The US state most often linked to climate change is Alaska. This is probably because it is seen as having an Arctic climate even though only about a third of the state actually lies within the Arctic Circle. In fact Alaska is no more northerly than Norway and its Aleutian Island chain stretches further south than London and Berlin. It has an area three times that of France but its population is less than that of Marseille, yet it has an extensive network of weather stations that is greater in data quality than that seen in many industrialized countries. Ordinarily this should be sufficient to determine the temperature change for Alaska to a high level of precision but it isn't. In fact the data is so inconclusive it is difficult to determine whether Alaska has warmed at all over the last one hundred years let alone quantify that warming and discern when exactly it occurred. This is because the natural variation in the long term temperature averages is far greater than the likely warming.
There are one hundred stations in Alaska with over 480 months of data before 2014 including seven long stations with over 1200 months of data. Of the 93 medium stations with over 480 months of data twenty have over 1000 months of data (for a full list of stations see here). The locations of these stations are shown in Fig. 139.1 below.
Fig. 139.1: The (approximate) locations of the 100 longest weather station records in Alaska. Those stations with a high warming trend between 1911 and 2010 are marked in red while those with a cooling or stable trend are marked in blue. Those denoted with squares are long stations with over 1200 months of data, while diamonds denote medium stations with more than 480 months of data.
The map in Fig. 139.1 shows that most of the temperature data for Alaska come from stations that are outside the Arctic Circle. In fact of the one hundred longest stations in Alaska only eight are actually inside the Arctic Circle. And while the remainder are fairly evenly distributed geographically, there are significant clusters of stations around Anchorage, Fairbanks and the panhandle along the coast in the southeast between the the border of Canada and the Alexander Archipelago. As usual for simplicity I will disregard this clustering and assume it makes very little difference to the measured temperature change as it only affects the contribution or weighting of about 15% of stations.
In order to quantify the changes to the climate of Alaska the temperature anomalies for all stations with over 480 months of data before 2014 were determined and averaged. This was done using the usual method as outlined in Post 47 and involved first calculating the temperature anomaly each month for each station relative to its monthly reference temperatures (MRT), and then averaging those anomalies to determine the mean temperature anomaly (MTA) for the country. This MTA is shown as a time series in Fig. 139.2 below with the MRTs for each station calculated using data between 1961 and 1990, (again using the methodology outlined in Post 47).
Fig. 139.2: The mean temperature change for Alaska since 1900 relative to the 1961-1990 monthly averages. The best fit is applied to the monthly mean data from 1921 to 2000 and has a positive gradient of +0.31 ± 0.31 °C per century.
The data in Fig. 139.2 above illustrates the difficulty of determining a definitive temperature trend when the data is subject to significant variability over time. In this case choosing to fit to the data from 1921 to 2000 leads to a small positive gradient of 0.31°C per century, but this is no bigger than the uncertainty and so is not statistically significant. If other fitting intervals are chosen then the gradient can be significantly different. For example, an interval of 1921-1995 results in a gradient of 0.18°C per century while 1926-2005 produces 0.96°C per century. All of which poses the awkward question, which result is correct?
In my opinion there is no obvious answer, but there are two factors that we could consider that may shed some additional light on the problem. The first of these is to choose an appropriate fitting interval based on the cycle of the natural variations (i.e. fitting from peak to peak), while the second is to concentrate on data that is the result of averaging the greatest number of station records.
In Post 4 I explained how the best fit line to a single period of a sine wave gives a non-zero gradient (see Fig. 4.7) whereas fitting to a cosine wave does not. This is because a cosine wave is symmetric about the y-axis while the sine wave is anti-symmetric. As most temperature data tends to oscillate over time due to natural variations it therefore follows that the gradient of any fit to that data will depend on the interval chosen relative to the peaks of those natural oscillations.
In order to avoid biasing the gradient due to asymmetry in the fitting range, the range should be symmetric relative to the natural oscillations. These natural oscillations are seen most clearly in the 5-year moving average (see the yellow curve in Fig. 139.2). So the fitting range should be chosen so that it starts and ends on a peak in the 5-year average, or alternatively starts and ends on a trough. The best fit in Fig. 139.2 does not do this. It starts near a trough at 1921 and ends on a plateau in 2000. But if we change the fitting interval from 1914 to 2003 then the interval starts and ends on a peak in the 5-year average. The result is the best fit shown in Fig. 139.3 below.
Fig. 139.3: The mean temperature change for Alaska since 1900 relative to the 1961-1990 monthly averages. The best fit is applied to the monthly mean data from 1914 to 2003 and has a positive gradient of +0.71 ± 0.26 °C per century.
The gradient of the best fit in Fig. 139.3 is more than twice that in Fig. 139.2 even though the data hasn't changed. This is simply a result of changing the fitting interval. Of course the underlying reason why a change of fitting interval makes such a big difference in this case is that the natural fluctuations in the 5-year average are so large. These changes in temperature can exceed 2°C in less than five years. So we could ask, is the temperature rise of about 0.7°C indicated by the best fit in Fig. 139.3 really that significant in comparison?
Fig. 139.4: The number of station records included each month in the mean temperature anomaly (MTA) trend for Alaska in Fig. 139.2 and Fig. 139.3.
The second factor in determining any choice of fitting range is the quantity of data available. The graph in Fig. 139.4 above shows the number of stations included in the MTA in Fig. 139.2 and Fig. 139.3. From 1920 onwards there are over twenty stations each month. In the previous post and in Post 57 I argued that at least ten, and possibly over twenty-five stations are needed in order for the MTA to be reliable, so this condition is satisfied for all months after January 1920. The data before 1920 will therefore be much less reliable, but there is still enough data to allow us to calculate an approximate MTA as far back as the 1820s. This is shown in Fig. 139.5 below.
Fig. 139.5: The mean temperature change for Alaska since 1820 relative to the 1961-1990 monthly averages. The best fit is applied to the monthly mean data from 1911 to 2010 and has a positive gradient of +0.73 ± 0.22 °C per century.
The data in Fig. 139.5 indicates that it is possible to calculate and MTA as far back as 1829, but before 1900 there are gaps in the data and most of the MTA data for this period is based on an average of anomaly data from less than three different stations. So that raises questions over its reliability.
So how should we interpret this data? The station frequency data in Fig. 139.4 suggests only data after 1900 or even 1920 is sufficiently reliable. As for the data after 1900, there are many ways to interpret it. For example, if we just look at data from 1901 to 1975 the best fit (as determined from trough to trough) is strongly negative (see Fig. 139.6 below). But after 1975 the temperature appears to increase abruptly by about 1°C. So is this interpretation of the temperature trend any more believable than those shown in Fig. 139.2 or Fig. 139.3? It is hard to tell, again because of the high level of natural variability in the data which could be varying on multiple timescales. Such multi-frequency variability is potentially indicative of chaotic or fractal behaviour as I discussed in Post 9, Post 17 and Post 42.
Fig. 139.6: The mean temperature change for Alaska since 1900 relative to the 1961-1990 monthly averages. The best fit is applied to the monthly mean data from 1901 to 1975 and has a negative gradient of -0.52 ± 0.33 °C per century.
If we next consider the change in temperature based on Berkeley Earth (BE) adjusted data we get the MTA data in Fig. 139.7 below. This again was determined by averaging the anomalies for each month from the one hundred longest stations in Alaska and suggests that the climate of Alaska has warmed by over 1°C since 1870, but with large natural variations of up to 1.5°C in the 10-year average.
Fig. 139.7: Temperature trends for Alaska based on Berkeley Earth adjusted data. The best fit linear trend line (in red) is for the period 1876-2010 and has a positive gradient of +1.00 ± 0.06°C/century.
Comparing the curves in Fig. 139.7 with the published Berkeley Earth (BE) version for Alaska in Fig. 139.8 below we see that there is good agreement between the two sets of data as far back as 1880. This indicates that the simple averaging of anomalies used to generate the BE MTA in Fig. 139.7 using adjusted data is as effective and accurate as the more complex gridding method used by Berkeley Earth in Fig. 139.8. In which case simple averaging should be just as effective and accurate in generating the MTA using raw unadjusted data in Fig. 139.2 and Fig. 139.5. In other words, any discrepancy between the adjusted data in Fig. 139.7 and the unadjusted data in Fig. 139.5 cannot be due to the averaging process. Any form of weighted averaging would also not affect the results.
Fig. 139.8: The temperature trend for Alaska since 1820 according to Berkeley Earth.
Most of the differences between the MTA in Fig. 139.6 and the BE versions using adjusted data in Fig. 139.7 are instead mainly due to the data processing procedures used by Berkeley Earth. These include homogenization, gridding, Kriging and most significantly breakpoint adjustments. These lead to changes to the original temperature data, the magnitude of these adjustments being the difference in the MTA values seen in Fig. 139.5 and Fig. 139.7.
Fig. 139.9: The contribution of Berkeley Earth (BE) adjustments to the anomaly data in Fig. 139.7 after smoothing with a 12-month moving average. The blue curve represents the total BE adjustments including those from homogenization. The linear best fit (red line) to these adjustments for the period 1921-2000 has a positive gradient of +0.262 ± 0.009 °C per century. The orange curve shows the contribution just from breakpoint adjustments.
The magnitudes of these adjustments are shown graphically in Fig. 139.9 above. The blue curve is the difference in MTA values between adjusted (Fig. 139.7) and unadjusted data (Fig. 139.5), while the orange curve is the contribution to those adjustments arising solely from breakpoint adjustments. The overall adjustment from 1920 to 2000 is small, about +0.2°C. Nevertheless, it can be seen in the difference in the 5-year means (see Fig. 139.10 below) for the unadjusted data (blue curve) and the adjusted data (red curve). The difference, though, is about ten times less than the variability in the two MTAs over time. The data in Fig. 139.10 also highlights the difficulty in interpreting the data. If the data between 1940 and 1980 were missing or ignored, then one could postulate that Alaska has seen fairly consistent warming since 1900 amounting to about 1°C in total. But if the 1940-1980 data is included the data all looks very random.
Fig. 139.10: The 5-year mean temperature change for Alaska since 1900 based on the original raw data (in blue) and the Berkeley Earth adjusted data (in red).
Summary
The temperature data for Alaska demonstrates the difficulty in determining an accurate temperature trend for a region when the climate is subject to a high degree of variability.
It is possible that the climate has warmed by almost 1°C since 1900 (see Fig. 139.3), or it might not have warmed at all (see Fig. 139.2).
If the climate has warmed, this warming may have been fairly continuous (see Fig. 139.3), or it could have been fairly recent, occurring mainly after 1980 (see Fig. 139.6).
The one thing we can say is that the difference between the temperature rise based on Berkeley Earth adjusted data (see Fig. 139.7) and that based on the raw unadjusted data (see Fig. 139.5) is small (less than 0.3°C) and much less that the 5-year natural variability of the data (about 2°C).
Acronyms
BE = Berkeley Earth.
MRT = monthly reference temperature (see Post 47).
MTA = mean temperature anomaly.
List of all stations in Alaska with links to their raw data files.
