Frosty Morning

It was a chilly morning across interior and northern Alaska this morning, signaling that autumn is now under way.  Some temperatures I noted were:

19F  Atigun Pass SNOTEL
20F  Chandalar Shelf DOT
22F  Bettles Field SNOTEL
22F  Dahl Creek near Kobuk
22F  Goldstream Creek CWOP
23F  Goldstream Creek COOP
24F  Coldfoot SNOTEL
24F  Upper Nome Creek SNOTEL
25F  Tanana ASOS
29F  Keystone Ridge - lowest August temperature in the 19-year history
33F  Fairbanks Intl AP

Many of the higher elevation locations in the hills north of Fairbanks dipped below freezing, so it wasn't just a valley-bottom frost.

Autumn Warming - Part 2

Last week I examined the timing of the remarkable autumn warming that has affected Barrow in the past 12 years, and I showed that an October peak in unusual warmth has been observed not only at Barrow but throughout the northern part of Alaska.  Brian showed that the warming also extended down to the Bristol Bay region.  My original hypothesis that this is mainly an effect of the reduced autumn sea ice coverage in the Arctic proved a bit controversial, as readers suggested that changing wind patterns could also be responsible; so I've been thinking about further analysis that could shed light on the subject.

To highlight the issue again, below are charts showing the October mean temperature anomaly at Barrow and Fairbanks, relative to the 1981-2010 normal.  Barrow has seen a most extraordinary change since 2002, with October temperatures well above normal every year and with very little interannual variability compared to earlier years.  The warming has been less pronounced at Fairbanks, but is still quite noticeable, with 6 of the 12 most recent years being more than 5 °F above normal.

To investigate the cause of the warming at both locations, I first looked at changes in the distribution of 850 mb wind direction in October (see below).  At Barrow the changes between 1981-2001 and 2002-2013 were fairly small, with the more recent period seeing winds from the east more often.  There was also a small increase in the frequency of southwesterly and westerly winds, and a small decrease in the frequency of northwesterly flow.  Note that the area under the curves is not the same, because winds of 10 knots or stronger increased in frequency from 68% to 78% of the time.

The October changes at Fairbanks are more noticeable, as winds from compass headings between 120° and 240° have increased at the expense of winds from the west and from the east.  In and of itself this would be consistent with overall warming, as warmer air from the south is transported to Fairbanks a greater percentage of the time.

However, this is not the end of the story.  The scatter plots below show the relationship between wind direction and temperature anomaly at Fairbanks, for the 1981-2001 period (blue dots) and the 2002-2013 period (red dots).  The blue and red lines show the median temperature anomaly for each period, based on wind direction windows of +/- 10 degrees (and with a minimum count of 10 events to calculate a median).  The top chart shows the relationship for surface temperature, and the bottom chart shows the relationship for 850 mb temperature.  Remarkably, the results indicate that surface and 850 mb temperatures have warmed similarly regardless of the direction of the 850 mb wind; there are some subtle differences in the amount of warming depending on wind direction, but overall the warming is quite uniform.

These results are NOT what I would expect to see if the October warming in Fairbanks was directly caused by the reduction in sea ice.  If the sea ice reduction was causing the warming, then there should be relatively more warming in more northerly flow, and relatively less warming in more southerly flow, but if anything the opposite has occurred: air coming from the south has become considerably warmer in Fairbanks!  I interpret this to mean that the recent October warmth in Fairbanks does NOT have its origin in warmer Arctic air masses; instead, there has been general warming regardless of wind direction, AND southerly winds have become more frequent.

Looking now at Barrow (see below), we see a result that is more consistent with October warming caused at least partly by an open ocean.  The surface temperatures have warmed dramatically for wind directions between 300° and 60°; for example, the median temperature for winds from 0-20° or 320-340° has increased by more than 15 °F between 1981-2001 and 2002-2013.  Northerly winds in October just aren't very cold in Barrow any more.  Interestingly, however, the 850 mb temperatures show by far the most warming from the southwesterly direction (210-250°), which again suggests a much larger-scale warming effect rather than local heating from the open ocean.

One more point of interest in the top Barrow scatter plot (surface temperatures) is that in the past 12 years, the median temperature anomaly is remarkably independent of wind direction in October.  This fits perfectly with the collapse of interannual variance in October temperatures: no matter which direction the wind blows from, the low-level temperature outcome is about the same.  Remarkably, it appears there is no way for the October temperature anomaly to end up significantly different from where it has been in the past 12 years.

Finally - I promise - here is the change in the temperature difference between the surface and 850 mb at Fairbanks and Barrow, for each day of the year (but heavily smoothed).  Values above zero indicate that the surface has warmed relative to the 850 mb level in the past 12 years.  The surface warming has been very dramatic at Barrow, but quite insignificant at Fairbanks; and I would suggest that this represents a better indication of the true thermal effect of the reduced sea ice.  Also, we see that the surface warming peaks in early November, which distinguishes it from the mid-late October peak that we saw in the simple temperature anomalies.

First Frost Possible

A chilly air mass is about to make its way down to the Alaskan interior, and so there's a possibility that we might see the first sub-freezing temperature of the season at Fairbanks airport this weekend.  The National Weather Service shows temperatures no lower than 37 °F at the airport in the next several days, but last night's MOS (Model Output Statistics) forecast showed 30 °F for Sunday morning.  Of course, rural valley locations are much more likely to see a freeze.

A quick look at the upper-air history at Fairbanks reveals that 850 mb temperatures typically have to drop below freezing, and often several degrees below freezing, before a freeze is observed at Fairbanks airport.  The chart below shows the lowest 850 mb temperature on the day of the first frost, or in the two days prior to the first frost (to allow for a lag before skies clear out and permit frost).  Note that frost in August (red markers) is usually associated with 850 mb temperatures below -2 °C, and it's not yet clear whether it will get that chilly aloft this weekend (the model temperature forecast is also shown below).

With the calendar still showing August, and with no August frost seen at Fairbanks airport in the past 17 years, I'd say it's probably not going to happen; but with clear skies and light winds expected, it's not out of the question.  By the way, the 1981-2010 median date for first sub-32 temperature is September 11.

Daily Temperature Departures: 2004-2013

This is a follows-up on Richard's Autumn Warming post from a few days ago. Richard's main focus was the relationship between warming in October and how far from the Arctic Ocean its effects are felt due to a decline in sea ice coverage. However, seeing the autumn charts made me wonder what the rest of the year looks like. Therefore, let's see how the departures look for all 365 days of the year.

I replicated Richard's technique of averaging the departure from normal for a 10-year period (I used 2004-2013) for each station in Alaska. Only stations with at least 80% coverage during the 10 years were used. Stations also had to have 1981-2010 normal daily temperature values published by NCDC. In addition, I smoothed the daily data for each station using a 15-day moving average just as Richard did. Without smoothing, the data really jump around a lot. Figure 1 shows the smoothed (red) versus unsmoothed (blue) 10-year daily temperature departure for Fairbanks.

Figure 1. Average daily temperature departure from the 1981-2010 normal at the Fairbanks International Airport during the 2004-2013 time period. The blue lines are the raw daily averages and the blue line represents a 15-day smoothing of the raw data.

Statewide Departure Lines

Using the example from Figure 1, we can repeat the procedure for all other stations in Alaska with sufficient data. Figure 2 shows the result of all 112 stations in Alaska that had at least 80% data coverage and published normal values from NCDC.

Figure 2. Average daily temperature departure from the 1981-2010 normal for 112 stations in Alaska during the 2004-2013 time period. 

There are several interesting patterns that are clearly shown in Figure 2. First, there is a very high degree of variability during this 10-year period. Surprisingly high in my opinion. Over a course of 10 years I would expect the highs and lows to converge; but that is not the case at all. Second, there is a remarkable consistency to the patterns. Alaska has endured an unusual string of consecutive warm Februarys and Octobers and an unusual string of cold Novembers and Januarys. Anecdotally, I have noticed that during the cold half of the year, major patterns seem to last 30 days and they have quite remarkably coincided with calendar months. In the case of February, it seems to warm up dramatically around the 1st or February every year. Of course it is just a coincidence but it sharpens the peaks and troughs on the charts.

Even though Figure 2 clearly shows that the magnitude and direction of the anomaly vectors across the state are remarkably consistent, I thought it would be interesting to see how the trends vary across different regions. Figure 3 shows the average daily departure for all thirteen climate divisions (Bieniek et al.). Figure 4 shows the boundary of the regions. As with Figure 2, there is a remarkable consistency between climate divisions. The only notable exception is the North Slope region during November. Every other station showed strong negative anomalies during the 10-year period but the North Slope region was above normal. I would argue that if the statewide October anomaly is a result of the relativly ice-free Arctic Ocean, which is a reasonable assumption, that effect only lingers across the North Slope into November and the rest of the state no longer feels the influence of the uncovered water.

Figure 3. Average daily temperature departure from the 1981-2010 normal for 13 climate divisions in Alaska during the 2004-2013 time period.

Figure 4. Boundaries of the climate divisions of Alaska (Bieniek et al.). The yellow dots indicate the location of the 112 stations used in the analysis.


*** Update Section ***

Long-Term Precipitation Trends – Part II: Entire U.S.

Last week we posted a write-up of long-term precipitation trends in Alaska. Since it was subtitled "Part I: Alaska," some may have wondered if there was going to be a Part II. Normally if I run analysis for the entire U.S. I won't add it to the Deep Cold Blog since the focus is Interior Alaska but since a few people have found it to be of interest, I'll go ahead and put it up. This originally ran on a Weather Underground blog post a few days ago. 

*********

Significant research has been conducted in recent years regarding changes in precipitation amounts and patterns in a warming climate. From a theoretical perspective, warmer air holds more moisture so increases in temperature should lead to increases in precipitation. On the flip side, increased temperatures may dry out soils and lakes (sources of moisture), cause air currents to change, or lead to other situations that counter-balance the increase in atmospheric moisture.

A chapter from the recently released National Climate Assessment discusses the trends in long-term heavy precipitation events for the entire U.S. during the last century. In particular, they note how the proportion of annual precipitation from extreme events has increased since the 1950's. The map below shows Figure 2.18 from that report. The map shows that large increases in very heavy precipitation events have been observed in the eastern half of the country.

Figure 1. Map from National Climate Assessment (Figure 2.18) showing the observed change in heavy precipitation events.

I am interested in knowing how the rate of heavy precipitation events has changed at smaller geographical scales. Therefore, I decided to look at all airport stations in the U.S. that have a continuous record dating back multiple decades. In this instance, a beginning point of 1949 was chosen because 207 stations have complete precipitation records between 1949 and 2013 (additional stations with 1 or 2 missing months during the same time period will be added at a future date). This is also a long enough period of time to smooth out increases or decreases due to cyclical climate oscillations with short (<10 year) periods. Cooperative stations were excluded since the time of observation is not consistent from one station to the next and in some cases it changes intra-annually at single stations. Therefore, only airport stations with midnight-to-midnight reporting times were used.

The 207 stations are nicely distributed geographically with a slightly higher density east of the Rocky Mountains and a lower density west of the Rocky Mountain Front Range. The following map (Figure 2) shows the distribution of stations.

Figure 2. Locations of airport stations with complete precipitation data from 1949-2013. A total of 207 stations met the criteria. 

For the purpose of this analysis, we are not studying the temporal spread of singular heavy rain events – just the frequency of high precipitation events. In fact, the year with the highest precipitation event for the 1949-2013 time period at each station is not statistically significant when grouping the years into eight categories. Figure 3 shows the year range of the highest precipitation event for each of the 207 stations. There is a slight tendency for the records to be more frequent in recent years but the significance level (p-value) is only 0.15 and is therefore not significant at the 95% or 90% level. If there was an 85% significance category it would fall within that bound (See Figure 4).

Figure 3. Year when highest calendar day precipitation event for all stations during the 1949-2013 time period was recorded.

Figure 4. Number of maximum precipitation events grouped by year of occurrence for all 207 stations during the 1949-2013 time period.

Methodology

For each station, a linear regression line was fitted to the number of days per year that met or exceeded A) 0.05", B) 0.50", C) 1.00", and D) 2.00". The first value (0.05") was chosen as a proxy measure for the overall number of rainfall events per year. A smaller value was not used so that future research can extend the analysis to Cooperative stations. Those stations, especially early in their climate records, missed some small precipitation events. The 0.05" value allows us to determine if all precipitation events are increasing or decreasing – not just heavy events.

Once a linear regression was completed for each station at each of the four precipitation thresholds, a probability value (p-value) was computed. The p-value is a statistical measure of significance. A p-value less than 0.05 indicates that there is a less than 5% chance that the statistical trend is random. A p-value less than 0.10 indicates that there is a less than 10% chance that the statistical trend is random. By convention, a p-value greater than 0.10 is considered not statistically significant.

As an example, the Dallas Fort Worth International Airport (GHCN ID: USW00003927) saw a slight decrease in number of days with at least 0.05" of precipitation between 1949 and 2013. However, the p-value was 0.93 – indicating near total randomness in the distribution. Looking at the number of days with at least 0.50", there was an increase over time and the p-value was 0.045. Since this number is less than 0.05, the upward trend is considered significant at the 95% level. The p-value for the trend in days with at least 1.00" was 0.16 and for days with at least 2.00" was 0.70 – both not significant at the 95% or 90% levels. Collectively, we conclude that the Dallas Fort Worth International Airport has observed a statistically significant increase in the number of days with at least 0.50" of precipitation but all other thresholds were not significant.

Statistical Significance Maps

Instead of plotting percent change (or raw value change) for each station from 1949-2013, I decided to plot statistical significance – using the aforementioned p-value. For example, if a station showed at 20% increase in the number of days with 1.00" or more between 1949 and 2013, 1 or 2 years might be responsible for all of the increase. Therefore, the increase, in that example, is an aberration and not an actual trend. However, we can compute a statistical significance for that station's trend line and report back whether or not the 20% increase was meaningful at the 95% or 90% significance level. For all of the statistical significance calculations and maps, a station must have an average at least 0.5 days per year to calculate a trend – otherwise they are identified as "too few events."

Days per year with at least 0.05"

Most of the stations in the U.S. experienced no statistically significant increase or decrease in the number of days with at least 0.05" of precipitation. A band of stations from the Dakotas to the eastern Great Lakes saw statistically significant increases and some areas in the Southeast saw statistically significant decreases but most of the U.S. was nondescript.

Figure 5. Significance map of trend in number of days with 0.05" of precipitation or greater during the 1949-2013 time period.

Days per year with at least 0.50"

Using a threshold of 0.50", patterns begin to emerge. Many stations from northern Texas to the Dakotas and then eastward to include the entirely of New England saw a statistically significant increase in the number of days with at least 0.50" of precipitation. Much of the West consistently recorded a decrease in the number of days with 0.50" of precipitation but only a few stations were statistically significant.

Figure 6. Significance map of trend in number of days with 0.50" of precipitation or greater during the 1949-2013 time period.

Days per year with at least 1.00"

The statistical significance pattern is even more apparent when looking at days with at least 1.00" of precipitation. Nearly 90% of stations east of the Rocky Mountains saw an increase in the number of 1.00" precipitation days and approximately half of those stations met the 95% statistical significance threshold. Notice that some stations in the Intermountain West receive too few days per year (<0.5) to be included in the analysis.

Figure 7. Significance map of trend in number of days with 1.00" of precipitation or greater during the 1949-2013 time period.

Days per year with at least 2.00"

At the 2.00" threshold, the trend direction (positive or negative) and the significance levels are not nearly as distinct as they were for the 0.50" and 1.00" events. Nevertheless, a clear pattern exists in the northeastern portion of the country and a strong majority of stations east of the Rocky Mountains saw an increase in the number of days with at least 2.00" of precipitation. West of the Rocky Mountain Front Range, a majority of stations (65) receive too few days per year to make meaningful assessments.

Figure 8. Significance map of trend in number of days with 2.00" of precipitation or greater during the 1949-2013 time period.

All Stations Averaged Together

The primary purpose of this analysis was to assess changes in the frequency of heavy precipitation events in small geographical units. That being said, it is helpful to look at the results when all stations are averaged together. To do this, every station had an average value computed representing the average number of days with at least a certain amount of precipitation (e.g., >=0.05"). Then the value for each year was compared against that average and a percentage above or below the average value was recorded. If, for example, a station averaged 80 days per year with at least 0.05" of precipitation, a year with 88 days would be recorded as 110% of the average. This averaging technique was performed for all station, in all years, for each precipitation threshold. Using percentages prevents stations with large numbers of precipitation days (e.g., New Orleans) from overwhelming stations with small numbers of days (e.g., Las Vegas). Figure 8 shows the number of days per year with at least 1.00" of precipitation between 1949 and 2013 as an example of the spatial variability in heavy rainfall events.

Figure 9. Average number of days with 1.00" of precipitation or greater during the 1949-2013 time period.

As you can see, stations in the southeastern corner of the U.S. have far more days per year with at least 1.00" of precipitation. If, for example, the number of days in Mobile, AL, and Salt Lake City, UT, both increased by 2 days per year, using raw numbers masks the change in Salt Lake City whereas using percentages does not. Therefore, any methodology that does not normalize the data runs the risk of being a de facto analysis of only those stations that have large average annual precipitation amounts.

Days per year with at least 0.05"

The change in the number of days per year with at least 0.05" is pretty chaotic across the entire U.S. There are long periods with consistently upward or downward trends but overall the values are pretty flat. Beginning in 1998, the rate of change dropped noticeably. This also corresponds to period of record or near record worldwide temperatures. The p-value of 0.54 indicates that the overall trend is not statistically significant.

Figure 10. Annual average of each station's percentage from the long-term average number of days with at least 0.05" of precipitation.

Days per year with at least 0.50"

The nationwide change in the number of days per year with at least 0.50" consistently increased for most of the 65 year analysis period. As with the 0.05" chart, the rate of change dropped in 1998. The p-value of 0.01 indicates that the trend is statistically significant at the 95% (and even at the 99%) level over the course of the analysis period.

Figure 11. Annual average of each station's percentage from the long-term average number of days with at least 0.50" of precipitation.

Days per year with at least 1.00"

The change in the number of days per year with at least 1.00" was strongly positive. The increase in the number of days per year is greater than 10% and the post-1998 deviations from the prior two charts are just 1 or 2 year anomalies on the 1.00" chart. In fact, the p-value of 0.003 indicates that the trend is statistically significant at the 99% level. A total of 17 stations that do not average at least 0.5 days per year with 1.00" of precipitation or greater were excluded from the analysis.

Figure 12. Annual average of each station's percentage from the long-term average number of days with at least 1.00" of precipitation.

Days per year with at least 2.00"

The change in the number of days per year with at least 2.00" was even more strongly positive. The p-value of 0.0001 indicates a very high degree of statistical significance. A total of 65 stations that do not average at least 0.5 days per year with 2.00" of precipitation or greater were excluded from the analysis. Since the vast majority of the excluded stations are in the western U.S., this chart essentially reflects the statistical trend of the eastern half of the U.S. only. As Figures 5 and 6 demonstrate, the trends for the eastern half of the U.S. is much more prominent than for the western half.

Figure 13. Annual average of each station's percentage from the long-term average number of days with at least 2.00" of precipitation.

Conclusion

We showed that the rate of small precipitation events has not changed much in the last 64 years (see Figure 5). However, when the precipitation intensity rises, so does the strength of the statistical significance. Most of the eastern half of the U.S. has experienced an increase in the number of days with at least 0.50", 1.00", and 2.00" of precipitation. The western half of the country has, on average, seen a slight decline in the rate of those precipitation thresholds when enough observations are available for analysis– but not at a statistically significant level.

At the station level, the long-term trend of days at different intensity thresholds tells a more complete story than just looking at regional data using state boundaries. While data at an individual station is not sufficient to draw very many conclusions, aggregating station data in this manner allows us to draw new conclusions about how precipitation patterns change over space and time.

A complete list of the stations used in the analysis can be found Here.