e&h). These 106 results give confidence that WRF is capable of realistically simulating AR-induced heavy 107 precipitation along the West Coast of North America, thereby providing a basis for further 108 analysis of the influence of mesoscale SST on landfalling ARs and related precipitation 109 statistics. 110 To understand the relationship between landfalling ARs and heavy precipitation, we 111 selected a subset of landfalling ARs that are concurrent with heavy precipitation events 112 defined as those with daily precipitation rate exceeding the 75 th percentile of the area-113 averaged daily precipitation over the West Coast of North America. This subset of 114 landfalling ARs and heavy precipitation events is used in the analyses below. We note that 115 the results do not fundamentally change if extreme precipitation events (exceeding 90 th 116 percentile of daily precipitation) are used. Figure 2a&b shows the ensemble-mean 117 landfalling AR IVT accumulated over the heavy precipitation days divided by the total 118 number of winter days (150 days) in the SE CTRL and the corresponding value of SE 119 CTRL minus SE FILT, respectively. The reason for using the accumulated rather than 120 averaged IVT is that suppressing of mesoscale SSTs in FILT leads to a significant decrease 121 in both frequency and strength of landfalling ARs. Accumulated IVT takes account of both 122 AR frequency and strength change while averaged IVT undercuts the frequency change. 123 In fact, the total number of landfalling ARs detected in all 65 ensemble members drops 124 from 829 in SE CTRL to 631 in SE FILT. The accumulated IVT of the landfalling ARs is 125 increased by ~40% in SE CTRL compared to SE FILT (Fig. 2b). Figure 2c&d shows the 126 accumulated concurrent heavy precipitation divided by the total winter days and the 127 corresponding difference between SE CTRL and SE FILT. As expected, heavy 128 precipitation over high terrain is most significantly affected by the change in landfalling 129 ARs (Fig. 2d). The presence of mesoscale SST forcing results in up to a 30% increase in 130 heavy precipitation (Fig. 2e) due to the increase of landfalling ARs in the region. A further 131 support to this finding comes from an analysis on the relationship between strength of 132 mesoscale SST forcing and strength of landfalling ARs and heavy precipitation response. 133 Because of the shortness of the record, we simply grouped the 13 years (2002-2014) of SE 134 CTRL into two 4-year sets based on the strength of mesoscale SST forcing and compared 135 the landfalling ARs and heavy precipitation between these two sets (Methods). The results 136 show that the set with stronger mesoscale SST forcing produces stronger landfalling AR 137 and precipitation response (Fig. 3). The CE experiment allows us to further address the question of whether mesoscale 149 SST forcing along the KE front is responsible for the change in landfalling ARs and heavy 150 precipitation. The CE ensemble consists of 568 winter cyclone cases selected from the SE 151 CTRL such that they all passed through the KE region (Methods). Therefore, the majority 152 of ARs generated in this experiment are closely related to these cyclones that pass over the 153 mesoscale SST forcing along the KE. Remarkably, despite the short integration period, the 154 landfalling AR IVT is significantly increased in CE CTRL compared to CE FILT 155 (Extended Data Fig. 5), leading to a significant increase in the precipitation amount due to 156 the presence of mesoscale SST features (Fig. 4a&b). The two-week mean fractional 157 increase of precipitation (>40 mm day -1 ) in CE CTRL compared to CE FILT can reach 15% 158 ( Fig. 4c). More interestingly, the time-evolving pdf of the fractional precipitation change 159 along the West Coast of North America reveals a delayed response: negligible change 160 within the first 4 days, but a significant increase of up to 40% for the precipitation higher 161 than 40 mm day -1 afterwards (Fig. 4d). This indicates that the influence of mesoscale SST 162 forcing can occur on weekly time scales, thereby potentially affecting the predictability of 163 AR-related heavy precipitation events along the West Coast of North America. 164 The delayed precipitation response to the mesoscale SST forcing points to a remote 165 forcing mechanism on landfalling ARs that originates in the eddy-rich KE region. We 166 hypothesize that the key process involved in this remote mechanism lies in the strong 167 influence of mesoscale SST features along the KE (Extended Data Fig. 1) on the net 168 moisture supply to the developing cyclones over this region. This effect is particularly 169 strong following each initial cyclone (selected to initialize the twin ensemble runs) passing 170 through the KE region. In the wake of the initial cyclone's cold front, cold and dry air 171 descends over the KE region. Over warm mesoscale SSTs the atmosphere is destabilized, 172 intensifying vertical mixing and resulting a strong upward vertical moisture flux that 173 pumps moisture out of the PBL. Over cold mesoscale SSTs, however, such a moisture 174 pumping does not occur, because the atmosphere is more stable there. As a result, there is 175 a net increase of moisture above the PBL (Fig. 5a&b). A recent modeling study shows a 176 similar moisture increase when ocean eddy induced SSTs are included in a set of WRF 177 simulations 28 . This increase of moisture supply from the PBL over the KE acts to moisten 178 the precyclone environment for the next developing cyclone. As such, when the next 179 cyclone develops over the KE, the airflow within the warm sector of the cyclone, known 180 as the feeder airstream 39 , can transport more moisture into the cyclone. A branch of this 181 feeder airstream feeds to the warm conveyor belt ascent, contributing to cyclone 182 precipitation, while another branch turns away from the cyclone, exporting moisture from 183 the cyclone to form AR 39 . Thus, the ability of mesoscale SSTs to moisten precyclone 184 environment over the KE region can lead to increase in AR IVT. This mechanism also 185 explains the delayed precipitation response along the West Coast of North America, 186 because the initial cyclone does not produce major differences in ARs and associated heavy 187 precipitation due to the small difference in the precyclone environment between CTRL and 188 FILT during the initial stage. 189 The impact of mesoscale SSTs on cyclone-related AR IVT generation is revealed 190 by a composite analysis (Fig. 6). Despite of the small differences in intensity and structure 191 of the composite cyclone between CE CTRL and CE FILT (Fig. 6a&b), there is a 192 significant increase of AR IVT and precipitation in the warm sector of the composite 193 cyclone when mesoscale SSTs are present in CE CTRL (Fig. 6c-f). Since the majority 194 (>85%) of ARs is associated with extratropical cyclones in the simulations, we can 195 conclude that the cyclone-related AR IVT difference is largely responsible for the total AR 196 difference, including the landfalling AR difference, between CTRL and FILT. It indicates 197 that the presence of mesoscale SSTs can enhance AR genesis by increasing IVT associated 198 with extratropical cyclones even though cyclone intensity remains unchanged. 199 The enhanced moisture supply above PBL over the KE region is also observed in 200 ERA-Interim by contrasting the periods between high-and low-resolution SST forcing 201 (Extended Data Fig. 4g&h). As shown by both WRF and the reanalysis, maximum 202 moisture anomalies carried by the ARs occur at ~ 800hPa, and in WRF the value is about 203 10% higher in CE CTRL than in CE FILT during the first day of ARs (Fig. 5c). This 204 enhanced moisture supply promotes generation of stronger ARs in the region, as shown by 205 the AR genesis PDF (Fig. 5d) that displays a marked shift towards higher IVT values in 206 CE CTRL compared to CE FILT. Since the stronger ARs are more likely to survive the 207 journey across the Pacific, the fractional difference of AR moisture content between CE 208 CTRL and FILT is expected to increase (Fig. 5c) as ARs move westward, eventually 209 making landfall along the West Coast of North America and impacting rainfall in the region 210 ( Fig. 5e-h). The time scale associated with this remote mechanism is approximately 4-5 211 days, consistent with the result in Fig. 4d. This mechanism of enhanced vertical moisture 212 transport by warm mesoscale SSTs is further tested for its robustness by using different 213 PBL schemes in WRF. The results show that all the schemes produce a net increase of 214 water vapor content above the PBL in response to mesoscale SST forcing (Methods and 215 Extended Data European west coast to GS mesoscale SST anomalies (Extended Data Fig. 7). Given 225 differences in model physics and numerics between WRF and CAM, the consistency 226 between these modeling results further points to the robustness of the findings. 227 Collectively, these numerical simulations and observations support the hypothesis 228 that mesoscale SST forcing associated with oceanic fronts and eddies in western boundary 229 current regions substantially influences landfalling ARs and associated heavy precipitation 230 on S2S time scales. It indicates that the common practice of using non-eddy-resolving 231 monthly SST as forcing in atmosphere-only models can lead to underestimates of AR-232 induced heavy precipitation, even for high-resolution atmospheric models.  as a and b, but for precipitation (mm day -1 ). PDF of relative difference of precipitation 428 concurrent with landfalling ARs between SE CTRL and FILT in reference to SE CTRL 429 (the red curve in Fig. 1) (e). Heavy precipitation events are defined as area-averaged 430 (magenta box in Fig. 1c) daily precipitation events exceeding the 75 th percentile of the 431 value. The difference above 95% confidence level based on a two-sided student t-test is 432 shaded by gray dots. 433 Figure 3 | Relationship between ARs/precipitation response strength and SST forcing 434 strength in the seasonal ensemble. a and c, same as Fig. 2b and 2d, but for four strong 435 mesoscale SST forcing cases. b and d, same as Fig. 2b and 2d, but for four weak mesoscale 436 SST forcing cases. 437 Figure 4 | Response of heavy precipitation along west coast of North America to 438 mesoscale SST forcing in the cyclone ensemble. Two-week mean heavy precipitation 439 (mm day -1 ) associated with landfalling ARs in CE CTRL (a) and the corresponding 440 difference between CE CTRL and FILT (b). Two-week mean (c) and time evolving (d) 441 PDF of relative difference of 6-hourly precipitation (averaged in the magenta box in Fig.  442 1c) concurrent with landfalling ARs between CE CTRL and FILT in reference to CE CTRL. 443 The mean precipitation is computed as the sum of landfalling AR induced precipitation 444 over the heavy precipitation days divided by the two-week simulation period. The 445 difference above 95% confidence level based on a two-sided student t-test is shaded by 446 gray dots. same as a and b, but for composite of AR IVT (kg×m -1 s -1 ) associated with extratropical 472 cyclones. e and f, same as a and b, but for composite of AR-induced precipitation (mm 473 day -1 ). The difference above 95% confidence level based on a two-sided student t-test is 474 shaded by gray dots. 475 476 Methods 477

Seasonal Ensemble 478
The For CE, the anomalies were derived from 6-hourly IVT subtracting two-week mean of all 526 568 ensemble members. For the global CAM simulations, the anomalies were derived from 527 daily mean IVT subtracting winter season (DJF) mean of all ensemble members. The outer 528 edge of an AR is defined by a closed IVT contour of 250 kgm -1 s -1 . The length of an AR 529 must be longer than 2000 km while its width must be narrower than 1000 km. ARs were 530 designated as landfalling if the outmost contour insects coastline. The center of an AR is 531 defined as the geometric center of the IVT contour. Coherent AR object is stitched from a 532 Lagrangian tracking approach 61 to form an AR trajectory. We tested a different AR 533 detection algorithm using integrated water vapor and the tracking results by including or 534 not including a temporal requirement that tracked ARs must last for at least 18h，the 535 results showed no significant effects on the conclusion of this study. 536

Definition of Heavy Precipitation Events 537
Heavy precipitation events along the west coast of North America are defined as area-538 averaged (magenta box in Fig. 1c) daily precipitation events exceeding the 75 th percentile 539 daily precipitation rate. To test the robustness of the results, the analysis was repeated using 540 extreme precipitation events (exceeding 90 th percentile daily precipitation rate) and results 541 consistently show higher precipitation associated with landfalling ARs in SE CTRL than 542 FILT. Heavy precipitation events were used because it allows for a larger sample size, 543 increasing the robustness of the results. 544 IVT and precipitation between the two sets are shown in Fig. 3. 554

Comparison of ARs and Water Vapor in ERA-Interim and ERA5 555
The SST forcing in ERA-I was switched from low resolution (1°) to high resolution 556

Extratropical Cyclones (EC) Detection and EC/AR Composite 561
A widely-used winter storm detection approach 62 was used to identify and track ECs. 562 Centers of ECs were first identified by sea level pressure (SLP) minima within a closed 563 contour, with an additional requirement of at least 1hPa increase of SLP within 5° of the 564 center. 400hPa temperature was used to detect and eliminate those with a warm core so 565 that the identified ECs are distinct from tropical cyclones. Candidates are then stitched in 566 time to form paths, with a maximum distance of 6° between them. The identified ECs must 567 have a during of at least 2 days and a traveling distance of 10°. Following a recent study 63 , 568 if both an AR and an EC co-exist within a 4000km x 4000km box centered at EC center, 569 the related field including SLP, IVT and precipitation are used in the composite analysis. 570 The detection of extratropical cyclones and composite analysis were performed in the 571 whole North Pacific region([150°E-240°E, 20°N-60°N]). 572

Significance Test 573
A student t-test is applied for a given variable when comparing the difference between 574 CTRL and FILT, assuming each of the detected ARs is an independent sample. Data value 575 in regions without ARs occurrence is set to zero. heavy precipitation days and then divided by the total number of days (150x65) in SE 688 CTRL (a) and the corresponding difference between SE CTRL and FILT (b). c and d, same 689 as a and b, but for precipitation (mm day -1 ). PDF of relative difference of precipitation 690 concurrent with landfalling ARs between SE CTRL and FILT in reference to SE CTRL 691 (the red curve in Fig. 1) (e). Heavy precipitation events are defined as area-averaged 692 (magenta box in Fig. 1c) daily precipitation events exceeding the 75 th percentile of the 693 value. The difference above 95% confidence level based on a two-sided student t-test is 694 shaded by gray dots. 695 696 697 (mm day -1 ) associated with landfalling ARs in CE CTRL (a) and the corresponding 707 difference between CE CTRL and FILT (b). Two-week mean (c) and time evolving (d) 708 PDF of relative difference of 6-hourly precipitation (averaged in the magenta box in Fig.  709 1c) concurrent with landfalling ARs between CE CTRL and FILT in reference to CE CTRL. 710 The mean precipitation is computed as the sum of landfalling AR induced precipitation 711 over the heavy precipitation days divided by the two-week simulation period. The 712 difference above 95% confidence level based on a two-sided student t-test is shaded by 713 gray dots. 714 715