https://orcid.org/0000-0001-8743-3949
Figures
Abstract
The flooding of 1861-1862 in California and Oregon is the most severe flood event documented in the far western USA and stands as a benchmark for a worst-case atmospheric-river flooding event. In western Washington, historical data are sparser, and 19th-century flood events have consequently not been well documented. We found that rainfall observations from five locations spanning western Washington had no detectable bias when compared to nearby 20th and 21st-century comparator stations. Time series of the four-day precipitation sum revealed an event in December 1867 that was greater than any of the last century at three locations, and in the top two events at the other two locations. Summing over all locations, the regional three-day or four-day peak precipitation in 1867 exceeded the 150-yr recurrence magnitude by nearly 150 mm, indicative of non-stationarity of precipitation extremes. Newspapers and historical accounts document flood damage to settlements, farms, and bridges from the Columbia River to central Puget Sound. Reported high water levels at two locations indicate floodplains under more than a meter of water. Reanalysis data (20CRv3) is poorly spatially constrained in 1867, and underestimates the magnitude of this event, but it clearly shows the atmospheric-river cause of the event and supports snowmelt as a significant contributor to flooding. Compared to the most recent extensive flooding in 1996, the 1867 floods were likely of a similar extent but centered further north, and with notably more precipitation and enhanced by snowmelt. The 1867 rainfall amounts were also greater than those produced by the 2006 atmospheric river, though flooding in 2006 was not enhanced by snowmelt and record stream discharges were limited to mountain catchments. The combined rainfall and flood evidence from 1867 shows the potential for events more extreme than have occurred in recent history in the major urban corridors of western Washington.
Citation: Gavin DG, Bartlein PJ, Mock CJ (2023) Historical archives reveal record rainfall and severe flooding in December 1867 resulting from an atmospheric river and snowmelt, western Washington, USA. PLOS Clim 2(12): e0000324. https://doi.org/10.1371/journal.pclm.0000324
Editor: Ahmed Kenawy, Mansoura University, EGYPT
Received: September 7, 2023; Accepted: November 14, 2023; Published: December 19, 2023
Copyright: © 2023 Gavin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Daily precipitation data from GHCN Daily Summaries (version 3) can be found at https://doi.org/10.7289/V5D21VHZ. Data and R code used to generate Figs 2–6 and 9 can be found in the Supplemental Materials. Maps in Figs 1, 7, and 8 were made in QGIS with base map layers from map services and data available from U.S. Geological Survey, National Geospatial Program. Data layers are the US Geological Survey Small-Scale Hydrographic data set (https://www.usgs.gov/programs/national-geospatial-program/small-scale-data) and the 3DEPElevation tinted shaded relief from The National Map server (https://apps.nationalmap.gov/services/). In Fig 7, the following public domain data sources were used: 1) precipitation data from PRISM (https://prism.oregonstate.edu), 2) gage basin boundaries were from the BASINS II database (GAGES-II: Geospatial Attributes of Gages for Evaluating Streamflow; https://doi.org/10.5066/P96CPHOT), and 3) snow-water equivalent data are from the Natural Resource Conservation Service National Water and Climate Center (https://www.nrcs.usda.gov/wps/portal/wcc/home/). Lidar elevation data in Fig 8 can be found at the Washington State Lidar Portal (https://lidarportal.dnr.wa.gov). Fig 9 and Animation S1 in supplementary materials were based on data from the NOAA/CIRES/DOE 20th Century Reanalysis (V3), downloaded from the National Oceanic and Atmospheric Administration, Earth System Research Laboratories, Physical Sciences Division at https://psl.noaa.gov/data/gridded/data.20thC_ReanV3.html. Base-map shapefiles were calculated in R from the reanalysis model’s land-ocean mask. Moisture-flux calculations and images were made using the NCAR Command Language (NCL).
Funding: The author received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Atmospheric rivers (ARs), transitory and narrow (300-500 km wide) corridors of poleward water vapor transport, are a major cause of flooding along the western margins of midlatitude continents [1, 2]. Climate-model projections indicate that in a warmer world the frequency of ARs will increase [3, 4]. Along the Pacific coast of North America, AR intensity has been increasing since the 1950s, linked to increasing sea-surface temperature [5]. In western Washington, over 95% of peak river flows occur following intense winter rainfall from AR events when air temperatures are high enough such that most orographic precipitation falls as rain, often onto existing snowpack in the headwaters of the Cascade Range [6]. Such “rain-on-snow” floods impact suburban and urban areas along Puget Sound, where there is a large area (>10 km2) of developed land on alluvial fans and higher floodplains (Fig 1; [7]). The projection of increased AR-caused flooding in this region is a significant concern [8, 9].
Modern historical comparator sites (Global Historical Climate Network) are listed in Table 1. River abbreviations from north to south are Sn: Snohomish, Is: Issaquah Creek, Du: Duwamish, Ce: Cedar, Gr: Green, Wh: White, Pu: Puyallup, Ni: Nisqually, De: Deschutes, Ch: Chehalis, Co: Cowlitz, Le: Lewis. Snotel stations are ShCa: Sheep Canyon, SkCr: Skookum Creek. Map was developed in QGIS using base layers from the U.S. Geological Survey, National Geospatial Program, downloaded from https://apps.nationalmap.gov/downloader/ and https://elevation.nationalmap.gov:443/arcgis/services/3DEPElevation/ImageServer/WMSServer.
The 14 Dec 1867 event is highlighted in bold. Data for Fort Steilacoom, Fort Canby, and Fort Vancouver were obtained from the Midwestern Regional Climate Center (see footnote for data pretreatment). Data for American Camp was entered from digital scans of handwritten ledgers. GHCN-Daily stations were downloaded from https://www.ncdc.noaa.gov/cdo-web/ Site locations are shown in Fig 1.
Historical rainfall and flood events are important for understanding and assessing long-term trends in extreme events [10–12] and constraining frequency-magnitude relationships otherwise built on shorter records [13]. Reconstructing these events are also important for understanding recent flood history that may have left geomorphic evidence on floodplains [14], including in western Washington [15]. For example, the major flooding in 1861/1862 in central California has helped inform the “ARKstorm scenario” of converging flood-related disasters [16]. While the storms of 1861 were similarly intense in western Oregon [17], further north along the Pacific Coast historic records have not been collated to summarize similarly severe AR events. In this study, we summarized precipitation records from mid-1800s military forts in western Washington state. Meteorological logs from these sites provide daily observations of rainfall amounts during the first decades of Euro-American settlement. We compared these rainfall amounts to >100 yr of precipitation at nearby sites and assess potential bias in the historic data. We identified a single event in December 1867 by its combined magnitude and extent. We compiled newspaper and book accounts from this flood and compare the reported high-water levels in 1867 with current flood hazard estimates. We also assessed its significance with respect to recent large AR events using rainfall and snowmelt observations and reanalysis data.
Methods
Validation and analysis of 19th century precipitation observations
We examined daily weather logs from the 19th century at six locations (mainly U.S. military forts) in western Washington and northwestern Oregon (Fig 1). Measurements were likely made using a DeWitt conical rain gauge or a cylindrical rain gauge with a diameter between six and nine inches (150 to 225 mm). Guidance from the Surgeon’s General office indicated that rain gauges were to be located in clear openings and kept at least eight feet (2.44 m) high [18], which is considerably higher than the modern standard of 0.58 m [19]. This would likely result in about 10% reduction of rain measured in the gauge due to wind effects, thus providing a conservative low-end measurement of precipitation [20]. Daily data from three U.S. military forts (Fort Canby, Fort Vancouver, and Fort Steilacoom) were obtained from the Midwestern Regional Climate Center (https://mrcc.purdue.edu/) and subject to several quality control criteria and outlier verification [21]. As the Midwestern Regional Climate Center dataset does not yet contain all available data, we keyed in additional data from scanned images of original log sheets obtained from the U.S. National Archives from American Camp (San Juan Island), Fort Townsend, and, for 1865-1870, Astoria. The Astoria record is from the U.S. Coast Survey, and its precipitation gauge practices did not have the abnormally high gauge placement problem as the military forts. We conducted additional extreme-value checking by examining comments and temperature values on the data ledgers on the days with high rainfall. On four days (at two locations), high precipitation values were accompanied by a comment of “snow” and temperatures consistent with snowfall. On those days we reduced values to 10% of the reported value assuming a 10:1 snow water ratio.
The 19th century data were compared to nearby Global Historical Climatology Network (GHCN) stations (Fig 1; [22]). Comparator stations were selected as the longest records within several km from the 19th-century observations. For the Astoria comparators, we combined two GHCN stations (Astoria, 1892-1960; < 1 km away; Astoria Regional Airport, 1960-2023; 5 km away). For Fort Canby, American Camp, and Fort Vancouver, the comparator GHCN stations were located 8 km, 23 km, and 6 km, respectively, from the historic observations. For Fort Steilacoom comparators, we combined data from Puyallup 2W Experiment station (1914-1947; 17 km from Fort Steilacoom) and SeaTac Airport (1948-2023; 34 km from Fort Steilacoom). The GHCN daily data were processed as follows: 1) No observations were used if data quality flags indicated potential error, 2) 0.25 mm was substituted for days flagged as trace rainfall, 3) 0 mm was substituted for days flagged as ‘missing presumed zero’. Such flags occurred on a total of 11 days among three stations. An extreme-value checking of the GHCN data, as for the 19th century data, did not identify any suspicious values.
We assessed bias in the 19th century data in two ways. First, we plotted cumulative precipitation for the wet season (01 November to 31 March) for each year in the 19th century data with respect to quantiles and extreme values of cumulative precipitation at the comparator GHCN stations. Only years with observations on all 151 days of the wet season were included. We tested for differences in the median November-March cumulative precipitation between the 19th century data and GHCN data, in which each water year is an observation, using the Mann-Whitney U test. Second, we examined the statistical distribution of wet-day precipitation for the entire set of days with precipitation > 0.5 mm in the 01 November to 31 March period. For each site, we plotted the cumulative distribution for the set of 19th century daily observations and for the GHCN comparator data with respect to their exceedance probability. Differences in mean values may indicate non-stationary climate rather than observer bias, but we expect similarities in the shape of the distributions.
We summarized rainfall using three-day and four-day sums because of their strong correlation with peak streamflow in the region [23]. Frequency-magnitude plots of precipitation sums at the GHCN stations were calculated using the set of the n largest three-day and four-day sums, with no overlap of days, for the n years of record at each station. The resulting frequency-magnitude relationship is not based on annual maximum values, as more than one event may occur in any year. This “partial duration series” produces a more linear power-law relationship than produced by annual-maximum series [24]. Data and analyses are in S1 File.
Analysis of the 19th century data indicated that the greatest magnitude event, at four of the six sites, was during 12-15 December 1867. At Fort Townsend, the 14 December precipitation was not recorded, and it was only recorded for a six-day period at Fort Canby. We compared this event to the GHCN comparator stations by plotting it on the frequency-magnitude relationships constructed from the GHCN data. In addition, a regional rainfall precipitation series from the GHCN stations was constructed from the four locations with daily measurements through the 1867 event. The precipitation for the comparator sites of these four stations was summed daily over their common period (January 1914 through May 2016), then the three-day and four-day precipitation was calculated as for the individual stations and contrasted to the combined magnitude of the peak three-day (13-15 December) and four-day (12-15 December) precipitation in 1867.
Trend analysis
The long continuous precipitation series at four sites with daily observations through December 1967 (i.e., excluding Fort Canby) were analyzed for significant monotonic trends in 1) annual precipitation, 2) peak three-day precipitation sums, and 3) peak four-day precipitation sums. Each of these statistics was tested over three time periods: 1920-2020, 1970-2020, and all available data including the 19th century data. Annual precipitation was included only for years with fewer than three missing days. The number of peak three- and four-day precipitation events was set to the number of years of data in each data set (i.e., number of years of days with non-missing values). The Kendall test statistic was applied to each series. In cases with significant or nearly-significant trends (two-sided P value < 0.10), the slope of the trend was estimated using Thiel-Sen robust regression [25].
Newspaper and book accounts of flooding
Following the identification of the December 1867 event, we examined scanned data ledger sheets for that month from Fort Steilacoom, Fort Vancouver, Fort Canby, and American Camp for wind, temperature, and other observations. We compiled evidence of flooding from primary newspaper and book accounts, which provide important documentary information of flood locations, intensity, and damage [12]. We used the Library of Congress online archive (https://chroniclingamerica.loc.gov) to find flood reports in newspapers. We contacted two historical societies (White River Historical Society and Tolt Historical Society) to find additional sources. All locations of destroyed bridges, sawmills, homes, and farms were georeferenced. All notes of water levels relative to buildings or farms were converted to elevation based on the locations of the buildings as shown on General Land Office survey maps from the 1860s (https://glorecords.blm.gov) and lidar topography.
Comparing the 1867 flood to the 1996 and 2006 floods
Two recent major floods, in 1996 and 2006, provide context for the floods described in 1867. We mapped four-day precipitation sum [26], four-day snowmelt from the United States Department of Agriculture Snow Telemetry (SNOTEL) network, and the United States Geological Survey streamgage peak discharges. The peak discharges at each streamgage were ranked only for streamgages in western Washington and western Oregon with ≥ 40 years of observations between 1960 and 2020. In addition, we calculated the topographic enhancement of rainfall for the 1996 and 2006 events by comparison of low elevation (GHCN stations) with the SNOTEL network.
Reanalysis assessment of the meteorological controls of the 1867 flood event
We used data from the NOAA-CIRES-DOE Twentieth Century Reanalysis version 3 (20CRv3) data set [27, 28]. This reanalysis includes 3-hourly values of dozens of variables that describe the state of the atmosphere and land surface. These data are produced by assimilating surface-pressure observations, and prescribed sea-surface temperature, sea-ice concentration, and atmospheric radiative forcing as input to the National Centers for Environmental Prediction (NCEP) Global Forecast System Model. We used a subset of data archived at the NOAA Physical Sciences Laboratory (NOAA PSL) for the interval 1836-2015 (https://www.psl.noaa.gov/data/gridded/data.20thC_ReanV3.html). We used the National Center for Atmospheric Research (NCAR) Command Language (NCL, version 6.6.2) to calculate atmospheric moisture-flux variables, including vertically integrated moisture flux and moisture-flux divergence, and to map the data. Because the 20CRv3 data set at NOAA PSL does not provide values for snowfall, we estimated the fluxes of liquid water (rain) and snow using the approach of Hostetler and Alder [29], with Train = 4.0°C and Tsnow = -3.0°C.
There are a number of approaches for automated identification of rain-on-snow (ROS) and atmospheric-river (AR)/integrated-vapor-transport (IVT) events (i.e. moisture-flux events). For example, one approach for identifying ROS events uses daily reanalysis data and defines events as days with snow-cover fraction greater than 0.5 and total daily precipitation greater than 10 mm [30], while another uses daily surface observations and a simple definition of precipitation occurrence accompanied by a snow-depth decrease [31]. Automated approaches for recognizing ARs focus on recognizing the filamentous plumes of water vapor that are the striking characteristic of ARs [5, 32]. We found that when applied to 3-hourly data, the existing ROS-event procedures differed radically in the number of events identified, and it was beyond the scope of this project to create a catalog of IVT events. Consequently, we created two ad hoc event indices as follows: ROS events were defined when: (a) water-equivalent of snow depth was greater than 0.0 mm, (b) snow-cover percentage was greater than 30.0%, (c) precipitation was greater than 0.0 mm d-1 and exceeded snowfall rate, (d) snow phase-change heat flux was greater than 5.0 W m-2, and (e) 2-m air temperature was greater than 0.0°C. Conditions (a) and (b) assure that there is sufficient snow that, if melted, would produce runoff, condition (c) indicates that precipitation is not strictly snow, and (d) and (e) indicate that snow is melting. Moisture-flux (IVT) events were defined when: (a) moisture-flux magnitude (or IVT) was greater than 100 kg m-1 s-1, (b) 700 hPa vertical velocity (omega) was negative (signaling rising motions), (c) precipitation rate exceeded 5 mm d-1, and (d) moisture-flux divergence was negative (i.e. convergence of moisture was occurring). Condition (a) assures that atmospheric moisture levels are high, conditions (b) and (c) indicate that the synoptic situation is conducive for precipitation and that it is indeed occurring, and condition (d) allows us to discriminate between cases when high atmospheric moisture continent is derived from local sources as opposed to remote ones (via atmospheric rivers) [33]. Vertically integrated water vapor, often used to map atmospheric rivers, was not useful as an event indicator because it closely follows temperature over the year, which would require the definition of a seasonally varying threshold. Nevertheless, local (in time) peaks in vertically integrated water vapor help reinforce the designation of moisture-flux events. The threshold values listed above were determined by inspection of time-series plots of the data, including surface runoff. We also defined compound events when the conditions for both event indices were met.
Results
Validation and analysis of 19th century precipitation observations
Precipitation rate from the 19th century historic records compare well with the GHCN 20th and 21st century data (Fig 2). The cumulative precipitation of most years mostly fell within the central quartile of the GHCN data, with no clear bias towards high or low values. At four of the five 19th century precipitation records, the wet-season cumulative precipitation was not statistically different from comparator GHCN station records (Mann-Whitney U test, P>0.2). At American Camp, the 19th century median cumulative precipitation (340 mm) was less than at the comparator station (455 mm; Mann-Whitney U test, P=0.007). In addition, the cumulative distribution of daily precipitation was remarkably similar in slope and form for the 19th century and GHCN data (Fig 3). In four of the five comparisons, the distributions overlap over their lower range, and only depart in the highest 10% to 1% of days (generally > 20 mm), consistent with higher extreme precipitation in the 19th century. Only at Fort Steilacoom were observations greater at the lower end of the distribution (< 2 mm difference for days with < 5 mm).
Cumulative precipitation is restricted to the high-rainfall season from 01 November to 31 March. The thicker line is the 1867/1868 water year.
Data were limited to days with precipitation >0.5 mm and to the months of November through March, as in Fig 1.
The most extreme event emerging from analysis of three-day and four-day precipitation sums occurred in December 1867 (Fig 4). In Vancouver, WA, the three-day precipitation in December 1867 (194 mm), which was also reported in The Oregonian newspaper (S1 Text), was the greatest three-day sum compared to the most recent 121 years. At the mouth of the Columbia, Astoria’s three-day and four-day precipitation was greater than any since the GHCN record began in 1892, which exceeded the extrapolated 150-yr recurrence interval by > 40 mm. At nearby Fort Canby, only a single value was reported, representing a six-day period, but is consistent with data from Astoria. At Fort Steilacoom, near Tacoma, the three-day precipitation (122 mm) was the sixth-largest event when compared to Puyallup and Seatac Airport (107-year composite record), and the largest if using four-day sums (174 mm). At American Camp in the San Juan Islands, the second-greatest three- and four-day rainfall was recorded compared to the nearby 125-yr GHCN record. Overall, the 14 December 1867 AR produced the greatest or second-greatest four-day precipitation across western Washington, while no other of the top-five events occurred at two or more stations (Table 1).
Four-day precipitation sum is plotted in gray below the three-day sum. The December 1867 atmospheric river event is shown by points and dashed lines. The right column shows the frequency-magnitude relationship of three- and four-day precipitation from the weather stations. The frequency-magnitude plots use the n largest non-overlapping events from the n years of record. Fort Steilacoom and Fort Canby were compared to composites of nearby GHCN-Daily stations. Fort Canby only reported cumulative precipitation for 12-17 December 1867. See Fig 1 for site locations. Data coverage, pretreatment, and values of the largest five events at each site are in Table 1.
When summing data across four sites, the 14 December 1867 precipitation is an outlier compared to the GHCN regional composite (Fig 5). Both three-day and four-day precipitation sums from December 1867 exceed the largest GHCN regional precipitation sum (November 2006) by 150 mm, and exceed the regional February 1996 sum by ca. 200 mm. Furthermore, the 1867 magnitude falls well above (by ca. 120 mm) the frequency-magnitude plots extrapolated out to 150-yr return intervals.
The four-site daily time series was used to compute non-overlapping sums in three-day (green) or four-day (gray) periods, as in Fig 2. The end-dates of the two largest events are labeled. The precipitation summed over the four historical records for December 1867 are shown as dashed lines.
The daily progression of temperature and wind recorded three-times daily at five sites are consistent with the landfall of a warm atmospheric river (Fig 6). Southwest surface winds occurred at the mouth of the Columbia (Fort Canby) during the record rainfalls at nearby Astoria. Temperature rose from 7°C to 15°C at Vancouver during rainfall, though surface winds remained from the southeast, steered by the Columbia River gorge as is common at this site during winter. Thunderstorms, rare in atmospheric rivers, were reported on 14 December at American Camp and Fort Steilacoom, consistent with an unusually thick moist layer making landfall (Fig 6). Four days later, temperature dropped to freezing from Vancouver WA to American Camp, and snow was noted at sea level across the region, consistent with a cold front derived from the intense low pressure in the northeast Pacific that drove the atmospheric river.
Temperature and wind force and direction were measured three times daily. Wind force has unknown units. Cumulative rainfall curve in blue shows times of observations. Atmospheric pressure was recorded at Fort Steilacoom (black line, overlayed on precipitation curve). Rainfall at Port Townsend is not plotted because of a missing observation on 14 December. At Astoria, only precipitation data were available. S=snowfall, L=lightning observed.
Trend analysis
Two trends were detected in annual rainfall (S1 Table). At Astoria, annual rainfall decreased significantly from 1873 to 2022 over the 130 years with complete data (tau=-0.178; P=0.003, slope=-231 mm/100 yr). For the period from 1920 to 2020 at Astoria, annual rainfall decreased but not significantly (tau=-0.111; P=0.104, slope=-214 mm/100 yr). In contrast, annual precipitation from 1920-2020 marginally increased at Vancouver WA for the 90 years with complete data (tau=0.137, P=0.056, slope=132 mm/100 yr), but when considering the 124 years with complete data between 1850 and 2022, the trend was no longer marginally significant. There were no significant trends in peak three-day or four-day rainfall for all sites and time periods except for a marginally significant increasing trend in peak four-day rainfall from 1970 to 2020 at Seatac Airport (tau=0.183, P=0.059, slope=12.7 mm/50 yr).
Newspaper and book accounts of flooding
Historical archives (five newspaper articles and five books) claim that 14-15 December 1867 was devastating, and some claim it was the largest flood of the early settlement era (Fig 7A, S1 Text). High-water levels described in newspaper accounts could be located with high confidence using survey maps from the Bureau of Land Management. For example, accounts from the White River near present-day Kent WA (renamed Green River after the 1906 upstream avulsion of the White River into the Stuck River), stated that the water rose rapidly to seven feet (2.1 m) above the floodplain, which based on lidar topography implies that the full 3-km wide floodplain was inundated (Fig 8A). On the Cowlitz River at Longview WA, an account in The Oregonian stated that houses 30 feet (9.1 m) above mean levels survived but the pioneer town of Monticello was destroyed, implying widespread inundation of the lower Cowlitz River (Fig 8B; S1 Table). In contrast, the 1% probability flood on this reach of the Cowlitz River, from hydraulic models in the published flood insurance study reports (msc.fema.gov), is computed to the same elevation that we infer occurred from the 1867 flood accounts, though the 1% flood is constrained within levees. On Issaquah Creek, the detailed newspaper account of the water level also places the water level about 0.5 m above the 1% flood hazard (Fig 8C). On the Cowlitz River at the historic town of Olequa (Fig 8D), the same report in The Oregonian placed the 1867 water level well above the mapped 1% probability flood.
A) Western Washington and Oregon rivers, showing the locations of four-day rainfall observations for 12-15 December 1867 and mapped newspaper accounts of flood damage from 14-15 December 1867. Letter A-D show locations of reported high-water-levels: A: Issaquah Creek, B: Cowlitz River near Vader, C: White River near Kent, D: Cowlitz River near Longview. E: Skagit River near Concrete; date of the flood on Skagit River is uncertain. See S1 Text for all flooding accounts. Cities shown have newspaper accounts from 1867. B) Peak streamflow rank for the 05-08 February 1996 atmospheric river for USGS streamgage basins. Only streamgages with observations in 1996 and ≥ 40 yr of observations between 1960 and 2020 are shown. Contour lines show modeled four-day precipitation (PRISM 2023). The four-day change in snow-water-equivalent (square symbols) is from the SNOTEL network. C) Same as (B) but for the 04-07 November 2006 atmospheric river. Maps were developed in QGIS using base layers from the U.S. Geological Survey, National Geospatial Program, downloaded from https://apps.nationalmap.gov/downloader/ and https://basemap.nationalmap.gov:443/arcgis/services/USGSShadedReliefOnly/MapServer/WmsServer? and rainfall data from https://prism.oregonstate.edu.
Lidar elevation data from 2017 to 2021 show modified floodplains (dikes, elevated roads, and ditches) that were absent in 1867. A) Green/White River near Auburn; B) Cowlitz River near Longview; C) Issaquah Creek; D) Cowlitz River near Vader. Dashed line on profile b shows the channel depth. Locations of farms and houses were obtained from General Land Office survey maps and placed on lidar topography. Quotes on maps provide the basis for the inferred high-water elevation (blue areas on cross sections). The FIS 1% flood elevation are from hydrological models in published flood insurance study reports (msc.fema.gov), except for transect d, where only an approximate flood hazard zone was available. Note that dikes constrain the 1% probability floods at sites A (near Auburn WA) and site B (near Longview WA). Maps were developed in QGIS using elevation data from the Washington Lidar Portal (https://lidarportal.dnr.wa.gov) also available at https://apps.nationalmap.gov/downloader/.
Further north, on the Skagit River (Fig 1), an investigation into pre-settlement floods (pre-1878 in that region) used mud-stained tree trunks in connection with first-hand Native American accounts as evidence of flooding during recent years [34]. The mud-stained trees were reported to be very clear in 1878 but extremely faded in 1900. At Concrete WA (Fig 7A) the high-water mark from this event was 1.9 m higher and with 30% higher discharge than the next-greatest flood in 1897, and in 1878 was reported by Native Americans as devasting [34]. The year of the flood was estimated, using the age of a single riparian tree examined in 1923, to be ca. 1856 [34]. Given this poor basis for an age, an 1867 age of this record flood is possible.
Comparing the 1867 flood to the 1996 and 2006 floods
The topographic gradient of rainfall in combination with snowmelt observations for two recent floods shows that snowmelt may augment a moderate AR event to produce severe flooding. In 2006, there was a ca. 350% increase in the four-day precipitation sum from sea level to ca. 1150 m elevation (Seatac Airport to Skookum Creek Snotel: 125 to 472 mm; Vancouver to Sheep Canyon Snotel: 151 to 536 mm). In 1996, there was only a 220% topographic enhancement in the four-day precipitation sum (Seatac to Skookum Creek Snotel: 130 to 312 mm; Vancouver to Sheep Canyon Snotel: 193 to 437 mm). However, there was no measurable snowmelt at the Snotel sites during the 2006 AR, while in 1996 the added snowmelt (114 and 152 mm at Skookum Creek and Sheep Canyon, respectively) resulted in greater precipitation + snowmelt in 1996 than in 2006. The 1996 AR produced the record streamflow across much of southwest Washington and northwest Oregon (Fig 7). In contrast, the record streamflow of the 2006 event was mainly in the areas of most intense rainfall in the Cascade Range of Washington.
Reanalysis assessment of the meteorological controls of the 1867 flood event
We characterized the atmospheric conditions during the 1867 atmospheric-river event using the 20CRv3 reanalysis data set [28]. High values of vertically integrated moisture flux span 10° of latitude, and there was a loss of snow cover preceding and during the event (Fig 9; S1 Checklist). The peak precipitation rate in the reanalysis data for 1867 was lower than typical for recent atmospheric river events, likely because there is less support from the assimilated surface pressure data used in the earlier years of the reanalysis, and so we should therefore consider the reanalysis output as an underestimate of precipitation (S2 Text; [27]). This effect imparted an increasing trend before 1900 in rainfall amount in western Washington, which also resulted an increasing trend before 1900 in the number of ROS, IVT, and compound events (Fig B in S2 Text). However, the large-scale set up and day-to-day variations during 1867 are consistent with a compound large-scale atmospheric river and rain-on-snow event (Fig 9; S1 Checklist). The overall “synoptic situation” of the 1867 event is consistent with that of the more recent and better documented AR events of 1996 and 2006 [35].
Data are from NOAA-CIRES-DOE 20th Century Reanalysis (V3) provided by the NOAA/OAR/ESRL PSL, Boulder, Colorado, USA at https://www.psl.noaa.gov/data/gridded/data.20thC_ReanV3.html. The magenta square is centered on western Washington. Times are PST. The figure illustrates (left to right) the supply of moisture (moisture flux), and the large-scale atmospheric cooling mechanism (vertical velocity, rising implies cooling), that favors cloudiness and precipitation (precipitation rate), plus the additional contribution to the hydrologic response of snowmelt (change in snow cover).
Discussion
Record rainfall in western Washington
The observations at five historic forts indicate that the combined magnitude and extent of the December 1867 AR was greater than any in the historic record (Figs 4, 5). The four-day precipitation sum was remarkable across all western Washington (93 mm at American Camp in the San Juan Islands; 194 mm at Vancouver WA; 122 mm at Fort Steilacoom. American Camp and Fort Vancouver are 316 km apart, and Fort Steilacoom is approximately at the mid-point between these sites (Fig 1). This extremely wide swath of high rainfall contrasts with recent events, which are normally less than 200 km wide [23]. Indeed, the reanalysis data show the 1867 AR to have a wider landfall than the larger 1996 and 2006 ARs, though fewer observations constrain the 1867 simulation, preventing detailed comparisons of these events (Fig 9). The wide swath of high rainfall is consistent with the southwesterly flow shown in the reanalysis data, and observed at the times of the AR landfall on 14 December (Fig 6) as westerly flow would be blocked by the Olympic Mountains [6].
We rule out observer bias in the 1867 rainfall observations. Handwriting on the ledgers was clear and verified by an independent group [21]. For the record event in Vancouver WA, the precipitation amount on the ledger was reprinted in a newspaper (S1 Text). The magnitude and statistical distribution of daily precipitation was very similar to the nearby GHCN stations (Figs 2 and 3). Only Fort Steilacoom has a higher mean value across all percentiles of wet days (Fig 4), but the overall wet-season precipitation totals are not significantly different in comparison with the comparator GHCN stations. This greater discrepancy for Fort Steilacoom may be because the comparator GHCN stations were furthest from the 19th century station (17 km, or 34 km after 1948; Table 1), while the other GHCN stations were less than 10 km from the 19th century stations. This issue at Fort Steilacoom was not large enough to result in a significant time trend in the data (S1 Table) and may simply represent nonstationary climate rather than observation bias.
The documented impacts of the 1867 flood include lost livestock and destroyed bridges and mills from Seattle to Vancouver WA and very high water elevations (Fig 7A). European settlements had not yet begun north of Seattle, thus flooding may have been just as severe in northern Puget Sound, as record rainfall was recorded at American Camp to the north. At the four locations where water heights were noted, 1867 water levels were similar to or higher than the current 1% probability flood elevation, consistent with the greater magnitude of rainfall for this event than any in the last century (Fig 8). However, quantitative interpretation at each site is complicated by several factors. On the lower Cowlitz River (at the former town of Monticello on Fig 7A), the 1867 level was similar to the modern modeled 1% probability flood, despite that modern floods are constrained to the small cross section within dikes, suggesting a much larger discharge in 1867 (Fig 8B). Upstream dams have decreased peak flows on the lower Cowlitz River, greatly decreasing the 1% probability discharge. On the Cowlitz River at Pumphrey’s Hotel, the newspaper account suggests a very high water level and flooding of higher terraces well above the modern flood hazard (Fig 8D). However, only approximate methods have been used to map the 1% probability flood area on this reach of the Cowlitz River and recent high water levels for this reach (e.g., 1996 flood) could not be found. A constriction in the downstream valley could account for the exceptionally deep flooding on this reach of the Cowlitz (Fig 8D). On the White River, the inferred 2.1 m water level on the floodplain near Auburn is well above the modern 1% probability flood hazard elevation (Fig 8A). However, in 1906 the White River avulsed into the Stuck and Puyallup Rivers, and revetments have since maintained this route, such that this reach was renamed the Green River and the watershed has been reduced by ca. 60%, therefore significantly reducing the flood hazard. Lastly, on Issaquah Creek, the high-water elevation is accurately geolocated from land records and historic structures, indicating a higher water level than for the modeled flood (Fig 8C). However, past changes in valley cross-sectional area from fluvial fills, or the past presence of log jams, may have increased past flood heights, and recent channel modification (dredging, widening) of this urban stream has decreased flood risk.
Reconstructing the December 1867 AR
The 1867 AR had combinations of the 1996 AR (high snowmelt) and the 2006 AR (high rainfall) events. A topographic enhancement of 350% from near sea level to 1000 m, as occurred during the 1996 AR, applied to the four-day rainfall at Fort Steilacoom and Fort Vancouver, implies four-day sums at high elevations (>1000 m) of 550 to 650 mm across the Cascade Range from the Columbia River to central Washington. This is ca. 100 mm more than occurred during the 1996 event (Fig 7). According to the PRISM precipitation estimates, such intense precipitation only occurred in small areas during the 2006 AR. While there were no direct observations of snowmelt in 1867, newspaper reports implied snowmelt was a significant contributor to flooding (S1 Text). Indeed, 22 mm of rainfall at Fort Steilacoom, and 16 mm at Fort Vancouver, preceded the AR event.
The reanalysis data show that the December 1867 flooding appears to be the result of several compound rain-on-snow (ROS) and moisture flux (IVT) events (S1 Fig). Conditions in the region were relatively dry through the end of October 1867, with precipitation increasing gradually through November, accompanied by slight increases in snow cover. Following a several-day interval of well-below freezing temperatures, a noticeable rain and snow event produced a small runoff peak on 4 December. Although 2-m air temperature was just below freezing, a broad peak in snow phase-change heat flux is evident indicating that snowmelt was likely taking place. On 9, 11, and 14-15 December, compound ROS and IVT events can be seen. Each of these includes peaks in atmospheric moisture variables, precipitation occurring as a combination of rain and snow, peaks in snow phase-change heat flux, and decreases in snow cover. Snowmelt contribution to recent AR-related flooding has been documented in western Washington (Neiman) and British Columbia [36].
The synoptic meteorological situation that supports atmospheric river events can be seen in animations of mapped reanalysis data, as, for example, for the conditions prevailing at 13:00 PST on 14 December (Fig 9; S1 Checklist). At that time, there was a broad upper-level trough in the eastern Pacific, accompanied by a surface low in the Gulf of Alaska. These circulation features resulted in generally westerly flow aloft into the Pacific Northwest, with strong southwesterly flow at the surface. This pattern resulted in a strong flow of moisture into the region, accompanied by high values of column-integrated water vapor. The 700 hPa vertical velocity was negative, indicating large-scale rising motions supporting abundant precipitation were occurring, and 2 m air temperatures were above freezing. A similar combination of features prevailed on 23-24 December (registered in the fort observations in California and Oregon, but not Washington), although the upper-level flow was more zonal in character (S1 Fig).
Comparing the 1867 flood with other notable historic floods
We compared the 1867 western Washington flood to the early December 1861 floods in Oregon which are well-known as arguably the worst in Oregon history [17]. We find no historical evidence outside the Vancouver WA area of any river flooding in Washington State in 1861. In terms of highest three-day rainfall magnitudes, the 1867 event compares with that of the highest for 1861, centered from Nov. 30-Dec. 2. For 1861, three-day rainfall sums include Fort Hoskins (near Corvallis OR) at 185 mm and Fort Umpqua at 174 mm. The 1861 flood event had wet SE to SW winds with seasonally warm temperatures (near 14°C) typically characteristic of AR events [17]. Nine inches of snow occurred at Hood River OR, and 10 inches at La Grande OR within a week of the early December floods, suggesting that some melting snow could have contributed to flood heights despite being early in the winter season. More pronounced, however, for the 1861 event was a long continuous rain duration not as prominent as in 1867. For example, Fort Hoskins recorded measurable rain each day from Nov. 16-Dec. 3.
The set of historical rainfall records used in this study have poor coverage during the 1890s to 1910s (Fig 4) raising the possibility that an event in Western Washington may have been missed in our analysis. One detailed account by Stewart and Bodhaine [34] of the flood history on the Skagit River places the highest water levels with direct observations (since settlement in 1878) to a flood on 19 November 1897. However, rainfall records at Tacoma and Seattle at this time have a four-day precipitation of 60 mm, much less than record levels at those stations (S1 File). Stewart and Bodhaine [34] report the 1897 flood was notable for the rate it rose and fell (within 1.5 days), consistent with the mountainous terrain of the majority of the Skagit River watershed. We found no other reports of severe floods in western Washington within this period of low coverage in our time series.
Centennial scale trends in precipitation
Our analyses of precipitation showed few centennial scale trends in annual rainfall or in three- or four-day precipitation sums (S1 Table). The reanalysis data also showed no 20th century long-term trends in numbers of ROS, IVT, or compound events (Fig B in S2 Text). In contrast, Gerushnov et al. [5] found an increasing magnitude of IVT, related to increasing sea-surface temperature, since 1948. In addition, Sharma and Dery [37], using the AR catalog of Gershunov et al. [5], found an increasing trend in numbers of landfalling ARs in southern British Columbia. While the continental-scale analyses of Gershunov et al. [5] could identify an increasing trend in IVT, this trend did not impart a statistically significant trend in precipitation maximums at the four sites analyzed in western Washington. In addition, the >130-year records at Astoria and Vancouver revealed opposing trends in annual precipitation. These results are consistent with local expressions that contain too much local variance to reveal the regional long-term pattern.
Conclusions
Our analyses of primary data describing an intense AR with associated rain-on-snow in December 1867 were consistent in supporting this event as the largest recorded flood event in western Washington. The high rainfall observations and high water-level marks, combined with evidence of snowmelt, suggest this event had a combination of the severe rainfall of the 2006 AR and the rainfall/snowmelt of the 1996 AR. Reanalysis data was shown to be unreliable for precipitation magnitude in the 1860s but the precipitation patterns supported the rain-on-snow contribution to the 1867 flood. Our study shows that there is historical precedent for AR-flooding in western Washington greater than any observed over the last century, yet prior to the period of increasing moisture-flux related to increasing sea-surface temperatures. Thus, potential for more intense AR events exceeding the 1996 flooding has an existing precedent prior to the modern AR-producing climate system. The 1867 AR was large enough and covered a significantly large area that, if such an event were to occur in the urban corridors of western Washington, would cause significant damage and pose major challenges to emergency management.
Supporting information
S1 Checklist. Animation of reanalysis model for preceding and following the 14 December 1867 atmospheric river.
Data are output from the NOAA-CIRES-DOE 20th Century Reanalysis (V3) as described in the Methods main text. The animation is also available online, created using HAniS (HTML5 AnimationS, https://www.ssec.wisc.edu/hanis/). https://pages.uoregon.edu/bartlein/20CRv3/20CRv3_186712/20CRv3_186712.html.
https://doi.org/10.1371/journal.pclm.0000324.s001
(GIF)
S1 Fig. NOAA-CIRES-DOE 20th century reanalysis (V3) time series for western Washington, Nov 15 to Dec 31, 1867.
Data are from three-hourly time steps. We extracted data for a grid cell in Puget Sound. Precipitation and snowfall rate using the approach of Hostetler and Alder (Hostetler and Alder, 2016). All other values were extracted directly from model output.
https://doi.org/10.1371/journal.pclm.0000324.s002
(TIF)
S1 File. Data and R code for precipitation analyses and maps in Figs 2, 3, 4, 5, 9, S1 Table, and S1 Checklist.
https://doi.org/10.1371/journal.pclm.0000324.s003
(ZIP)
S1 Table. Kendall’s tau values for trend tests of precipitation time series.
n=number of years without missing values. All yrs includes all data in Fig 4. Ann=annual precipitation, 3d=peak three-day precipitation, 4d=peak four-day precipitation. Values in bold indicate significance at P<0.05.
https://doi.org/10.1371/journal.pclm.0000324.s004
(DOCX)
S1 Text. Newspaper and book accounts of the 14-15 December 1867 flood in western Washington.
The following excerpts were transcribed from scanned images of newspaper articles and from books. The quoted texts are followed by a summary of the relevance of the observation for flood interpretation and any discrepancies noted in the accounts.
https://doi.org/10.1371/journal.pclm.0000324.s005
(DOCX)
S2 Text. Trends in 20CRv3 precipitation data.
Our historical records indicate that the 1867 20CRv3 underestimates the intensity of the 1867 AR event.
https://doi.org/10.1371/journal.pclm.0000324.s006
(DOCX)
References
- 1. Gimeno L, Nieto R, Vázquez M, Lavers D. Atmospheric rivers: a mini-review. Front Earth Sci. 2014;
- 2. Payne AE, Demory M-E, Leung LR, Ramos AM, Shields CA, Rutz JJ, et al. Responses and impacts of atmospheric rivers to climate change. Nat Rev Earth Environ. 2020;1: 143–157.
- 3. Lavers DA, Ralph FM, Waliser DE, Gershunov A, Dettinger MD. Climate change intensification of horizontal water vapor transport in CMIP5. Geophys Res Lett. 2015;42: 5617–5625.
- 4. Wang S, Ma X, Zhou S, Wu L, Wang H, Tang Z, et al. Extreme atmospheric rivers in a warming climate. Nat Commun. 2023;14: 3219. pmid:37270607
- 5. Gershunov A, Shulgina T, Ralph FM, Lavers DA, Rutz JJ. Assessing the climate-scale variability of atmospheric rivers affecting western North America. Geophys Res Lett. 2017;44: 7900–7908.
- 6. Neiman PJ, Schick LJ, Ralph FM, Hughes M, Wick GA. Flooding in Western Washington: The connection to atmospheric rivers. J Hydrometeorol. 2011;12: 1337–1358.
- 7. Konrad CP. Geospatial Assessment of Ecological Functions and Flood-Related Risks on Floodplains along Major Rivers in the Puget Sound Basin, Washington. U.S. Geological Survey; 2015. Available:
- 8. Hamman JJ, Hamlet AF, Lee S-Y, Fuller R, Grossman EE. Combined effects of projected sea level rise, storm surge, and peak river flows on water levels in the Skagit floodplain. Northwest Sci. 2016;90: 57–78.
- 9.
Salathé EP, Mauger G. Climate change, heavy precipitation and flood risk in the western United States. In: Murphy C, Gardoni P, McKim R, editors. Climate Change and Its Impacts: Risks and Inequalities. Cham: Springer International Publishing; 2018. pp. 109–127. https://doi.org/10.1007/978-3-319-77544-9_7
- 10.
Lamb HH. Climate, History and the Modern World. 2nd ed. London: Routledge; 1995. https://doi.org/10.4324/9780203433652
- 11. Sutcliffe JV. The use of historical records in flood frequency analysis. J Hydrol. 1987;96: 159–171.
- 12. Brázdil R, Kundzewicz ZW, Benito G, Demarée G, Macdonald N, Roald LA. Historical floods in Europe in the past millennium. Changes in Flood Risk in Europe IAHS Special Publication 10. 2012; 21–166.
- 13. Slater LJ, Anderson B, Buechel M, Dadson S, Han S, Harrigan S, et al. Nonstationary weather and water extremes: a review of methods for their detection, attribution, and management. Hydrol Earth Syst Sci. 2021;25: 3897–3935.
- 14.
Toonen WHJ, Munoz SE, Cohen KM, Macklin MG. High-resolution sedimentary paleoflood records in alluvial river environments: A review of recent methodological advances and application to flood hazard assessment. In: Herget J, Fontana A, editors. Palaeohydrology: Traces, Tracks and Trails of Extreme Events. Cham: Springer International Publishing; 2020. pp. 213–228. https://doi.org/10.1007/978-3-030-23315-0_11
- 15. Gavin D. Annually-laminated lake sediment reveals late Holocene alluvial channel history and a record flood in 1867, Tolt River, western Washington. Geological Society of America; 2021. Available from: https://gsa.confex.com/gsa/2021AM/webprogram/Paper371459.html
- 16. Porter K, Cox D, Dettinger M, Martin Ralph F. Special Issue on the ARkStorm Scenario: California’s Other Big One. Nat Hazards Rev. 2016;17: A2016001.
- 17. Miller GR. The Great Willamette Flood of 1861. Oreg Hist Q. 1999;100: 182–207.
- 18.
Lawson T. Army Meteorological Register for Twelve Years from 1843-1854. Washington, D.C.: Surgeon General’s Office; A. O. P. Nicholson Public Printer; 1855.
- 19. Mock CJ. Drought and precipitation fluctuations in the Great Plains during the late nineteenth century. Gt Plains Res. 1991;1: 26–57.
- 20. Robinson PJ, Henderson-Sellers A. Contemporary Climatology. Longman; 1999.
- 21. Kunkel KE, Andsager K, Conner G, Decker WL, Hillaker HJ, Knox PN, et al. An expanded digital daily database for climatic resources applications in the midwestern United States. Bull Am Meteorol Soc. 1998;79: 1357–1366.
- 22. Menne MJ, Durre I, Korzeniewski B, McNeill S, Thomas K, Yin X, et al. Global Historical Climatology Network - Daily (GHCN-Daily), Version 3. NOAA National Centers for Environmental Information; 2012.
- 23. Warner MD, Mass CF, Salathé EP. Wintertime extreme precipitation events along the Pacific Northwest coast: Climatology and synoptic evolution. Mon Weather Rev. 2012;140: 2021–2043.
- 24. Kidson R, Richards KS. Flood frequency analysis: assumptions and alternatives. Progress in Physical Geography: Earth and Environment. 2005;29: 392–410.
- 25.
Theil H. A rank-invariant method of linear and polynomial regression analysis. Henri Theil’s Contributions to Economics and Econometrics Advanced Studies in Theoretical and Applied Econometrics, vol 23. Dordrecht: Springer; 1992. pp. 345–381. Available: https://doi.org/10.1007/978-94-011-2546-8_20
- 26. Daly C, Doggett MK, Smith JI, Olson KV, Halbleib MD, Dimcovic Z, et al. Challenges in observation-based mapping of daily precipitation across the conterminous United States. J Atmos Ocean Technol. 2021;38: 1979–1992.
- 27. Slivinski LC, Compo GP, Sardeshmukh PD, Whitaker JS, McColl C, Allan RJ, et al. An evaluation of the performance of the twentieth century reanalysis version 3. J Clim. 2021;34: 1417–1438.
- 28. Slivinski LC, Compo GP, Whitaker JS, Sardeshmukh PD, Giese BS, McColl C, et al. Towards a more reliable historical reanalysis: Improvements for version 3 of the Twentieth Century Reanalysis system. Q J R Meteorol Soc. 2019;145: 2876–2908.
- 29. Hostetler SW, Alder JR. Implementation and evaluation of a monthly water balance model over the US on an 800 m grid. Water Resour Res. 2016. p.
- 30. Cohen J, Ye H, Jones J. Trends and variability in rain-on-snow events. Geophys Res Lett. 2015;42: 7115–7122.
- 31. McCabe GJ, Clark MP, Hay LE. Rain-on-snow events in the western United States. Bull Am Meteorol Soc. 2007;88: 319–328.
- 32. Guan B, Waliser DE. Detection of atmospheric rivers: Evaluation and application of an algorithm for global studies. J Geophys Res Atmos. 2015;120: 12514–12535.
- 33. Dacre HF, Clark PA, Martinez-Alvarado O, Stringer MA, Lavers DA. How do atmospheric rivers form? Bull Am Meteorol Soc. 2015;96: 1243–1255.
- 34.
Stewart JE, Bodhaine GL. Floods in the Skagit River Basin Washington. Washington, D.C.: Geological Survey Water-Supply Paper No. 1527; 1961.
- 35.
Mastin MC, Gendaszek AS, Barnas CR. Magnitude and extent of flooding of selected reaches in western Washington, January 2009. Reston, VA: U.S. Geological Survey Scientific Investigations Report 2010-5177; 2010. https://doi.org/10.3133/sir20105177
- 36. Gillett NP, Cannon AJ, Malinina E, Schnorbus M, Anslow F, Sun Q, et al. Human influence on the 2021 British Columbia floods. Weather Clim Extrem. 2022;36: 100441.
- 37. Sharma AR, Déry SJ. Variability and trends of landfalling atmospheric rivers along the Pacific Coast of northwestern North America. Int J Climatol. 2020;40: 544–558.