Problems with the Use of Climatological Data to Detect Climatic Change at High Latitudes

S. A. Bowling
(Alaska Climate Research Center, Geophysical Institute, University of Alaska Fairbanks)
(Originally presented at the International Conference on the Role of the Polar Regions in Global Change, held June 11-15 1990 at the Univwersity of Alaska Fairbanks, and published on pp 206-209 of the Proceedings for that conference.)

Abstract
Although warming due to increased amounts of CO2 and other greenhouse gasses in the atmosphere is predicted to be greatest in high latitudes, results of the GISS model have already indicated that the ratio of warming to interannual variability will be relatively small, which will make any change hard to detect (Hansen, 1988). In addition, the climatological data set at high latitudes is scanty and subject to most of the same problems as those in the temperate zone. In fact, the extreme ground inversions, low sun angles, and seasonal polar night or continuous daylight conditions may lead to systematic errors with magnitudes much greater than would be predicted from midlatitude experience.

The Alaskan record demonstrates possible magnitudes of some of these systematic errors. Both winter and summer heat island effects are large. Site changes (including documentation problems) may have unexpectedly large effects, and virtually the entire state is affected by a change in time zones in 1983.


Introduction
Instrumental records from high latitudes suffer from the same problems as those elsewhere: changes in observation methods, observation sites, instrumentation and site surroundings can all affect the validity of apparent long-term trends. Some of these problems are more difficult to correct for at high latitudes. Thus low station density makes it unusually difficult to correct for station location changes, and extreme variation in sunset and sunrise times is expected to complicate correction for observation time. In Alaska, this correction must be applied to almost every station, as time zone changes in October 1983 produced observation time changes of 1 to 2 hours at every station in Alaska except Yakutat. In other cases, problems may actually be of greater magnitude at high latitudes. For instance, extreme ground inversions lead to large temperature changes over short distances (magnifying the potential effects of station changes) and unexpectedly large urban heat island effects.

This paper deals with three examples of local variations or apparent "climate changes" probably due to urban effects, plus one possible real variation in Alaskan climate.

An apparent long-term summer warming
While looking for differences in the character of summer and winter temperature records across the state, we noticed that virtually all of the large negative summer minimum temperature anomalies at Fairbanks occurred during the early years of the record, while the large positive anomalies were recent. When trend lines were fitted to individual monthly mean minimum and maximum temperatures, the May, June, July, and August minimum temperatures at Fairbanks showed significant upward trends with time


Figure 1. Summer-month mean daily minimum temperatures at Fairbanks. This is an extremely heterogeneous record, with major station moves in 1929, 1942, 1943, and 1951, but no similar trends appear in other months or at other stations checked. Correlation coefficients, shown in the table below, are significant at the 99.9% confidence level.
MonthJuneJulyAugustSeptember
Slope (degrees F / year)0.0840.0910.1020.075
Correlation coefficient0.510.640.770.56

No trends were noted in other seasons, in summer maxima, or at other stations. As the Fairbanks record is far from uniform, we repeated the analysis using only data from the current Fairbanks International Airport location, and found similar trends, though with slightly weaker correlations

Figure 2. Same as Figure 1, but only for the recent, homogeneous period of record. For 38 years of record (36 degrees of freedom) a correlation coefficient of roughly .32 would be significant at the 95% level, and one of .4 at the 99% level.).
MonthMayJuneJulyAugust
Slope (degrees F/year)0.0950.0850.1430.078
Correlation coefficient.39.37.64.33

As there is no sign of such effects in areas with smaller populations, it appears that the trend is most likely due to a combination of station changes and a summer climatological heat island (due to the increasing amount of building materials in the vicinity of the airport) of previously unsuspected intensity.

The Fairbanks area is subject to strong urban heat island effects (measured to be as much as 13 °C) when skies are clear in winter (Bowling and Benson, 1978). Summer heat islands have hardly been looked at, but they now appear likely to have more effect on recorded climate than do the large winter ones, which are masked by large year-to-year variability in winter and seem to be more confined to the city core.

Why doesn't Anchorage show a similar effect? The most probable culprit is the Good Friday Earthquake of 1964. The earthquake leveled the Control Tower, which required that the instruments be relocated. The max/min thermometers were at Point Campbell, farther from pavement and nearer the waters of Cook Inlet, by 1971, but there is apparently no published record of their location during the intervening period.

Local variability during January 1989
January of 1989 was a record setter in much of Alaska. Not, however, officially in Fairbanks, where the official temperatures are recorded downwind of the city. Unofficial temperatures on the upwind side of Fairbanks were much lower, as were many at surrounding climatological stations


Figure 3. Daily mean temperature anomalies at Fairbanks and nearby stations during a severe cold spell. Heavy line is Fairbanks International Airport, elevation 133 m. Other stations are: Col Ob = College Magnetic Observatory, on the West Ridge of the University of Alaska at elevation 189 m; Eielson = Eielson Air Force Base, about 40 km SE and upvalley of the Airport at elevation 167 m, Ex Sta is the University Experimental Station, near the base of the West Ridge at elevation 145 m, No Pole is North Pole, between Fairbanks and Eielson at elevation 145 m, and Col 5 NW is in the uplands 5 miles NW of the two campus sites, elevation 290 m.).

All of the stations shown are within a radius of 40 km, with Eielson being the only one that far from the Fairbanks site. The difference between the experimental station and the airport can probably be attributed to the heat island effect combined with a time of observation error; the spread among the other lowland stations indicates local variability. The difference between College 5 NW and the other stations during the first half of the cold spell indicates a strong inversion, especially on the 23rd. [n.b. The figure above is missing 2 stations: North Pole (low lying and colder than those shown) and College 5 NW, which is at a higher elevation and warmer than Fairbanks during the early part of the period shown. I'll get a version with all 6 stations in when I can find the file.]

The Misplaced Move and the Encroaching City
Recorded station moves are not necessarily consistent among different sources, nor are the dates always accurate. Take the case of the downhill move at the University Experiment Station. The station list in Climatological Data, Alaska shows an elevation change from 500 feet to 475 feet in the summer of 1947. Was it real, or the result of re-surveying the area? With known current winter inversion strengths in the University area, such a move could have produced a decrease of a few degrees in recorded winter temperatures. Comparison of the University record for December and January with the Fairbanks record from Weeks Field (near where the Borough Library is now located) did indeed show either a decrease in University temperature or an increase in the Weeks Field temperature, but suggested a change in the summer of 1946 rather than 1947. The actual station history for the University Experiment Station confirms that the move was real, but it took place in 1933, 15 years before it was finally brought up to date in Climatological Data.

The real trend of increasing temperature at Weeks Field relative to University affects all summer, fall, and early winter months (Table 1).

Trends of temperature differences (Weeks Field - University), 1943-1951.
Month PointsSlope, °F/yr Correlation CoefficientSignificance
January8 .76 .70Less than.90
February8.08.17Less than .80
March8.89.59>.80
April8.05.05Less than.80
May 9 .21.51>.80
June 9 .29 .74 >.95
July 9 .27.73 >.95
August 8 .34 .84 >.99
September8.42.95>.999
October8 .27.85>.99
November8.65.67>.90
December8.68.71>.95

The trend is most logically explained as due to an increasing heat island effect as Fairbanks grew around Weeks Field. Although the time period is very short for high significance (May 1, 1943 through mid-August 1951) the trends are extremely consistant from month to month outside of the spring period, with an increase of almost half a degree per year


Figure 4. Annual mean differences between Fairbanks Weeks Field temperatures and University Experimental Station temperatures, showing increasing heat island as the city of Fairbanks grew around Weeks Field.

Some of the large deviations from the trend line can even be explained: December 1946 and January 1947, both of which show large positive deviations from the trend line, are known to include a record-setting cold spell which, from previous studies, is known to be an ideal setting for a large heat island effect. But unanswered questions remain. Why does the heat island appear to weaken during late winter and spring? Why do so many months have large deviations from the trend line in 1947?

The 1976 temperature step: a real change?
The 1976-77 winter in Alaska was astonishingly warm. At Fairbanks, pussy willows bloomed in November (author's observation) and daily average temperatures never reached -30°F. Subsequent winters followed the same trend to a lesser degree, with degree days below -40°F showing a reduction quite noticable to long-term residents. An average of four Alaskan stations with good, continuous records, Anchorage, Barrow, Fairbanks, and Nome, show what appears to be a change in mean annual temperatures at around this time


Figure 5. Annual mean temperatures, average of Anchorage, Barrow, Fairbanks, and Nome, since all four airports stabilized positions in the mid 50's. (Changes in instrument location at Anchorage may have led to apparent warmer winters and cooler summers, but Anchorage does not in any way dominate the trend shown, which is strongest at Fairbanks and Nome. Note that the version shown here has been updated from the original paper.

.All of these stations are in areas that could be affected by urban effects, but a similar step, which there appears to interrupt an overall downward trend, appears at McGrath, with a population of under 1000 people. Is it real? And will it last?

In an attempt to find a somewhat independent measure of climate shift, be it temporary or permanant, we used 700 mb grid point heights at 60°N 150°W, 65°N 160°W, 65°N 140°W, and 70°N 150°W to estimate changes in geostrophic flow. The usual net north-south and east-west flows would have been ambiguous as to whether an increase in net southerly flow, for instance, was caused by an increase in the frequency or intensity of southerly flow or a decrease in northerly flow. To get around this, we summed the positive differences between 160°W and 140°W as northerly flow and the negative differences as southerly flow at 65°N over Alaska, dividing by the actual number of soundings used to compensate for missing data. The differences between 60°N and 70°N at 150°W were similarly treated, with positive differences contributing to westerly flow and negative differences to easterly flow.

The somewhat unexpected result was that all four indices decreased in absolute value over the period from 1965 through 1986, suggesting a general reduction in the intensity of circulation over Alaska. The only significant decrease, however, was that in northerly flow, which showed a fairly consistent high level early in the time series, rather violent oscillations for several years around the apparent temperature change, and a second steady period, at around two-thirds the previous level, in the first half of the '80's


Figure 6. Northerly wind index over Alaska, annual values from 1965 through 1986 calculated as described in text. A least-squares regression (not shown) gives a slope of -.4 m yr-1 and a correlation coefficient of .68, significant at the 99% level. Note that this index cannot by its nature be negative, so this trend represents a decrease in northerly flow intensity of more than a third over 20 years..

It thus seems likely that the observed warming in the mid '70's in Alaska is real and associated with a circulation shift. Whether it is part of an unusually long-period oscillation or has something to do with carbon dioxide is still a mystery.

Conclusions
Use of high-latitude instrumental data to deduce long-term trends is risky at best, and should be attempted only with reference to the fullest available station histories and some knowledge of the local topography, settlement history, and microclimates. Changes documented by more than one type of data (e.g., soundings as well as surface temperatures) can be considered better supported than simple temperature measurements, but such data are rarely available for really long time series.



References
Hansen, J.; I. Fung; A. Lacis; D. Rind; S. Lebbedeff; R. Ruedy and G. Russell, 1988: Global climate changes as forecast by Goddard Institute for Space Studies three-dimensional model. J. of Geophys. Res. 93, 9341-9364.

Bowling, S. A., 1986: Climatology of high-latitude air pollution as illustrated by Fairbanks and Anchorage, Alaska. J. of Climate and Appl. Meteor. 25, 22 34.

Bowling, S. A. andC. S. Benson, 1978: Study of the subarctic heat island at Fairbanks, Alaska. EPA Report EPA-600/4-78-027, 149 pp.