My hat is off to scientists who have to work in research where accurate data must be teased out of a boat load of data that is inaccurate. It’s what I jokingly call separating the fly poop out of the pepper. They both look the same but one of them you wouldn’t want to use in your cooking.
I was reading an abstract written by Eric J. Steig and published in the Nature Journal of Science titled “Warming of the Antarctic ice-sheet surface since the 1957 International Geophysical Year” wherein Steig makes the claim that the entire Antarctic has been warming since 1957. What caught my attention is global warming alarmists have been citing this study quite passionately while quite a number of other highly respected studies and data sources, including highly accurate satellite measurements of ice extents and temperatures indicate that the Antarctic has been cooling for a number decades. How can one data set say the Antarctic is cooling and Steig’s data set says it is warming?
It is all in the quality and quantity of the data. What many don’t realize is that most, if not all of the data used in climate analysis is normalized or reconstructed data. Put another way, the sensors that provide the data experience many technical problems including data dropouts, data transmission errors, sensor drift, calibration errors, low SNR (signal to noise ratio) events, and a host of other data damaging problems.
Climate scientists use an advanced form of standard deviation analysis of data called Regularized Expectation Maximization or RegEM to fill in missing or errored data. Expectation Maximization (EM) figures out the normal range and trend of data and fills in the missing blanks where holes appear. However it has its limits. When too many holes appear in the data EM fails. Regularization must then be used to recreate the missing data. The problem is regularization can often introduce bias into the data. RegEM is a statistically dangerous form of data infilling.
Given the non-linearity of the inputs to climate models, a small amount of bias at the input can result in a large error at the output. The nature of data perturbations in Automatic Weather Station (AWS) and satellite based measurements are such that they seem to always bias the data towards underestimating negative forcings or overestimate positive forcings in the models. The result is the model spits out results that bias towards a hotter climate. The farther in time the model forecasts the greater the error in its prediction. Put in layman’s terms, when errors are made, things turn hot!
But Steig’s models aren’t forecasting into the future, they’re modeling the past through to the present. How does the above explanation fit? It’s simple really, if your data sources are relatively few and the data contains a high number of holes and errors, viola - we have bias.
In studying Steig’s full paper, it is evident that he is using only a very small handful of ground station measurements providing data prior to 1980. These stations had numerous problems with the accuracy of their data. Most were positioned in coastal locations with only one or two (mostly non-reporting) positioned in the interior of the Antarctic. Statistically, they provided a very small data set that was full of holes with almost no data for the Antarctic inland. Given that the Antarctic is roughly the size of the United States, this is like taking a few samples from Key West, San Diego, and Portland and building a model from the data to represent the climate for the entire United States! The model couldn't possibly account for much colder temperatures in inland North Dakota.
You may now be getting the picture - erratic and poorly representative data from ground stations prior to 1980 and no data from satellites prior to 1982. Using RegEM, Steig has to reconstruct (fill in) most of the missing AWS data prior to 1980 and all of the missing satellite data prior to 1982. That’s 25 years worth of data that had to be made up! What I find most interesting is if you remove the RegEM infilled years prior to 1982 from Steig's model and use his better correlated data set from 1982 forward, the data shows no relative warming whatsoever. The first 25 years of largely made up data skewed the study to show warming when there wasn't any! Personally, I can't imagine how Nature Journal could have possibly peer reviewed Steig's abstract prior to publishing it with such glaring statistical problems.
Now enter the Anthropogenic Global Warming (AGW - aka. Al Gore Warming) folks who compare Steig's data to the equally (and identically) flawed current global climate models and we now have yet another scientific study that proves global warming is worse than ever imagined.
I’ve tried to avoid getting too technical but felt it necessary to show how data errors and poor methodologies can creep into a study and bias it to produce global warming when, in fact, the Antarctic has been well documented to be cooling all these years. But this small fact won’t stop global warming alarmists from jumping all over Steig’s study and crying “the ice is melting... the Polar Bears are falling off their ice sheets and drowning... global warming crisis is upon us!”