"Right after the year 2000," climate change skeptic Tony Heller claimed last month, federal climate scientists "dramatically altered US climate history, making the past much colder and the present much warmer….This alteration turned a long term cooling trend since 1930 into a warming trend." Heller (nom de blog Steven Goddard) says that these adjustments " cooled 1934 and warmed 1998, to make 1998 the hottest year in US history instead of 1934."
Heller's assertions induced a frenzy of commentary, attracting the attention of The Drudge Report, the Telegraph, The Daily Caller, and Fox News. A few days later, the hullabaloo was further stoked by reports that scientists at the National Climatic Data Center (NCDC) had quietly reinstated July 1936 as the hottest month on record in the continental U.S. instead of July 2012. (For the record, the National Oceanic and Atmospheric Administration—the NCDC's parent agency—has declared 2012 the hottest year on record for the lower 48 states, and the months between August 2011 and July 2012 as the hottest 12-month period on record. The year 2012 was also the warmest year in the 36-year satellite temperature record.)
In response to the brouhaha, the NCDC press office sent out a rather defensive statement noting that its new U.S. temperature dataset based on climate division adjustments has, indeed, restored July 1936 to its hellish pinnacle. "We recalculate the entire period of record to ensure the most up-to-date information and to ensure proper comparison over time," said the press release (which, oddly, is not available online). "In this improved analysis, July 1936 is now slightly warmer than July 2012, by a similar very small margin." It added that this "did not significantly change overall trends for the national temperature time series" and that the "year 2012 is still easily the warmest on record."
But never mind the quibbling over which month in the past century was the hottest. Is Heller right when he claims that NCDC scientists are retrospectively fiddling with the national thermostat to bolster the case for man-made global warming?
The answer is complicated.
When Heller produced his temperature trend for the continental United States, he basically took the raw temperature data from the U.S. Historical Climatology Network from 1895 to the present and averaged them. He made no adjustments to the data to take into account such confounders as changes in location, equipment, time of observation, urban heat island effects, and so forth. Heller argues that these changes more or less randomly cancel out to reveal the real (and lower) trend in average U.S. temperatures.
In contrast, the researchers at the NCDC have spent years combing through U.S. temperature data records trying to figure out ways to adjust for confounders. In 2009, the NCDC researchers detailed how they go about adjusting the temperature data from the 1,218 stations in the Historical Climatology Network (HCN). They look for changes in the time of observation, station moves, instrument changes, and changes in conditions near the station sites (e.g., expanding cities). They filter the data through various algorithms to detect such problems as implausibly high or low temperatures or artifacts produced by lazy observers who just keep marking down the same daily temperatures for long periods.
They've clarified a lot this way. For example, simply shifting from liquid-in-glass thermometers to electronic maximum-minimum temperature systems "led to an average drop in maximum temperatures of about 0.4°C and to an average rise in minimum temperatures of 0.3°C." In addition, observers switched their time of observation afternoon to morning. Both of these changes would tend to artificially cool the U.S. temperature record.
Urban areas are warmer than the countryside, so previous NCDC researchers had to adjust temperature datasets account for the effects of urban growth around weather stations. The center's 2009 study conceded that many HCN stations are not ideally situated—that they now sit near parking lots, say, or building HVAC exhausts. Such effects tend to boost recorded temperatures. The researchers argue that they do not need to make any explicit adjustments for such effects because their algorithms can identify and correct for those errors in the temperature data.
Once all the calculating is done, the 2009 study concludes, the new adjusted data suggests that the "trend in maximum temperature is 0.064°C per decade, and the trend in minimum temperature is 0.075°C per decade" for the continental U.S. since 1895. The NCDC folks never rest in their search for greater precision. This year they recalculated the historical temperatures, this time by adjusting data in each of the 344 climate divisions into which the coterminous U.S. is divvied up. They now report a temperature trend of 0.067°C per decade.
The NCDC have also developed a procedure for infilling missing station data by comparing temperatures reported from the nearby stations. Why? Because as many as 25 percent of the original stations that comprised the HCN are no longer running. Essentially, the researchers create a temperature trend for each missing station by interpolating temperature data from nearby stations that are still operating. Skeptics like Heller argue that that the virtual "zombie stations" that infill missing data have been biased to report higher than actual temperatures.
Some sort of infilling procedure needs to be done. Let's say that there are records from five stations, all of which report time series of 1, 2, 3, 4, and 5. The average of each therefore comes to 3. If two stations fail to report on the second day, missing records of 2, then the average of their remaining four records is now 3.25 instead of 3. In trying to address the problem of missing data from closed stations, the NCDC folks average other stations to fill in the absent 2s. According to climate change skeptic blogger Brandon Shollenberger, what Heller does is the equivalent of averaging the raw data from the notional five stations to report 3, 3, 3, 3.25, and 3.25. "He'd then accuse the people of fraud if they said the right answer was 3, 3, 3, 3, 3," Shollenberger writes.
Let's assume that all of the NCDC's adjustments are correct. What do they reveal? The center's 2009 study concluded, "Overall, the collective effect of changes in observation practice in the U.S. HCN stations is the same order of magnitude as the background climate signal (e.g., artificial bias in maximum temperatures is about -0.04°C per decade compared to the background trend of about 0.06°C per decade). Consequently, bias adjustments are essential in reducing the uncertainty in climate trends." In other words, the asserted bias is almost as big as the asserted trend. Even with the best intentions in the world, how can the NCDC be sure that it has accurately sorted the climate signal from the data noise such that it has in fact reduced the uncertainty in climate trends?
Well, for one thing, other scientists have found a similar trend. Another group of researchers at Berkeley Earth use a different statistical method in which any significant changes to the temperature record of any station are treated as though a new station had been created. They use eight times more data than the NCDC does. Via email, Berkeley Earth researcher Zeke Hausfather notes that Berkeley Earth's breakpoint method finds "U.S. temperature records nearly identical to the NCDC ones (and quite different from the raw data), despite using different methodologies and many more station records with no infilling or dropouts in recent years." He is also quite critical of Heller's simple averaging of raw data.
The NCDC also notes that all the changes to the record have gone through peer review and have been published in reputable journals. The skeptics, in turn, claim that a pro-warming confirmation bias is widespread among orthodox climate scientists, tainting the peer review process. Via email, Anthony Watts—proprietor of Watts Up With That, a website popular with climate change skeptics—tells me that he does not think that NCDC researchers are intentionally distorting the record. But he believes that the researchers have likely succumbed to this confirmation bias in their temperature analyses. In other words, he thinks the NCDC's scientists do not question the results of their adjustment procedures because they report the trend the researches expect to find. Watts wants the center's algorithms, computer coding, temperature records, and so forth to be checked by researchers outside the climate science establishment.
Clearly, replication by independent researchers would add confidence to the NCDC results. In the meantime, if Heller episode proves nothing else, it is that we can continue to expect confirmation bias to pervade nearly every aspect of the climate change debate.