AGCI Session II: Characterizing and Communicating Scientific Uncertainty
Session Chairs: Dr. Richard H. Moss and Dr. Stephen H. Schneider
July 31 to August 8, 1996
Uncertainties in Observed Changes in Climate Variables
Tom Karl
National Climate Data Center
Asheville, North Carolina
Karl discussed uncertainties in observed changes in global temperature, precipitation, and other climate variables, and the three -star approach to ranking confidence levels used in the 1995 IPCC assessment. In general, the level of confidence has been raised by additional data since the 1992 assessment. In particular, there is now a high degree of confidence (three stars) that the global temperature increase has been between 0.3 and 0.6°C from the late 19th century to the present. On the other hand, there is less confidence in our understanding of the behavior of clouds and water vapor, and estimates in these areas were assigned one star in the confidence ranking. In Karl's IPCC Working Group I discussions, there was very little serious disagreement about what confidence levels were appropriate for the various estimates. Figure 2.10 illustrates and summarizes the basic consensus reached on stratospheric cooling, tropospheric warming, retreat of mountain glaciers, and other climate variables.
The uncertainty range is largely related to both systematic and random errors in data sets. Systematic errors include urban heat island effects, poor exposure of instruments (particularly for 19th century land -based data), uninsulated or poorly insulated buckets used to measure ocean temperature from ships, and differences between these data and ship engine intake measurements and hull contact thermistors. Random errors include inadequate spatial sampling, observer time changes (may be systematic or random), switch to automated instruments or other instrument changes, changes in local land use (such as effects of desertification), and instrument re-locations. While many errors have been corrected for, there is still a degree of uncertainty with regard to some data.
Figure 2.11 shows the combined land-air and sea surface temperature anomalies from 1860 to the present. This data set raises the question of how to decide when in the record we should begin having confidence in data. Older data are not as reliable for reasons including the errors mentioned above and because spatial sampling increases after 1900. Karl treats data collected in the 19th century with skepticism. It is thus unclear what beginning date should be used. Karl says that selection of the initial date should be dependent on the question(s) being asked.
In general, the
level of confidence has been raised by additional data since the 1992
assessment. In particular, there is now a high degree of confidence
(three stars) that the global temperature increase has been between
0.3 and 0.6°C from the late 19th century to the present.
An analysis of minimum and maximum temperatures at non-urban stations reveals that minimum temperatures are increasing by 1.33°C per 100 years while maximum temperatures are increasing by 0.83°C per 100 years. Comparisons between rural, urban and metropolitan stations show that the urban heat island effect is discernible even in small cities, though the effect is not large. The greatest warming has been measured in metropolitan areas, with less warming in smaller cities, and still less in rural areas.
There is a high
degree of confidence (three stars) in the data that show that alpine
glacier mass is decreasing globally, and similarly, that snow cover
is decreasing.
Assuming major errors are independent, with a 0.1°C/century error in land data, 0.1°C/century in ocean data, and another 0.1°C/century due to sampling inadequacy, the error interval is 0.17°C. Additionally, there is an urban heat island residual bias of less than 0.05°C/century. With the warming calculated as ~0.55°C/century (quoted as 0.5°C /century, after consideration of residual urban affects), the uncertainty band established by IPCC of 0.3 to 0.6°C/century in 1990 was not altered in 1995. The question was raised as to why this range includes significantly more on the down side than the up side (a 0.3 to 0.6°C range given for a calculated rise of 0.5°C). Karl expressed the belief that this range is likely to change in the next IPCC assessment but that there was not enough new data to make the changes at this time.
There is a high degree of confidence (three stars) in the data that show that alpine glacier mass is decreasing globally, and similarly, that snow cover is decreasing. Satellite data of snow cover changes over the Northern Hemisphere from 1972 to 1992 reveal a strong correlation (r=0.82) between temperature rise (in area-mean maximum) and reduced snow cover. This provides added confidence in near-surface global warming.
In regard to the question of whether the climate has become more variable or extreme, there are inadequate data and analyses to say anything about global scale changes. In some aspects, the climate has become more extreme, but in other aspects, there is evidence of little change or even a decrease in extremes. For example, there is a clear trend toward increasing precipitation rates in the U. S. and northeast Australia, while other areas (such as China) show little change. Other trends include increased intensity in extratropical cyclones in the North Atlantic since the late 1980s, and a small decrease in hurricane frequency and intensity in the North Atlantic over the past 40 years.
In the Working Group I discussions, there were some controversies about what can be said about extreme events. For temperature, the statement was made that a general warming tends to lead to an increase in extremely high temperatures and a decrease in extreme lows ( e. g., frost days). It was also stated that small changes in the mean (or climate variability) can produce large changes in the frequency of extremes; a small change in variability has a stronger effect than a similar change in the mean. Karl asks, "What does this mean? The units are not even similar." A change in variability (variance) of 50 percent has a smaller effect on extremes than does a change in the temperature of 1.5°C (related to the extreme temperatures in Chicago, for example). Variability is not necessarily more important than changes in the mean with respect to extremes, Karl says, arguing that the language poorly reflects our knowledge.
Regarding mid-latitude storms, conclusions regarding extreme events are uncertain. The pole-to-equator temperature gradient has increased at high elevations but decreased at low elevations. There is some evidence for a recent increase in storminess around the North Atlantic (e .g., the 1988/89 abrupt increase) but a more general increase since the 1970s, implying that the North Atlantic Oscillation may not be dominant. There is not clear evidence of any uniform increase, however, and all of this led to controversy on this subject. As for tropical storms, no assessment is possible about what to expect in the future.
The number of frost days is decreasing in many parts of the world; is that an extreme event? The Chicago heat wave of 1995 was certainly an extreme event. Karl compared annual 2-day episodes of maximum temperatures over the past four decades for Chicago and found a discernible upward trend. He also used a statistical model to examine the effect of changing four variables (mean temperature, variance of daily temperature, interannual variance, and day-to-day persistence) on the occurrence of such events. Results suggest that increases in the mean and in variability increase the probability of extreme heat waves. One striking result of this research is that under present conditions, the 1995 heat wave is a 1 in 20 year occurrence for the extreme high, but less than a 1 in 1000 year event for the elevated minimum temperature during the 2 days of most intense heat. Raising the mean temperature to levels that may be realized by the end of the next century raises the probability of such an event to a 1 in 3 year event for the extreme high and a 1 in 200 year event for the elevated minimum temperature still quite unusual.
Results suggest
that increases in the mean and in variability increase the
probability of extreme heat waves.