Wednesday, December 31, 2014

Not as Warm as 2002

The unusual and persistent warmth in Fairbanks grinds on, with temperatures rising above freezing this early morning at Fairbanks airport.  Some very light freezing rain has also been reported.  Climatologically speaking, above-freezing conditions occur at Fairbanks in December about once every other year.

It's interesting to compare the winter-to-date mean temperatures this year with those in 2002, which saw by far the warmest early winter on record.  The two charts below show the comparison; 2002 was considerably warmer even than this year, with many days seeing temperatures more than 1 standard deviation above normal.  The common factor between the two years?  An extremely positive phase of the PDO.

How did the rest of the winter turn out in 2002-2003?  Except for some brief cold weather around the turn of the year, it continued warmer than normal for most of January and February; early February was particularly warm.

Regular readers should note that I'm on the road and posts will be brief and limited until next week.

Sunday, December 28, 2014

Cold After Snow

Snowfall on Thursday night and Friday finally brought the snowpack in Fairbanks to above 12 inches in depth, with 5.2" of accumulation measured at the airport.  In the wake of the snowfall, the temperature dropped sharply and reached -20 °F at the airport, the coldest observed so far this mild winter.

It's a common occurrence to see temperatures drop rapidly when clouds dissipate above fresh snow cover, and I got to wondering if this "post-snow" temperature drop could be quantified.  For example, in the latest event the temperature dropped from 22° to -20°, a drop of 42 °F, and after the early December snowfall the temperature dropped 38 °F from 25° to -13 °F.  Are these numbers typical?  To look at this, I searched the winter months (November-March) since 1930 for snowfall events of 4 inches or greater, with two-day totals included.  I then removed overlapping events and obtained the largest temperature drop within two days after snow ended in each case (243 cases in 85 years).  A histogram of the results is shown below.

The historical data reveal that a drop of 40 °F in one or two days is very typical, with the long-term median being 37 °F.  It's not at all surprising to see that large temperature drops are quite common, because the thick cloud cover and warm advection associated with these relatively "heavy" snow events initially brings elevated temperatures, but then clearing skies and cold advection behind the system allow for rapid cooling.  Fresh snow cover is also very favorable for radiational cooling.

Out of curiosity I looked quickly at the long-term history of this phenomenon by taking a trailing median of the temperature drop associated with the prior 20 snow events, and then applying an extra 7-event smoother, see below:

There have been some interesting variations in the past two decades, with snow events generally being followed by less cooling in the 1990s but considerably more cooling since 2000.  I have no useful suggestions on how to interpret this change, so I mention it merely as a point of curiosity.  But it's clear that this weekend's chill was exactly in line with the Fairbanks climate of the past decade.

Friday, December 26, 2014

Forecast Bias Revisited

It was a week ago now that Fairbanks experienced its first day this winter with an official (midnight-to-midnight) high temperature below zero.  This was 32 days later than the long-term median of November 17 and was also the second latest on record after 2002 (December 21).  The chilly conditions also ended a 31-day sequence of days that saw a higher maximum temperature than indicated by the NWS forecast from the previous morning.  So far in December, the NWS day-1 high temperature forecasts are averaging 7.7 °F too cold at Fairbanks airport.  We previously discussed this puzzling cold bias in the NWS temperature forecasts here and here.

To look at this issue a little more closely, I obtained the archive of MOS (Model Output Statistics) computer forecasts since 2000; these are purely statistical forecasts with no subjective modification by a human forecaster.  The MOS forecasts often serve as a first guess for a human forecast, and are undoubtedly a key component of the NWS forecast process.  As the chart below shows, the NWS day-1 high temperature forecasts follow the MOS numbers quite closely, although the NWS temperatures average 2.7 °F higher than MOS.  Note that the diagonal line shows the 1:1 (perfect match) line, not the best-fit line.

The systematic difference between the two sets of forecasts reveals that the day-1 MOS bias is even greater than the NWS bias, i.e. the NWS forecasters are removing part of the cold bias from the MOS values.  However, there is a nuance here that I realized when reading the MOS documentation.  The MOS high temperature forecasts are valid for the "daytime" period 7am - 7pm local standard time, and this means that it is not appropriate to use the midnight-to-midnight maximum temperatures for verification.  On average (2000-2014), the Fairbanks midnight-to-midnight maximum is 2.6 °F higher than the 7am-7pm maximum in December through February, because the 24-hour period has several hours either side of the "daytime" period in which the temperature can be higher (and this often happens with little solar heating in deep winter).

The bias of the MOS forecasts each winter since 2000-2001 is shown in the chart below, both for the midnight-to-midnight verification and the (correct) 7am-7pm verification.  The red diamonds indicate the mean winter temperature anomaly, which we expect to show some inverse relation to the bias.  We see that the MOS forecasts were actually too warm back in the winters of 2003-4 through 2005-6, but since then the cold bias has been persistent even when using the correct "daytime" verification.

An interesting conclusion we can draw from the recent MOS cold bias is that recent winters have seen a different statistical relationship between the airport temperatures and the GFS model predictors than during the MOS development (fitting) period.  The MOS regression equations for temperature were last updated in March 2010, but this seems to have made things worse as the cold bias was quite notable in the subsequent winter despite colder than normal temperatures.  As it happens, the next MOS update is scheduled to occur on January 14, 2015, in tandem with a major upgrade to the GFS model itself; so it will be very interesting to see if the MOS forecast bias is reduced at that time.  If it is, then it's likely that the NWS forecasts will also see sudden improvement.

Tuesday, December 23, 2014

Negative PDO Seasonality

My recent post on the seasonality of the PDO's influence on Fairbanks temperatures revealed an interesting feature of the Fairbanks climate: according to historical data since 1930, late December often brings relief from the colder than normal conditions that are typically associated with a negative PDO phase.  In other words, when we look at temperatures observed in Fairbanks concurrently with a negative PDO phase, we find a notable absence of cold on or around December 21st.  This was illustrated in the following figure from the earlier post (note that I discovered a minor error in the calculation, and the new results show a slightly less pronounced peak in late December):

To dig into this a little deeper, I examined the evolution of the 500 mb height pattern in the years that saw a significantly negative (bottom quintile) PDO index on December 21, which is the peak of the anomalous "warmth" in the figure above.  Fortuitously, all 17 of the years occurred between 1948 and the present, which is the era of the NCEP/NCAR reanalysis data; none were found in the period 1930-1947.

The sequence of maps below shows the 500 mb height pattern at 10-day intervals for the 17 negative PDO years, from November 1 to January 30 (left to right, then top to bottom).  Blue shading indicates where the height was below normal in the majority of years, and red shading corresponds to above-normal heights.  The upper-level circulation appears to have followed a somewhat consistent sequence in these years; for example, an upper-level trough was frequently in place over Alaska on November 11.  The interesting feature in connection with the late December warmth is the ridge that develops south of Alaska in December in the majority of years; the December 21 map shows that above-normal heights tend to encompass the southwestern half of Alaska on that date.  However, the anomaly is quickly erased by the end of the year.

The temperature anomalies in Fairbanks in the 17 negative PDO years are shown in the chart below.  Note that the 17 years are the same throughout, unlike in the top figure above, where the 17 years were determined separately for each date based on the lowest PDO values for that date.  The chart below shows a much reduced cold signal in November, which I take to mean that the PDO values often change quite significantly over the course of just a month or two.  Of course, the PDO index partly reflects a response of the sea surface temperatures to recent weather conditions, so we can't think of the PDO as simply an independent forcing.  It may be, for example, that a sequence of upper-air anomalies similar to that shown above is particularly favorable for a strongly negative PDO on December 21; it's hard or impossible to extricate cause and effect, so my aim here is just to show the historical connection.

I also created a parallel set of maps showing the reanalysis temperature anomaly patterns in the (static) negative PDO years, see below.  The cold anomaly over the interior on December 1 and January 10 agrees with the Fairbanks chart above, and we see that on December 21 warmth is common over the Bering Sea.  The overall impression of the weather pattern is of volatility, with cold episodes quickly giving way to normal or even warmer than normal conditions over Alaska.  This is consistent with what we saw in the previous post, that winter temperatures are more variable during the negative PDO phase than during the positive phase.

Saturday, December 20, 2014

Howard Pass National Wind Chill Record

** This is a working draft as the National Park Service may have suggested edits or additions **

On February 14, 2014, a remarkable event occurred along the northern slopes of the Brooks Range above the Arctic Circle. Namely, a purported wind chill of -97°F was observed by an automated wind instrument located at Howard Pass operated by the National Park Service. The 2014 event was described and modeled in great detail by Richard here and here. I encourage you to read those posts.

The station is part of the Remote Automatic Weather Stations (RAWS) network and is referred to as Howard Pass RAWS. This event was well publicized by the National Weather Service (NWS) as a possible state and national record. Both Alaska and national newspapers carried the story and for a few days it generated a fair amount of discussion nationally. Unbeknownst to all but a handful of people at the time was an even lower wind chill recorded at the same station one year earlier. This earlier event was unknown at the time of occurrence due to the malfunctioning of transmission equipment. The onsite data logger continued to function during the event and when the data were retrieved several months later, the dramatic observations were revealed.

At 10 p.m. on February 21, 2013, a 2-meter air temperature of -47.5°F was observed along with a sustained wind of 53.7 miles per hour at the Howard Pass RAWS weather station. The combination of these two meteorological conditions produced a staggering wind chill of -99.8°F! The following sections of this report describe the setting where the station is located and the meteorological conditions present at the time.

Howard Pass

The Howard Pass area is currently uninhabited but indigenous people have occupied the region for thousands of years. The pass is an important caribou migration corridor and hence historic and prehistoric hunters utilized this location for subsistence and other traditional activities. The National Park Service participates in a program to establish and maintain weather stations in remote sections of Alaska on their land holdings. Howard Pass is within the six million acre Noatak National Preserve, created in 1980 by the Alaska National Interest Lands Conservation Act (ANILCA). The Noatak National Preserve is managed by the National Park Service and part of the justification of the Preserve's establishment was to promote scientific research in the area (ANILCA Section 102 (8)(a)). The National Park Service decided in 2007 to place an automated station at Howard Pass for the following reasons: 1) there are no nearby stations, 2) it is a major caribou migratory corridor, and 3) there are significant archaeological resources in the pass. This site location is also more accessible by helicopter than the pass itself due to the boggy nature of the pass.

The Howard Pass RAWS station is located at latitude 68.156°N, longitude 156.895°W, and at an elevation of 2,062 feet in the northwestern Brooks Range of Alaska. The station sits near the top of a hill on the eastern flank of the pass approximately 300-400 feet above the pass level and is approximately 12 miles south of the northern limit of the Brooks Range and 85 miles north of the southern limit. The station was placed above the pass level to prevent migrating caribou from damaging the station. Figure 1 shows a map of the station’s location and Figure 2 shows an exaggerated relief map of the area. Figure 3 shows a picture of the station.

Figure 1. Howard Pass RAWS station location in northern Alaska. Image is a screenshot from Google Earth.

Figure 2. Howard Pass RAWS station vertical relief. The viewpoint of the image is facing north-northeast. Image is a screenshot from Google Earth.

Figure 3. Photograph of Howard Pass automated weather station looking toward the south. Image courtesy Ken Hill.

The station’s location makes it extremely susceptible to both low temperatures and strong winds. Each of the winter months has an average temperature between +5°F and -10°F and an average wind speed between 17 and 25 miles per hour. Figure 4 shows the monthly averages between 2011 and 2014. Just looking at the month of February, the combination of temperature and wind produces a month-long average wind chill of -36°F!

Figure 4. Monthly average temperature and wind at the Howard Pass RAWS station (2011-2014).

All of northern Alaska is characterized by long, severe winter conditions. Snow lies on the ground from October through May and temperatures are below zero for nearly one third of the year. The passes of the Brooks Range are also subjected to intense winds that are funneled through topographically constrained areas. The cold, dense nature of the airmass facilitates a semi-permanent inversion layer that provides a vertical constraint on the wind movement. The combination of the horizontal constriction due to topography and vertical restrictions due to an inversion layer causes tremendous winds to occur when very low temperatures are present and a north-to-south pressure gradient exists. This setup frequently produces exceptionally low wind chill values.

February 21, 2013

If you only had a satellite image to look at, nothing about February 21, 2013, would stand out. The MODIS satellite image from this date (see Figure 5) reveals nothing out of the ordinary. The entire region is snow covered and few, if any, clouds are present.

Figure 5. MOSIS Natural Color (12:43 p.m. Alaska Standard Time on February 21, 2013). Image area is approximately 105 x 70 miles.

What does not show up in the MODIS image is the dynamic nature of the atmosphere. The map shown in Figure 6 is a plot of surface weather conditions at 9 p.m. on February 21, 2013, from the Weather Prediction Center (WPC). A strong area of high pressure is centered far to the north of Alaska while a strong low pressure system was present in the Gulf of Alaska. These two features generated a tight pressure gradient resulting in strong winds across the northern one third of the state. Also evident are low temperatures and strong northeast winds.

Figure 7 is a map of temperature and wind chill values at 10 p.m. on February 21, 2013. The map illustrates the intensity of the cold air and the severity of the wind chill at the peak of the event. Many stations reported wind chills in the -40°s, -50°s, and -60°s. Anaktuvuk Pass, in the central Brooks Range, reported a wind chill of nearly -70°F. At this same time, the Howard Pass RAWS station reported a wind chill of -99.8°F. 

Figure 6. Surface map of Alaska on February 21, 2013 at 9 p.m. Alaska Standard Time (AST). Map courtesy of the Weather Prediction Center.

Figure 7. Temperature and wind chill values reported on February 21, 2013, at 10 p.m. Alaska Standard Time (AST).

We should note that this was not a short-lived event. For the 9-day period between February 15, 2013, and February 24, 2013, the 2-meter temperature at Howard pass averaged -32.6°F and the 3-meter wind speed averaged 33.1 miles per hour (mph). The average wind chill during those 9 days was -71°F. Figure 8 shows the temperature and wind chill for Howard Pass RAWS during the time period described above.

Figure 8. Hourly temperature and wind chill values observed at the Howard Pass RAWS station between February 16, 2013, and February 24, 2013.

For nearly 48 hours, the wind chill at Howard Pass RAWS oscillated between -90°F and -100°F – with the peak value of -99.8°F reported at 10 p.m. on February 21st. During those 48 hours, the average temperature was -45.1°F and the average wind speed was 49.1 mph. We should note that Howard Pass has reported winds in excess of 80 mph on numerous occasions. The event of February 2014 that produced a wind chill of -97°F included a stretch of time with temperatures in the -30°s and sustained winds up to 90 mph. Interestingly, once wind speeds exceed 50 mph, wind chill values do not change very much. Figure 9 shows the conditions observed at Howard Pass during the low wind chill event of February 2014 that produced a minimum reading of -97°F.

Figure 9. Hourly temperature and wind chill values observed at the Howard Pass RAWS station between February 11, 2014, and February 15, 2014.

Station Equipment

Is a -99.8°F wind chill even possible or should we immediately treat the observations as suspicious and figure out what went wrong? Let us first look at the equipment present at the site (also see photograph in Figure 3).The following bulleted points were relayed from National Park Service staff:

  • Station installation date: 7/13/2011. Station blew over on 8/16/2011 and was not repaired until 7/16/2012. Most sensors were not operating correctly August 2011 - July 2012.
  • There are two air temperature sensors at the station. The primary sensor is a YSI ThermX and the backup sensor is a Vaisala HMP155. For the period 2011 through July 2013 the backup sensor (Vaisala) was transmitting instead of the primary sensor (YSI). 
  • The transmitting air temperature sensor (Vaisala) failed on August 3, 2012 and was not repaired until summer 2013. During the summer 2013 field visit, the primary sensor data were recovered from the data logger.
  • Additional air temperature sensor information:
    • Model YSI 44211
    • Height: 2.0 meters
    • Linear Range: -55°C to +85°C
    • +/- 0.18° at -55°C
    • +/- 0.02°C at +85°C
  • Additional wind sensor information:
    • Model RM Young 05103
    • Height 3.0 meters
The station is clearly located in an exposed area. Strong winds have disabled the station several times in the past and have sand-blasted the equipment. This alone lends credence to extreme wind observations. Nothing in the list of station equipment though stands out as a disqualifier for any observations.

Event Reconstruction

To look at the plausibility of this event, we conducted a model simulation using the Weather Research and Forecasting (WRF) model for the time period in question. The model was initialized using February 21, 2013, 0.5°GFS data from 00:00 UTC. A recursive, nested grid structure of 27-9-3-1 kilometers were modeled over a 96-hour period. The model uses 30 arc-second topography – equivalent to 1 square kilometer grid cells – to represent the terrain.

The model showed very cold temperatures, strong winds, and very low wind chills during the time period in question. Figure 10 shows the estimated temperatures for the portion of the model area centered on Howard Pass. Cold air is clearly advected part of the way through the pass. The temperature only drops about 5°F in the 12 miles from the northern entrance of the pass until the weather station. 

Figure 10. WRF model estimated temperatures for February 21, 2013, at 10 p.m. Alaska Standard Time (AST). Model initialized at 2/21/2013 at 0 UTC. Scene is 156 kilometers wide. Inner box contains 1.3 km grid cells. Areas outside of box contains 4 km grid cells. Units are degrees Fahrenheit.

Figure 11 shows the modeled sustained wind speed. The wind was only estimated to be around 35 miles per hour through the pass. Note how tightly packed together the wind streamlines are. There is a rather large discrepancy between the observed wind speeds and the modeled wind speeds. The hill where the station resides is not accounted for in the rather coarse (1 km) topographic dataset that the model uses. Interestingly, the model predicts an increase in winds of 5-10 miles per hour 100-200 meters above the pass level. Since the hill where the station is located is about 100-200 meters above the pass level, this is a reasonable, but still too low, proxy estimate for the wind speed at the station location. Figure 12 shows the modeled vertical wind profile for the station’s location.

Figure 11. WRF model estimated sustained winds for February 21, 2013, at 10 p.m. Alaska Standard Time (AST). Model initialized at 2/21/2013 at 0 UTC. Scene is 156 kilometers wide. Inner box contains 1.3 km grid cells. Areas outside of box contains 4 km grid cells. Units are miles per hour (mph).

Figure 12. WRF model estimated vertical wind profile for February 21, 2013, at 10 p.m. Alaska Standard Time (AST). Model initialized at 2/21/2013 at 0 UTC. 

Finally, Figure 13 shows the wind chills for the portion of the model area centered on Howard Pass. There are several black regions at the northern entrance of Howard Pass. Those grid cells represent wind chill values between -90°F and -95°F. These are the lowest values for any portion of the model domains. At the location of Howard Pass RAWS, the estimated wind chill was -85.4°F. As noted in the earlier paragraph, the model’s elevation dataset is not detailed enough to represent this area of complex topography.

Figure 13. WRF model estimated wind chills for February 21, 2013, at 10 p.m. Alaska Standard Time (AST). Model initialized at 2/21/2013 at 0 UTC. Scene is 156 kilometers wide. Inner box contains 1.3 km grid cells. Areas outside of box contains 4 km grid cells. Units are degrees Fahrenheit.

Overall the model did a reasonable job of representing this event with the exception of under forecasting the winds. Table 1 summarizes the difference between the actual observations and the model estimates. We suspect that with a higher resolution topographic dataset, the modeled wind speeds would approach the observed wind measurements.

Table 1. Comparison of modeled and observed conditions on February 21, 2013, at 10 p.m. Alaska Standard Time (AST).

Wind Chill Record

Is the -99.8°F wind chill a U.S. record? The -97°F wind chill from February 2014 was informally determined to be a statewide and a national record. Unfortunately there is no database of lowest wind chill readings. The National Climate Data Center (NCDC) does not track low wind chills as part of their state or national records database.

Back in 2001, the National Weather Service (NWS) adopted a new formula for computing wind chill values. Prior to 2001, wind chill values as low as -120°F were observed at several location. However, the “new” formula generally has higher (warmer) wind chill readings that the “old” formula given the same temperature and wind speeds. When those older observations were evaluated with the new formula, they all came back in the -90°s range.

So what is the lowest reading and what criteria are used? Would you believe that the previous low wind chill reading in Alaska was also -99.8°F? The city of McGrath, Alaska, reported an air temperature of -72°F and a wind speed of 7 mph on January 27, 1989. Plugging these numbers into the wind chill formula gives us the -99.8°F value. Figure 14 shows the minimum wind chills in Alaska during the great January 1989 cold snap. All values were rounded to whole numbers – hence the -100°F value shown for McGrath. Figure 15 shows the hourly observations at McGrath on January 27 and January 28, 1989.

Figure 14. Minimum wind chills in January 1989 for all stations in Alaska that reported hourly observations using. For some stations, many observations are missing so a few data point should be considered suspect.

Figure 15. Hourly temperature and wind speed for McGrath, Alaska, on January 27 and January 28, 1989. Both the new and old wind chill values are displayed. The -99.8°F (-100°F) value using the new formula is shown.

The McGrath number comes with a qualifier though. With a wind of only 7 mph, should it even count? The Alaska NWS offices do not even issue Wind Chill Advisories or Wind Chill Warnings unless the sustained winds are, or are forecasted to be, 15 mph or greater for at least three hours. Maps of all Alaska advisory criteria can be found here. So, can a wind chill be a record if it wouldn't even qualify for a Wind Chill Advisory? 

Several NWS offices have lists of statewide wind chill records. When the Minnesota Climatological Working Group discussed extreme wind chills, several examples they gave used wind speeds of 6 or 7 mph. The Montana Climate Atlas uses 10 mph as a threshold for their monthly probability maps. The lowest statewide wind chill value in their atlas was -80.9°F.

The NWS office in Lacross, Wisconsin, has a climatology of wind chills in the Northern U.S. (Lower 48). They use a 10 mph filter to develop their probability maps. In their report, they state, "[t]his speed was chosen as it is the minimum threshold currently used throughout most of the NWS for the issuance of Wind Chill Advisories or Warnings." Since their report is looking at climatological probabilities, it makes sense that they have a wind speed cutoff. Their report does not contain a list of extreme wind chills.

Given that no wind chill values in the Northern states claim to even approach -100°F, it seems unlikely that any place in the Lower48 has approached this value using the 2001 formula.

Another consideration is the nature of the formula itself and whether it is even applicable at such extreme values. It is worth noting that the original research that went into developing the new wind chill formula did not use air temperatures lower than -40°F and that the fitting of a formula to the observational data is the only reason it can be extended backward. Nevertheless, one of the pioneers of the new wind chill research, Randall Osczevski, discussed the New formula (which was actually developed in the 1990s) in this paper and uses values as low as -100°F with air temperatures as low as -60°F. He also discusses how the new formula is more realistic at extreme low temperatures. Also, since NOAA calculates wind chill values for any temperature and any wind speed (over 3 mph), we can assume an implicit endorsement of the formula at extreme low temperatures. Therefore, we consider the McGrath wind chill value of -99.8°F a valid wind chill record. 


Circumstantial evidence supports the observations from the Howard Pass RAWS station during the February 2013 low wind chill event. We therefore consider the wind chill observation to be valid and consider the reading a statewide and national record – along with the aforementioned reading from McGrath in 1989. If the National Park Service is able to keep the station up and running for a number of years, we expect this record to be broken several times in the future.


The National Park Service generously provided the raw station data, site photographs, and instrument specifications for this analysis.

Thursday, December 18, 2014

Winter Analogs

Regular reader Eric suggested that it would be useful to look at some analogs to see if we can find any indication on how long the unusual warmth may or may not persist this winter in Fairbanks.  Under certain circumstances, such as when major climate forcing mechanisms can be identified, analog forecasts can be at least as useful as very-long-range computer model forecasts.

The first analog I looked at was based on past years in which the PDO was significantly positive (at least +0.5 PDO index value) in both November and December.  I also required the November 1 - December 15 mean temperature in Fairbanks to be at least 3 °F above the preceding 30-year normal.  The chart below shows the daily temperature anomalies in thin gray lines and the median of the 10 years in black.  There is a wide range of temperature outcomes for January, but the median is generally on the warm side as we would expect.  I wouldn't regard any of the January peaks and troughs as being statistically significant, but the peak in early February is larger and it certainly looks like warmth is favored in the first third of February.

If we consider the fact that El NiƱo is under way at present, and combine that with the positive PDO in December, we find a slightly more robust indication of warmth through most of December and the first part of January, and again in early February.  It is interesting to note that large cold anomalies are notably infrequent among the analog years for most of December and February, but cold shows up for some of the years in January.

Another aspect of the current climate phase space is that the Quasi-Biennial Oscillation (QBO) is strongly negative.  Combining this criterion with the positive PDO in December produces a forecast that is generally similar, although the warm signal in January looks slightly more significant (see below).  The median temperature is considerably above normal for much of January, but cold extremes also show up in some of the years.

The similarity of the PDO-QBO forecast to the others suggests that perhaps the QBO has only a marginal influence.  I tested this by also finding analogs with positive PDO and positive QBO, i.e. the opposite QBO phase.  Under these conditions the warm signal seems notably less pronounced for January (see below), so perhaps we shouldn't discard the QBO as a useful factor in the forecast.

Finally, there has been much discussion lately in forecasting circles about the very large anomaly in Siberian snow cover this autumn.  In the past decade or so, it has been discovered that snow cover over Eurasia in October is a skillful predictor of the subsequent winter's Arctic Oscillation (see e.g. here), and October 2014 produced a near-record snow cover extent in Eurasia.  The implication is that the AO phase is very likely to be negative this winter (especially after January 1st), which would imply higher than normal pressure over the Arctic and lower than normal pressure in parts of the mid-latitudes, including the North Pacific; this in turn would suggest warm conditions in southern Alaska.  I looked at 7 analog years and found only small signals for Fairbanks (see below), although warmth in early February again appears more likely than cold.

Putting it all together, the grand ensemble of all the analog years produces the result shown below.  Warm conditions are the most likely outcome for most of January and the first half of February, and the probability of warmth appears to be highest for early February.  The range of uncertainty is large for January, so it shouldn't be a surprise if unusual cold shows up for at least a time.

In terms of monthly mean temperatures, the overall ensemble of 23 distinct analog years produces above-normal temperatures 65% of the time in January, 74% of the time in February, and 57% of the time in March.