Page 1
1
Hydroclimatic Conditions Preceding the March 2014 Oso Landslide 1
Brian Henn1, Qian Cao
2, Dennis P. Lettenmaier
2, Christopher S. Magirl
3, Clifford Mass
4, J. Brent 2
Bower5, Michael St. Laurent
6, Yixin Mao
1, and Sanja Perica
6 3
4
Manuscript prepared for Journal of Hydrometeorology as an expedited contribution 5
6
1Department of Civil and Environmental Engineering, University of Washington, Seattle, WA 7
2Department of Geography, University of California, Los Angeles, CA 8
3U.S. Geological Survey, Tucson, AZ 9
4Department of Atmospheric Sciences, University of Washington, Seattle, WA 10
5NOAA/National Weather Service, Seattle, WA 11
6Hydrometeorological Design Studies Center, NOAA/National Weather Service, Silver Spring, 12
MD 13
14
15
Corresponding Author: Brian Henn, Department of Civil and Environmental Engineering, 16
University of Washington, 201 More Hall, Box 352700, Seattle, WA, 98195; 17
[email protected] ; (510) 301-3147 18
19
PEER REVIEW DISCLAIMER: This draft manuscript is distributed solely for purposes of 20
scientific peer review. Its content is deliberative and predecisional, so it must not be disclosed or 21
released by reviewers. Because the manuscript has not yet been approved for publication by the 22
U.S. Geological Survey (USGS), it does not represent official USGS finding or policy. 23
Page 2
2
Abstract: The 22 March 2014 Oso landslide was one of the deadliest in U.S. history, resulting in 24
43 fatalities and the destruction of more than 40 structures. We examine synoptic conditions, 25
precipitation records, soil moisture reconstructions, and observed streamflow in the days, 26
months, and years preceding the landslide. Atmospheric reanalysis shows a period of enhanced 27
moisture transport to the Pacific Northwest beginning on 11 February 2014. The 21- to 42-day 28
periods prior to the landslide had anomalously high precipitation; we estimate that 300-400 mm 29
of precipitation fell at Oso in the 21 days prior to the landslide. Relative only to historical periods 30
ending on 22 March, the return periods of these precipitation accumulations are large (25-88 31
years). However, relative to the largest accumulations from any time of the year (annual 32
maxima), return periods are more modest (2-6 years). In addition to the 21-42 days prior to the 33
landslide, there is a secondary maximum in the precipitation return periods for the 4 years 34
preceding the landslide. Reconstructed soil moisture was anomalously high prior to the landslide, 35
with a return period that exceeded 40 years about a week before the event. 36
37
Classifications: precipitation, extreme events, hydrometeorology, soil moisture, complex terrain 38
Page 3
3
1. Introduction 39
On 22 March 2014, a high-mobility landslide was initiated from a 190 m-tall bluff of 40
unconsolidated glacial till and outwash 6 km east of the town of Oso, Washington. The landslide, 41
with an estimated volume of 8×106 m
3, traveled almost 2 km, crossed the North Fork of the 42
Stillaguamish River floodplain and State Route 530, and covered an area of 1 km2 to a depth as 43
great as 20 m (Iverson et al., in press). It destroyed over 40 structures in a rural neighborhood 44
and killed 43 people (Keaton et al. 2014; Magirl et al. 2014). In terms of loss of life, the Oso 45
landslide (also called the SR 530 Landslide by the State of Washington), was the second worst in 46
U.S. history, surpassed only by the 1985 Mameyes, Puerto Rico landslide (Jibson 1992). 47
Landslides from Pleistocene bluffs have been widespread along the floodplain of the 48
North Fork of the Stillaguamish River (Tabor et al. 2002; Dragovich et al. 2003; Haugerud 2014; 49
Keaton et al. 2014), where unstable deposits of glacial till and outwash intermix with clay-rich 50
lacustrine base units, creating conditions conducive to deep-seated landslides and slumps 51
(Thorsen 1989; Miller and Sias 1997; Dragovich et al. 2003). The 2014 Oso landslide initiated 52
from the recurrently active Hazel landslide, which had slumped several times in previous 53
decades (Shannon 1952; Benda et al. 1988; Thorsen 1989; Miller and Sias 1997, 1999; Keaton et 54
al. 2014). Analysis of historical precipitation data combined with groundwater modeling (Miller 55
and Sias 1997) indicated that landslide instability was sensitive to 1-year cumulative 56
precipitation, although Iverson (2000) argues that groundwater pressure distribution, not rainfall, 57
is the salient factor in triggering large landslides. The landslide site is characterized by a wet 58
climate on the west slope of the Cascade Range, where seasonal rainfall maxima normally occur 59
between late fall and late winter. The 2014 Oso landslide came after six weeks of heavy rainfall 60
Page 4
4
with nearby precipitation totals approaching gauge record maxima for 45-day and 4-year 61
intervals ending in March (Iverson et al., in press). 62
We examine the hydroclimatic conditions preceding and at the time of the landslide, 63
using historical records of precipitation, regional model-reconstructed soil moisture, and 64
groundwater storage inferred from streamflow records in the vicinity of the landslide. In so 65
doing, we seek to place the event in a hydroclimatic context. We do not discuss failure 66
mechanisms pertinent to the landslide, which we leave to others. Rather, we focus on the 67
synoptic setting, precipitation patterns over timescales from days to years, and associated soil 68
moisture indicators that may have contributed to local moisture fluxes associated with the 69
landslide. 70
71
2. Methods 72
Our source of regional synoptic information was the NOAA Climate Forecast Model 73
version 2 (CFSv2) analysis fields for 1 October 2013-22 March 2014. Anomalies were 74
constructed from the CFSv2 reanalysis (CFSR) for 1 October 1980-22 March 2011 (Saha et al. 75
2010). CFSv2 analysis fields, and the CFSR fields are available from the National Climatic Data 76
Center. 77
Our primary sources of precipitation observations were a gauge at the North Fork of the 78
Stillaguamish River near Oso (Oso gauge), approximately 3 km west of the landslide, and the 79
NOAA Cooperative Observer (COOP) gauge located at the Darrington Ranger Station 80
(Darrington gauge) approximately 18 km to the east of Oso (Fig. 1). The Oso gauge has been 81
operated by the Washington Department of Ecology since 30 October 2011; while the 82
Darrington gauge has been in operation since 1931. The Oso gauge is a Hydrological Services 83
Page 5
5
America tipping bucket; the Darrington gauge is an NWS Standard 8 in diameter (non-recording) 84
Rain Gauge (SRG). 85
In addition to the primary precipitation stations, we identified nine long-term stations (all 86
SRGs) within a radius of about 80 km of Oso (Fig. 1) and with homogeneous precipitation 87
frequency distributions, using the criteria of Hosking and Wallis (2005). We obtained daily 88
precipitation accumulations for the Oso gauge from Washington Department of Ecology records 89
(fortress.wa.gov/ecy/wrx/wrx/flows/station.asp?sta=05B090) and for the Darrington and regional 90
gauges from the Western Regional Climate Center archives 91
(www.wrcc.dri.edu/summary/Climsmwa.html). 92
We calculated precipitation accumulations for 28 time windows ranging from 1 day to 10 93
years, with each window ending on 22 March 2014. Historical and 2014 accumulations were 94
compared in two ways: by considering only windows ending on 22 March, or the largest annual 95
accumulation among windows ending on any day of that water year. For each precipitation 96
gauge site and time window, we fit a probability distribution – the Generalized Extreme Value 97
(GEV) distribution – to the past years’ accumulations. Then, we compared 2014 accumulations 98
with the fitted distributions to estimate their return periods, with reference to both the 22 March 99
and annual maximum series. 100
To ensure that our frequency analysis results for 2014 precipitation were robust, we also 101
carried out a regional frequency analysis using the set of nine gauges plus the Darrington gauge. 102
This involved creating a pooled set of annual maximum precipitation accumulations from all 103
sites, with the maxima from each site scaled by mean precipitation at the site. Estimating return 104
periods using a pooled sample from multiple sites helps to reduce the variance from an at-site 105
analysis. 106
Page 6
6
Soil moisture reconstructions were taken from the University of Washington’s Drought 107
Monitoring System for the Pacific Northwest 108
(www.hydro.washington.edu/forecast/monitor_west), which has a 1/16° (latitude and longitude) 109
spatial resolution. It provides reconstructed soil moisture in three zones from the Variable 110
Infiltration Capacity (VIC) model (Liang et al. 1994) for the period 1920-2014. We analyzed the 111
historical time series of reconstructed total column soil moisture for the grid cell that 112
encompasses the landslide site (Fig. 1) to determine whether the estimated soil moisture in 2014 113
was anomalous. 114
The Supplemental Materials (SM) provide greater detail on the approach used to fit 115
probability distributions to the precipitation accumulations, the regional frequency analysis and 116
the method of estimating return periods at the Oso gauge. SM also provide details of the 117
hydrologic model used to generate the long-term series of soil moisture. 118
119
3. Results 120
a. Synoptic setting 121
Water year 2014 (beginning 1 October 2013) was characterized by two extremes prior to 122
the landslide on 22 March: a period of high pressure and unusually dry conditions in the fall and 123
early winter, followed by an anomalously wet period from mid-February through late March, in 124
which moist, onshore flow dominated. Fig. 2 shows the 500-hPa geopotential height, precipitable 125
water, and zonal wind anomalies from climatology (1981-2010) for two periods: 1 October 126
2013-10 February 2014 (dry), and 11 February through the date of the landslide (wet). During 127
the autumn and early winter dry period, a strong positive height anomaly reaching approximately 128
110 m was centered over the southern Gulf of Alaska, resulting in enhanced upper-level 129
Page 7
7
northerly flow over the west coast of North America (Fig. 2a). The column-total precipitable 130
water for that period was below normal over the western U.S. (Fig. 2c). 131
The approximately six weeks of wet weather in the Pacific Northwest preceding the 132
landslide were characterized by a different synoptic pattern, with a positive 500-hPa height 133
anomaly of about 60 m over central California, and below-normal heights over the north-central 134
Pacific and western Canada (Fig. 2b). The resulting geostrophic wind anomalies during this 135
period were thus southerly over a large portion of the eastern Pacific and westerly over the 136
Pacific Northwest, a pattern that promoted the advection of tropical and subtropical moisture into 137
the region. This is supported by the precipitable water anomaly for that period (Fig. 2d), which 138
shows a plume of enhanced moisture values (2-6 kg m-2
) extending from Hawaii (Fig. 2d). Zonal 139
wind anomalies show below-normal westerly flow in the early dry period and enhanced westerly 140
flow during the preceding six weeks (Fig. 2e and 2f); increased westerlies enhance upslope flow 141
on the roughly north-south oriented Cascade Range. 142
b. Precipitation accumulations 143
Precipitation totals (Fig. 3, red lines) immediately (<3 days) prior to the event were light, 144
but in the week prior totals were in the range of 50-150 mm at most gauges. Cumulative 145
precipitation does not increase greatly as the window is extended to 14 days prior to the event, 146
but does increase rapidly from 14 to 21 days prior. In the 21 days prior to the landslide, the Oso 147
gauge received 334 mm of precipitation and the Darrington gauge received 401 mm. Cumulative 148
42-day precipitation was 506 mm and 682 mm at Oso and Darrington, respectively. These 149
patterns were consistent across the nine other gauges in the regional group (Fig. 3, gray lines). 150
Precipitation totals increase more slowly as the window is extended to 180 days, reflecting the 151
generally dry conditions prior to 11 February, but at 365 days, precipitation totals were at or 152
Page 8
8
somewhat above climatological norms. Above-average precipitation accumulations occurred in 153
each water year from 2011 to 2014, despite the anomalously dry start to water year 2014. 154
c. Precipitation return periods 155
Return periods of the 2014 accumulations at Darrington, with reference to historical 156
accumulations ending on 22 March, are shown in Fig. 3 (solid line with squares). The return 157
periods of several years or more indicate that the precipitation leading up to the landslide was 158
unusually heavy for this time of year. The peak return period is for the 21-day window prior to 159
the event (88 years). We also found large (>20 year) return periods for the 17- through 45-day 160
windows. As the window is extended to 180 days, return periods decrease due to the earlier dry 161
period, but then begin increasing again and exhibit a second maximum at the 4-year 162
accumulations window, which has a return period of 9.5 years. Note that for multi-year 163
accumulation windows, sample sizes are reduced as we consider only non-overlapping multi-164
year periods prior to the date of the landslide. 165
From the perspective of causal mechanisms, it may be more important to place the 166
precipitation prior to the landslide in the context of heavy precipitation occurring at any time of 167
the year, rather than windows ending on 22 March. For Darrington, return periods based on these 168
annual maximum accumulations are also shown in Fig. 3 (solid line with crosses). When 169
considered relative to precipitation accumulations occurring at any time of year, the 2014 170
accumulations are less unusual. For windows of less than one year, the return periods peak at 2.9 171
years for the 42-day accumulation, indicating that the 2014 accumulations were barely above the 172
median annual maximum 42-day precipitation accumulation. Based on the results at Darrington, 173
the precipitation accumulations in the 3-6 weeks prior to the landslide would be expected to 174
occur sometime in the year on average about once every 3 years. 175
Page 9
9
The solid line with circles in Fig. 3 shows the return periods at Darrington estimated 176
using the regional frequency analysis. The results are nearly identical to those from the single-177
site analysis, with the exception of the multiyear windows, for which the regional frequency 178
analysis yields greater return periods than the at-site analysis (16 years vs. 9.5 years for the 4-179
year accumulation). While the use of regional frequency analysis is often advocated to improve 180
the robustness of the estimated return periods (see e.g. Hosking and Wallis 2005), in this case 181
there was not a great difference between the at-site and regional return periods. For the Oso 182
precipitation gauge, the return periods are shown as the dashed line with circles in Fig. 3. The 183
return periods are greater than those estimated at Darrington, with a peak return period of 5.3 184
years for the 21-day window. While higher, the pattern and magnitudes of the return periods at 185
Oso are not dramatically different from those at the Darrington gauge. 186
d. Soil moisture 187
Reconstructed soil moisture in March 2014 was above the median daily value for the 188
entire month, and exceeded or was close to the 10-year return period for the 10 days preceding 189
the landslide (Fig. 4; note that soil moisture return periods are for specific dates, rather than the 190
entire year). The maximum came on 16 March, with a return period of about 43 years, slightly 191
less than the maximum of record for that date. On 22 March, the soil moisture return period was 192
about 10 years. Soil moisture remained more or less constant (with slight variations in response 193
to wet periods) until about a month after the landslide, and then decreased rapidly with the onset 194
of the dry season (not shown). The reconstructed soil moisture reflects the precipitation in the 195
wet period prior to the landslide, as soil moisture responds roughly to accumulated precipitation 196
in the prior months, less effects of evapotranspiration, which are modest during the wet season. 197
Page 10
10
The signature of the dry period earlier in the water year is also evident in Fig. 4, as the soil 198
moisture was below average prior to the beginning of the wet period on 11 February. 199
200
4. Discussion and Conclusions 201
Precipitation in the 21 to 42 days prior to the Oso landslide was heavy, reflecting the 202
transition during the second week of February to strong onshore flow and anomalously high 203
precipitable water over the region. On a climatological basis, we estimate that the 21-day 204
accumulations had a recurrence interval of 2.1 years. However, when compared to the same 21 205
calendar days prior to the event in the gauging record, the return period was much larger (88 206
years). This reflects the fact that heavy precipitation of this duration tends to occur earlier in the 207
water year, typically from November through January. At the two precipitation gauges nearest to 208
the landslide location, the accumulated 21-day precipitation was 300-400 mm. We also found a 209
secondary multi-year maximum in precipitation for the 4 years prior to the landslide. A regional 210
analysis yielded essentially the same results as an examination of individual gauges. 211
Soil moisture reconstructed from a regional model indicates that it was anomalously high, 212
in excess of the 10-year recurrence interval just prior to the landslide. Six days prior to the 213
landslide, soil moisture was at its water year 2014 maximum, with an estimated return period for 214
that date of about 43 years. An examination of groundwater storage inferred from summer 215
baseflow analysis was inconclusive (see SM). 216
While the precipitation observed in the lead-up to the 2014 Oso landslide was heavy, 217
based on the return periods estimated here, the 21- to 42-day precipitation accumulations would 218
be expected to occur as often as every 3 years, or no more rarely than every 5-6 years. This 219
shorter window of heavy precipitation occurred in the context of a 4-year period of above-220
Page 11
11
average precipitation. It is noteworthy that the period of high precipitation in the weeks before 221
the landslide came near the end of the seasonal wet period, when soil moisture (arguably a better 222
indicator of hazard than precipitation alone) was already high. 223
We did not examine the mechanisms that caused the landslide; such analyses are left to 224
others. However, by placing the hydroclimatic conditions during the event in historical context, 225
our analysis should contribute to a better understanding of the events leading to the landslide. 226
227
Acknowledgements 228
Support for the lead author was provided by a University of Washington Valle 229
Scholarship. Additional funding was provided to the U.S. Geological Survey by the Federal 230
Emergency Management Agency. 231
Page 12
12
References 232
Benda, L., G. Thorsen, and S. Bemath, 1988: Report of the ID Team Investigation of the Hazel 233
Landslide on the North Fork of the Stillaguamish River. 234
Dragovich, J. D., B. W. Stanton, W. S. J. Lingley, G. A. Greisel, and M. Polenz, 2003: Geologic 235
Map of the Mount Higgins 7.5-minute Quadrangle, Skagit and Snohomish Counties, 236
Washington. 237
Haugerud, R., 2014: Preliminary interpretation of pre-2014 landslide deposits in the vicinity of 238
Oso, Washington: U.S. Geological Survey Open-File Report. 239
Hosking, J. R. M., and J. R. Wallis, 2005: Regional frequency analysis: an approach based on L-240
moments. Cambridge University Press,. 241
Iverson, R., 2000: Landslide triggering by rain infiltration. Water Resour. Res., 36, 1897–1910, 242
doi:doi:10.1029/2000WR900090. 243
Iverson, R. M., and Coauthors, Landslide mobility and hazards: implications of the 2014 Oso 244
disaster. Earth Planet. Sci. Lett.,. 245
Jibson, R. W., 1992: The Mameyes, Puerto Rico, landslide disaster of October 7, 1985. Rev. Eng. 246
Geol., 9, 37–54. 247
Keaton, J. R., J. Wartman, S. Anderson, J. Benoit, J. DelaChapelle, R. Gilbert, and D. R. 248
Montgomery, 2014: The 22 March 2014 Oso Landslide, Snohomish County, Washington. 249
Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based 250
model of land surface water and energy fluxes for general circulation models. J. Geophys. 251
Res., 99, 14415–14428. 252
Magirl, C. S., M. K. Keith, J. E. O’Connor, R. Aldrich, S. W. Anderson, and M. C. Mastin, 2014: 253
Preliminary Assessment of Aggradation Potential in the North Fork Stillaguamish River, 254
Washington, downstream of the SR 530 Landslide: U.S. Geological Survey Open-File 255
Report. 256
Miller, D., and J. Sias, 1997: Environmental Factors Affecting the Hazel Landslide. 257
——, and ——, 1999: Hazel/Gold Basin Landlsides: Geomorphic Review Draft Report. 258
Saha, S., and Coauthors, 2010: The NCEP climate forecast system reanalysis. Bull. Am. 259
Meteorol. Soc., 91, 1015–1057. 260
Shannon, W. D., 1952: Report on Slide on North Fork Stillaguamish River Near Hazel, 261
Washington. 262
Page 13
13
Tabor, R. W., D. B. Booth, J. A. Vance, and A. B. Ford, 2002: Geologic Map of the Sauk River 263
30- by 60-Minute Quadrangle, Washington. 264
Thorsen, G. W., 1989: Landslide provinces in Washington. Engineering Geology in Washington: 265
Washington Division of Geology and Earth Resources Bulletin 78, R.W. Galster, Ed., 71–266
89. 267
268
Page 14
14
Figures 269
270 Fig. 1 – Area map of the Oso landslide, showing precipitation gauge locations, the soil moisture 271
model grid cell and stream gauge used in this study. 272
273
Page 15
15
274
Fig. 2 – a) and b), anomalies from climatology for 500 hPa geopotential height (m); c) and d), 275
precipitable water (kg m-2); and e) and f), zonal wind (m s-1); a), c) and e), 1 October 2013-10 276
February 2014; b), d) and f), 11 February-22 March 2014. Anomalies are based on NCEP 277
Climate Forecast System Reanalysis. The flux of large amounts of water vapor into the PNW 278
during 11 February-22 March 2014 was coupled with stronger-than-normal zonal (east-west) 279
winds. 280
281
Page 16
16
282 Fig. 3 – Precipitation return periods preceding the 2014 Oso landslide. Return periods 283
calculated using several different approaches at the Darrington and Oso gauges are shown in 284
blue. Accumulated precipitation at both gauges is shown in red, and accumulations at the other 285
COOP gauges are shown in gray. 286
287
Page 17
17
288 Fig. 4 – Reconstructed soil moisture for the VIC model grid cell that encompasses the Oso 289
landslide site. Modeled daily soil moisture for water year 2014 is compared to the daily median, 290
10-year return period value and maximum soil moisture for 1920-2013. Soil moisture is given as 291
total water in the grid cell soil column. 292
293
Page 18
18
Supplemental Materials 294
a. Precipitation accumulation calculations 295
We included data for a given time window in a given year if at least 80 percent of the 296
days in the window had non-missing data. The Darrington station has relatively complete records 297
beginning in 1931, although approximately four percent of all days were not directly observed 298
but instead reported later as multi-day accumulations; we redistributed these accumulations over 299
the unobserved days using daily precipitation distributions at the nearest reporting gauge. Up to 300
20 percent of missing days in the window were filled by using data from the nearest adjacent, 301
complete station, which were scaled by the ratio of average monthly precipitation at the target 302
station to precipitation at the adjacent station. Windows with less than 80 percent of days present 303
were treated as missing. 304
We first calculated the return periods of the 2014 precipitation at Darrington. For each 305
window length, we determined the number of valid (non-missing) past years in the record, which 306
ranged from 77 to 80 years depending on the window. We fit Generalized Extreme Value (GEV) 307
distributions to the data using the L-moments approach (Hosking and Wallis 2005), for both the 308
accumulations ending on 22 March, and the annual maximum accumulations. We used the GEV 309
distribution, because it has been widely used, including in the most recent revisions to NOAA 310
Atlas 14: Precipitation-Frequency Atlas of the United States (Perica et al. 2013). We then 311
compared the 2014 accumulations with the fitted distributions to infer the frequency or return 312
period of the 2014 precipitation, with the 22 March distributions necessarily generating larger 313
return periods than the annual maximum accumulation distributions. 314
b. Regional frequency analysis 315
Page 19
19
For the regional frequency analysis, we first set out to identify a set of homogenous 316
stations according to the guidelines of Hosking and Wallis (2005). Using their homogeneity 317
score (a test for relative similarity of the probability distributions of precipitation between 318
stations), we first eliminated stations in the coastal lowlands such as Anacortes, Chimacum, 319
Coupeville and Port Townsend. Doing so improved the group homogeneity score of the 320
remaining stations, presumably because the excluded stations are significantly drier than the 321
remaining mountain stations (even though the regional growth curve approach normalizes by 322
mean values at each station). The Monroe station was removed because it failed the discordancy 323
test of Hosking and Wallis (2005). The Arlington and South Fork of the Tolt River stations were 324
eliminated because Mann-Kendall trend tests revealed strongly increasing annual maximum 325
accumulations, a potential sign of inconsistencies of those stations with others in the region. The 326
remaining 10 stations, all on the west slope of the Cascades, showed acceptable homogeneity and 327
formed the region for our regional frequency analysis. These stations (see Fig. 1) are: Baring, 328
Concrete, Darrington, Diablo Dam, Newhalem, Ross Dam, Sedro Woolley, Snoqualmie Falls, 329
Startup and Upper Baker Dam. All of these stations had at least 45 years of data, and between 330
91.7 and 99.6 percent data completeness. 331
The regional frequency analysis was carried out by normalizing each station’s annual 332
maximum accumulations by their mean values, pooling the normalized values into a single 333
dataset, and then fitting a probability distribution to the pooled values following Hosking and 334
Wallis (2005). A goodness-of-fit test statistic was computed for each window for use in 335
identifying the best three-parameter probability distribution to represent the likelihood of annual 336
maximum precipitation accumulations. GEV and the generalized logistic distributions were 337
identified as the best three-parameter distributions. The return periods calculated by the two 338
Page 20
20
distributions show little difference, and we therefore adopted the GEV distribution as our 339
regional probability distribution. The GEV distribution was fit separately for each accumulation 340
window, creating regional growth curves, which may be more robust than distributions fitted to 341
data from a single station. The 2014 accumulations were then compared with the regional growth 342
curve to estimate their return period. The year for annual maximum accumulations for windows 343
of 1 year or longer was defined to end on 22 March, hence the fixed and annual maximum return 344
periods are the same for these windows. 345
Finally, we used the regional growth curve and the Oso precipitation gauge to estimate 346
the return period of observed precipitation at that site, which was closest to the landslide. To do 347
so, we followed the approach of Schaefer et al. (2002) and first estimated the at-site means of the 348
annual maximum accumulations at the Oso gauge site, which were needed to scale the 349
dimensionless regional growth curve to the local precipitation quantities. The Oso at-site means 350
were estimated by regressing the means from a larger set of 21 COOP precipitation sites against 351
PRISM (Parameter-elevation Relationship on Independent Slopes Model, 800m 1981-2010 352
climate normals; Daly et al. 1994) mean annual precipitation, as well as against elevation. The 353
relationship between at-site means and these two predictors was strong (mean R2 = 0.97), so we 354
concluded that these regression equations were appropriate for estimating the unknown at-site 355
means at the Oso gauge. Once the Oso at-site means were estimated, the regional growth curves 356
were rescaled to reflect them, and the 2014 accumulations from the Oso precipitation gauge were 357
used to estimate the return periods of the accumulations according to the regional curve. 358
c. Soil moisture 359
We examined soil moisture model output from the University of Washington Drought 360
Monitoring System for the Pacific Northwest 361
Page 21
21
(http://www.hydro.washington.edu/forecast/monitor_west) which has a 1/16° (latitude-362
longitude) spatial resolution. The system simulates daily soil moisture in near real-time using the 363
Variable Infiltration Capacity (VIC) macroscale hydrologic model (version 4.1.2g; 364
documentation and code are available from hydro.washington.edu/Lettenmaier/Models/VIC/.). It 365
also uses a long-term climatology simulation (1916-2013) to provide the basis for estimating soil 366
moisture percentiles. We retrieved historic simulated soil moisture from the climatology 367
simulation, and 2014 values from the real-time estimates. The VIC model, which attempts to 368
approximate the physical processes that govern the accumulation of moisture in the soil column, 369
is driven by observed daily precipitation and temperature maxima and minima from 370
approximately 725 stations across the region. These stations are selected for reporting reliably in 371
real-time and for having lengthy historical records (at least 50 years in most cases). The station 372
data were obtained from the NOAA Applied Climate Information System and Environment 373
Canada. Oso is located nearly in the center of a 1/16 degree grid cell, so we used the modeled 374
soil moisture data from 1920 to 2013 for this grid cell. The 10-year return period curve shown in 375
Fig. 4 was estimated empirically by interpolating the annual soil moisture values for each day in 376
the period of simulation to the 0.9 cumulative probability level. We emphasize that the model 377
soil moisture provides a regional index at a relatively coarse spatial resolution (about 6 km in the 378
north-south direction); it does not attempt to represent local effects such as topography, and is 379
subject to a number of simplifying assumptions summarized in Liang et al. (1994). 380
d. Baseflow analysis 381
In addition to precipitation observations and reconstructed soil moisture, we examined 382
daily streamflow observations on the North Fork of the Stillaguamish River near Arlington 383
Page 22
22
(waterdata.usgs.gov/nwis/inventory/?site_no=12167000) for evidence of elevated groundwater 384
storage prior to the landslide. Following Brutsaert (2008, 2010), 385
𝑆 = 𝐾𝑦, (S1) 386
where S is subsurface storage, y is baseflow, and K is a time scale parameter. Hence, 387
groundwater storage in a catchment is proportional to baseflow at the outlet. Both elevated base 388
flow and reduced baseflow recession rates can indicate high levels of groundwater storage 389
(Konrad 2006; Brutsaert 2008). As in Brutsaert (2010), we examined annual minimum 7-day 390
average flows as a proxy for the baseflow rate y. 391
Fig. S1a shows the minimum 7-day low flows through water year 2013 (the water year 392
2014 minimum occurred in the summer after the landslide, hence is not relevant). In the interest 393
of partitioning the low flow period (some years have their lowest flows in October) we analyzed 394
an “extended water year” that ended on 31 October. Fig, S1a shows that the analysis was 395
inconclusive – although the low flows for the several years prior to the landslide were slightly 396
higher than the average of the previous two decades, they are not particularly unusual in the 397
context of the 85-year record of low flows, which also appears to have a fairly strong downward 398
trend. In contrast to the low flows, the date of their occurrence (Fig. S1b) appears to have no 399
trend, although there perhaps is some interannual persistence. 400
The absence of elevated baseflow in the recent wet period may be the result of several 401
factors. First, the annual minimum flows may be indicative of regional groundwater during the 402
dry recession period, rather than reflecting conditions at the much wetter time of the landslide. 403
Second, the lowest flows appear to be somewhat related to the length of the summer dry period 404
(Fig. S1c), although the relationship is not statistically significant. Finally, the low flows reflect 405
conditions over the entire watershed, particularly from the high-elevation portions of the basin 406
Page 23
23
from which most of the flow is contributed, and less so from its downstream reaches in the 407
vicinity of Oso. We attempted to locate wells in the region that would support a more direct 408
analysis of groundwater levels, but were unable to locate any observation wells (as opposed to 409
pumping wells) with sufficient observations to support such an analysis. 410
411
References 412
Brutsaert, W., 2008: Long-term groundwater storage trends estimated from streamflow records: 413
Climatic perspective. Water Resour. Res., 44, doi:10.1029/2007WR006518. 414
——, 2010: Annual drought flow and groundwater storage trends in the eastern half of the 415
United States during the past two-third century. Theor. Appl. Climatol., 100, 93–103, 416
doi:10.1007/s00704-009-0180-3. 417
Daly, C., R. Neilson, and D. Phillips, 1994: A statistical-topographic model for mapping 418
climatological precipitation over mountainous terrain. J. Appl. Meteorol., 33. 419
Hosking, J. R. M., and J. R. Wallis, 2005: Regional frequency analysis: an approach based on L-420
moments. Cambridge University Press,. 421
Konrad, C. P., 2006: Longitudinal hydraulic analysis of river-aquifer exchanges. Water Resour. 422
Res., 42, doi:10.1029/2005WR004197. 423
Liang, X., D. P. Lettenmaier, E. F. Wood, and S. J. Burges, 1994: A simple hydrologically based 424
model of land surface water and energy fluxes for general circulation models. J. Geophys. 425
Res., 99, 14415–14428. 426
Perica, S., and Coauthors, 2013: NOAA Atlas 14, Volume 9, Version 2: Precipitation-Frequency 427
Analysis of the United States, Southeastern States. NOAA, National Weather Service, Silver 428
Spring, MD,. 429
Schaefer, M., B. Barker, G. Taylor, and J. Wallis, 2002: Regional Precipitation-frequency 430
Analysis and Spatial Mapping of Precipitation for 24-hour and 2-hour Durations in 431
Western Washington. 432
433
Page 24
24
434 435
Fig. S1 – Period of record annual 7-day low flows for North Fork of the Stillaguamish River 436
near Arlington (USGS station 12167000): a) annual 7-day low flow (by water year); b) date of 437
occurrence; c) annual 7-day low flow versus its preceding dry period length (defined as the 438
number of days before the occurrence of low flow when daily precipitation was less than 20 439
mm). The R2 of the correlation between low flow and dry period lengths is 0.027 (not statistically 440
significant). 441