The National Climatic Data Center and numerous other sources list the 15.20-in. (386 mm) rainfall observed at the Angoon, Alaska, cooperative weather station on 12 October 1982 as the state record for a single calendar-day precipitation amount. However, a close inspection of the precipitation data recorded during 1982 in Angoon reveals a pattern of suspect values, calling into question the validity of the data collected during this time period. Our analysis shows that errors may be present in the Angoon precipitation record, and therefore consideration should be given to removing those values from the official climate database. This study evaluates Angoon precipitation observations over a 12-month period starting in February 1982 using two objective analytical techniques (statistical and climatological) and a 2012 interview of the Angoon cooperative station observer during the time period in question. If the Angoon precipitation values are deemed erroneous, then the 12 October 1982 observation will no longer hold the distinction as the single largest precipitation event. The second place event is a 15.05 in. (382 mm) observation measured in Seward, Alaska, on 10 October 1986.
The validity of the Alaska record for single-day maximum precipitation is studied using two objective analytical techniques.
The climate record of Alaska and the rest of the United States is a vital tool for public and private decision makers. When should crops be planted to avoid freezing temperatures? When should concrete be poured for roads or other construction activities? How many resources should a government agency set aside for snow removal? How far above a river channel should development be permitted? These are a few of the many questions that are partially answered by analyzing the vast repository of climate data collected during the previous 100-plus years. In Alaska there are 21 first-order climate stations—some with high-quality, long-period data records. A few of these include Barrow, Nome, Fairbanks, Anchorage, Bethel, and Juneau. The remaining data are collected by volunteers in the National Weather Service (NWS) Cooperative Observer (COOP) program.
The NWS provides training and meteorological instruments to volunteers who wish to participate in the COOP program (NOAA 2010). The observers record data daily on a standard form (www.nws.noaa.gov/os/coop/forms/b-91.pdf) and then send the completed forms to the National Weather Service at the end of each month. The forms are eventually sent to the National Climatic Data Center (NCDC) in Asheville, North Carolina, which is responsible for preserving, monitoring, assessing, and providing public access to the data. This paper focuses on a single station in southeast Alaska, Angoon, whose COOP precipitation observations fall outside of climatological bounds and where one observation established the Alaska single-day precipitation record on 12 October 1982. We attempt to establish the validity of the observations at this station and whether the state-record precipitation event is erroneous.
The community of Angoon is located on the west side of Admiralty Island in southeast Alaska, approximately 100 km south of Juneau, Alaska (see Fig. 1). The Angoon region has been intermittently inhabited for over 3,000 years (De Laguna 1960). All of the residents of Angoon live on a peninsula of land bounded by Chatham Strait on the west and Favorite Bay on the east. The 2010 U.S. Census population estimate for Angoon showed 459 full-time residents (data taken from the U.S. Census Bureau; see http://factfinder2.census.gov/faces/tableservices/jsf/pages/productview.xhtml).
An understanding of the basic geography and climatology of Alaska is useful for contextualizing the observed occurrence of specific precipitation events. There are many climatic descriptions and classification systems for Alaska (Fitton 1930; Evangelista 1991; Shulski and Wendler 2007; Bieniek et al. 2012). To describe large-scale precipitation patterns, we use six broad regions. Northern Alaska and interior Alaska are areas of low precipitation. The orographic barriers of the Brooks Range and Alaska Range and low annual temperatures combine to inhibit moisture. Western Alaska and the Aleutian Islands experience frequent, stratiform precipitation. Very few instances of 100-mm daily precipitation events have been recorded in these regions. The southern portion of mainland Alaska and the entirety of southeast Alaska are prone to much larger precipitation events. Proximity to the Gulf of Alaska, prevailing storm tracks, and windward orographic effects result in large precipitation events for many locations. However, many local orographic effects inhibit precipitation in leeward locations. Figure 2 shows the normal annual precipitation for all stations near Angoon.
Angoon is situated in the expansive temperate rain forest of southeast Alaska that is dominated by Sitka spruce (Picea sitchensis) and western hemlock (Tsuga heterophylla). The cool, moist flow from the eastern North Pacific Ocean supplies copious rain and winter snow to all locations. Differences in local topography dramatically alter precipitation patterns across relatively small horizontal distances. The wettest portions of southeast Alaska receive over 200 in. (5080 mm) of annual precipitation, while the driest portions receive substantially less. All of southeast Alaska receives the greatest proportion of annual precipitation during the months of September, October, and November (40% of the annual total). Angoon lies in the rain shadow of mountainous Baranof Island and has the lowest annual precipitation of any location in southeast Alaska with a reliable climate record. The climate of Angoon is best described as cool and moist but not overly wet. The average July temperature is 56°F (13.4°C) and the average January temperature is 27°F (−2.5°C). Approximately 40 in. (1016 mm) of precipitation (rain and melted snow) are observed and 61 in. (1549 mm) of snow are recorded in a typical year (Schwartz and Miller 1983). However, close inspection of the Angoon climate record reveals a small period of time with precipitation observations that appear anomalously large. Those precipitation observations are the subject of the remaining sections of this paper.
ANGOON POWER STATION AND THE 1981–83 PRECIPITATION OBSERVATIONS.
A continuous record of weather observations exists for Angoon between 1941 and 1985. The observations were recorded on the western side of Angoon immediately adjacent to Chatham Strait at the Angoon Assembly of God church. The station's formal name was Angoon Power Station. The site coordinates were 57°29′56″N, 134°35′11″W, and the station was located at an elevation of 8.5 m. The station burned down in the 1990s, but a cousin of the last observer knew the location and escorted the lead author to the site in August 2013.
Several community volunteers recorded weather observations in support of the COOP program over the years in Angoon. These observations were typically very reliable; however, the period between August 1981 and June 1983 has precipitation observations far larger than are expected based on the recorded climatology for the location. Beginning in August 1981, a new observer was responsible for all measurements until June 1983, when observations were suspended.
To demonstrate the data integrity issues, we focus on a 12-month period beginning 1 February 1982—from now on called 1982 data. We chose this time period to highlight systematic issues with the data. The data were transcribed directly from the scanned E-15 forms and compared to the official data published by the NCDC. Both sets of data were identical, thereby eliminating the possibility of transcription error.
A key technique for establishing climate observation reliability is to perform statistical analysis on the data to determine the probabilistic likelihood of those observations occurring. The National Climatic Data Center performs statistical data assessments on all climate data (Reek et al. 1992; Peterson and Vose 1997; Durre et al. 2010; Menne et al. 2012). These techniques are discussed in greater detail later in this paper. Several other studies have identified additional methodological techniques for flagging suspicious precipitation observations (Eischeid et al. 1995; You et al. (2007); and others). In those studies, observations that fell outside a statistical threshold were identified as suspicious based on the probabilistic likelihood of occurrence.
In the current study, we first looked at the number of standard deviations that the 1982 recorded monthly totals deviate from corresponding long-term averages. Table 1 shows the long-term average precipitation, in inches, for Angoon prior to 1981. This study uses “average” as opposed to “normal” precipitation due to the lack of published normal tables during the study period in question.
The discrepancies between the 1982 values and the long-term average values are immediately noticeable. In fact, the 1982 precipitation of 213.29 in. (5418 mm) in Angoon is more than 5 times greater than the average annual precipitation. This is 25.10 standard deviations above the average annual precipitation. For normally distributed data, 95% of observations fall within two standard deviations of the mean. While annual precipitation is not precisely normally distributed, the Angoon precipitation values between 1950 and 1980 closely resemble a bell curve.
When the National Oceanic and Atmospheric Administration (NOAA)'s Hydrometeorological Design Studies Center reassessed collected precipitation data for a frequency project for Alaska, which was done jointly with the University of Alaska Fairbanks (Perica et al. 2012), they noticed a number of high outlier values at the Angoon station. The values were far outside the expected range, and their inclusion in the analysis dramatically affected precipitation frequency estimates, which represent precipitation magnitudes associated with specific average recurrence intervals or annual exceedance probabilities for a given location. When all recorded 1-day data were used, the 24-h, 100-yr precipitation estimate for Angoon was approximately 9.0 in. (229 mm). When suspect observations were removed, for reasons discussed later, the 24-h, 100-yr precipitation estimate lowered to 3.64 in. (92 mm), which is more in line with frequency estimates for nearby locations, as it is comparable to the 100-yr estimate of 3.53 in. (90 mm) from the study by Miller (1963). Estimates of 24-h precipitation frequency for 100- and 1000-yr average recurrence intervals (1% and 0.1% annual exceedance probabilities, respectively) from Perica et al. (2012) are shown in Fig. 3. As shown in the right panel of the figure, the 1000-yr estimate for Angoon is approximately 116 mm or 4.57 in. The NOAA Atlas 14 precipitation frequency values provide a meaningful baseline from which to assess the precipitation values collected in Angoon during the study time period in question.
From February 1982 to the end of January 1983, the recorded observations in Angoon exceeded the 24-h, 100-yr precipitation event 17 times and the 1000-yr precipitation amount was exceeded 9 times—even though the probability that the 1000-yr amount is exceeded in any year is only 0.1%. Also, the 12 October 1982 precipitation amount of 15.20 in. (386 mm) exceeded the 24-h probable maximum precipitation (PMP) amount of 14 in. (356 mm) from the report by Schwartz and Miller (1983), which represents the theoretically greatest depth of precipitation for a given duration that is physically possible over a given size storm area at a particular geographical location at a certain time of year. PMP values are no longer calculated by NOAA, but the exceedance of this theoretical value is noteworthy.
Precipitation events of that magnitude all in the same year are statistically improbable. Four of the precipitation events are flagged in the Global Historical Climatology Network (GHCN) database as climatological outliers. The day 12 October 1982 was one of the four calendar days whose precipitation readings were flagged. Nevertheless, it remains in the Alaska single-day precipitation record.
Four stations in close proximity to the Angoon Power Station site are used in this study to assess the regional validity of Angoon's data during the 12-month period ending on 31 January 1983. The stations are 1) Sitka Airport, 2) Juneau Airport, 3) Little Port Walter, and 4) Five Finger Light Station. Each station is located within 130 km of Angoon, and each station represents a variety of local topographic settings that influence precipitation patterns slightly differently. Two stations exhibit windward characteristics (Sitka Airport and Little Port Walter), while the remaining stations exhibit topographic sheltering caused by mountainous terrain and are only moderately exposed to moisture-bearing winds (Juneau Airport, Five Finger Light Station, and Angoon Power Station).
A summary of the 1982 monthly data for all five stations is shown in Table 2. For each station the average monthly precipitation for the 1950–80 period is shown on the left and the 1982 measured precipitation is shown on the right. Each of the stations, except for Angoon, observed below-average precipitation during this 12-month period. In several cases the deficit was substantial. Nearly every month at Angoon showed significantly higher-than-average precipitation.
When data are normalized using standard deviations, the variability in monthly precipitation values are transformed into easily comparable values. For example, in 1982, Little Port Walter received 182.94 in. (4647 mm) of precipitation. Even though this was 38.64 in. (981 mm) below average, it was only 1.54 standard deviations below average. During that same time period, Juneau was 11.27 in. (286 mm) below average for precipitation (−1.39 standard deviations). While the deficit in Juneau was much lower in magnitude than Little Port Walter, the deficits as measured in standard deviations were very similar (−1.54 vs −1.39). Figure 4 shows the precipitation surplus or deficit in standard deviations from the average for the monthly and annual precipitation totals for each of the five stations. Monthly correlations between all observation stations, except Angoon, range between strong and very strong.
All of the stations near Angoon reported annual precipitation totals below the long-term average. This indicates an areawide incidence of below-average precipitation for the time period in question. The National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis data (Kalnay et al. 1996) also indicated an annual precipitation deficit for all of southeast Alaska. While the reanalysis data are coarse for this level of analysis, the regional trend confirms the below-average precipitation regime for the region.
The most noteworthy discrepancy between the recorded observations during this period is on 12 October 1982, when the state record of 15.20 in. (386 mm) of rain was recorded in Angoon. Table 3 indicates that each station received considerable rainfall on 12 October 1982. All stations recorded precipitation values greater than would normally be expected for the date. However, the 12 October 1982 rainfall does not represent a daily record for any of the comparison sites except for Little Port Walter—and this is substantially lower than the monthly record value for October.
A review of the Juneau Empire and Sitka Sentinel daily newspapers for October and November 1982 reveal no news stories about heavy rainfall or flooding in those cities or the region. If a historic precipitation event occurred, then property damage, road damage, and river flooding would have been observed. In addition, no reference was made to Angoon or the rainfall values recorded there in either newspaper. However, since the Angoon data form was submitted on a monthly basis to the National Weather Service, the likelihood of the measured precipitation amount appearing in a news story is difficult to establish.
THE OCTOBER 1982 COOPERATIVE DATA FORM AND THE OBSERVER.
To further evaluate the data integrity issues with the official observations, we begin by looking at the original cooperative data form. Figure 5 shows a portion of the October 1982 scanned, handwritten WS Form E-15 for the Angoon Power Station site. October is one of the wettest months of the year in southeast Alaska. However, the monthly table for Angoon shows that 16 days achieved a precipitation total over 1 in. (25 mm), 7 days had over 2 in. (51 mm), and 12 October had 15.20 in. (386 mm).
If we stipulate that the observations are erroneous, then the question becomes, what procedure was used to measure the precipitation? Was the water level in the rain gauge inaccurate? Was the gauge malfunctioning, or was the method of measuring the precipitation inaccurate? The question of how these measurements were recorded is difficult to ascertain without interviewing the person who made the observations. Fortunately, the Angoon volunteer observer is still alive and generously spoke with the lead author of this paper.
The following section provides a summary of the correspondence with the observer (anonymous observer 2012, personal communication).
The observer who recorded measurements before 1981 was a Pastor at the Assembly of God church in Angoon. After he passed away, another individual at the church took over the recording of observations. The observer stated that the measuring stick looked similar to the one shown in Fig. 6.
The observer did remember the heavy rainfall that is indicated on the monthly form shown in Fig. 5. He described the rainfall as “very heavy” and indicated that 1) several gravel roads around town sustained damage and 2) some cisterns for collecting rain water from house roofs overflowed.
The observer openly wondered if the observations were correct. The measurement instrument was a 20-in. (508 mm) metal stick that was inserted into the central cylinder of the rain gauge. The same stick was used to measure both rainfall and snowfall. He mentioned that there were different numbers on the backside of the stick and his wife indicated they were unsure which numbers to use.
Based upon a review of the monthly weather forms during the time period in question and the correspondence with the cooperative observer, an explanation has emerged that accounts for the anomalous precipitation observations attributed to the observer. The measuring stick is 20 in. (508 mm) long and at each 1-in. (25 mm) increment, the marking on the stick increases by 0.10 in. (2.5 mm). This produces a disconnect in logic. If a standard ruler is held next to the measuring stick, the major increments appear 1.0 in. (25 mm) apart—even though the written markings indicate that increments are 0.10 in. (2.5 mm) different. When recording snow using the measuring stick, the observer treats the major markings as if they are 1.0 in. (25 mm) apart. The liquid precipitation was measured using the same logic; that is, the markings on the measuring stick were treated as 1 in. (25 mm) apart.
The measurement technique described above creates an immediate exaggeration of the precipitation. The rain gauge has a large funneled opening to focus the precipitation into a narrower inner cylinder. This amplifies the precipitation by a factor of 10 for ease of measurement; therefore, a 20-in.-tall (508 mm) cylinder full of water corresponds to 2.0 in. (51 mm) of precipitation. However, the observer assumed that a 20-in.-tall (508 mm) cylinder full of water was indicative of 20 in. (508 mm) of precipitation. Therefore, precipitation totals were generally exaggerated by a factor of 10. Also, all of the observations greater than 1.0 in. (25 mm) end with the digit 0 (2.80, 1.30, 15.20, etc.). This lends credence to the misreading hypothesis.
Interestingly, the observations on the monthly forms under 1.0 in. (25 mm) are almost always less than 0.1 in. (2.5 mm). For example, in October 1982, 10 different days are shown as having received between 0.01 and 0.09 in. (0.25–2.3 mm) of precipitation; yet, no days are shown as receiving between 0.10 and 0.99 in. (2.5–25 mm). We feel that the measurements less than 0.1 in. (2.5 mm) were correctly recorded and that the other measurements are erroneous.
If the observer overstated all precipitation values greater than 0.1 in. (2.5 mm) by a factor of 10 as described above, then we can adjust those raw values and provide a new estimate for precipitation during that period. When we presented the observer with our hypothesis about the measurements and the markings on the measuring stick, he acknowledged that it was certainly possible that he treated the 0.1-in. (25 mm) markings as 1.0 in. (25 mm), and he also did not realize that the funnel located at the top of the gauge amplified the precipitation.
If all the precipitation observations over 1.0 in. (25 mm) are divided by 10, over the course of a year, then this yields an adjusted precipitation total of 25.89 in. (658 mm), which represents a deficit of 13.85 in. (352 mm). This deficit value corresponds closely with the NCEP–NCAR estimated deficit and is similar in magnitude with other stations in the region (see Fig. 4).
DATA MANAGEMENT AND QUALITY CONTROL.
The handwritten E-15 observation forms from Angoon were sent to either the National Weather Service office in Juneau or the Alaska region headquarters before being sent to the NCDC for digital transcription. While a specific history for the forms from Angoon does not exist, the protocol was for manual entry of the information into a TD-3200 computer form. At a later time, automated quality control measures were performed by NCDC on all temperature and precipitation values (Reek et al. 1992). Approximately 0.07% of precipitation values were identified as suspect in a detailed test of 1,300 stations. Many of those values were corrected and others were flagged as problematic.
In the early 1990s, climate data for all stations with a minimum of 10 years of observations were incorporated into the GHCN, including Angoon. Periodic quality control checks (see Peterson and Vose 1997; Menne et al. 2012) are performed on temperature and precipitation data within the GHCN and climatologically questionable values are flagged as “failed climatological outlier check.” The climatological outlier check evaluates a variable at a station by comparing the observations to an expected distribution and identifying instances that are beyond a preset maximum allowable distance from the expected range of values (Durre et al. 2010). Four values for Angoon Power Station between February 1982 and February 1983 were flagged based on this test.
This paper identified problems with the Angoon Power Station cooperative weather observations for precipitation measured between August 1981 and June 1983, and it questions the validity of the state-record 1-day precipitation amount of 15.20 in. (386 mm) recorded on 12 October 1982. The evidence indicates that the measurement technique overestimated most precipitation values. We presented three independent rationales for this assessment. First, the precipitation observations are statistically unlikely. Second, the precipitation observations do not match observations at nearby stations. Third, the measurement techniques described by the observer indicate a consistent overestimate of the actual precipitation.
For the reasons stated earlier, the 12 October 1982 rainfall of 15.20 in. (386 mm) is not considered accurate. This precipitation measurement is also the state of Alaska's single-day record event. The National Climatic Data Center has established a mechanism for reviewing statewide temperature and precipitation records through the creation of a State Climate Extremes Committee (Shein et al. 2013). The authors of this paper recommend an NCDC State Climate Extremes Committee (www.ncdc.noaa.gov/extremes/scec/records) review of this record. If the committee concurs with the findings within this study, then the second highest precipitation event on record, 15.05 in. (382 mm), of rain recorded on 10 October 1986 in Seward, Alaska, should be investigated as the largest calendar-day event in the Alaska climate record.
This case study in observation verification has applications for other datasets that contribute to the climatological record. Extreme events establish the limits by which society makes many planning decisions. However, the purported records are not infallible (see El Fadli et al. 2013). Evaluating the reliability of these events is crucial for the public to have confidence in the climate record.