For more information on Berkeley Earth, see www.BerkeleyEarth.org For more information on Novim, see www.Novim.org
I begin by talking about
Prior groups at NOAA, NASA, and in the UK (HadCRU) estimate about a 1.2 degree C land temperature rise from the early 1900s to the present. This 1.2 degree rise is what we call global warming. Their work is excellent, and the Berkeley Earth project strives to build on it.
Human caused global warming is somewhat smaller. According to the most recent IPCC report (2007), the human component became apparent only after 1957, and it
amounts to "most" of the 0.7 degree rise since then. Let's assume the human-caused warming is 0.6 degrees.
The magnitude of this temperature rise is a key scientific and public policy concern. A 0.2 degree uncertainty puts the human component between 0.4 and 0.8 degrees – a factor of two uncertainty. Policy depends on this number. It needs to be improved.
Berkeley Earth is working to improve on the accuracy of this key number by using a more complete set of data, and by looking at biases in a new way.
The project has already merged 1.6 billion land surface temperature measurements from 16 sources, most of them publicly available, and is putting them in a simple format to allow easy use by scientists around the world. By using all the data and new statistical approaches that can handle short records, and by using novel approaches to estimation and avoidance of systematic biases, we expect to improve on the accuracy of the estimate of the Earth's temperature change.
I'll now talk about potential
Bias in Data Selection
Prior groups (NOAA, NASA, HadCRU) selected for their analysis 12% to 22% of the roughly 39,000 available stations. (The number of stations they used varied from 4,500 to a maximum of 8,500.)
They believe their station selection was unbiased. Outside groups have questioned that, and claimed that the selection picked records with large temperature increases. Such bias could be inadvertent, for example, a result of choosing long continuous records. (A long record might mean a station that was once on the outskirts and is now within a city.)
To avoid such station selection bias, Berkeley Earth has developed techniques to work with all the available stations. This requires a technique that can include short and discontinuous records
In an initial test, Berkeley Earth chose stations randomly from the complete set of 39,028 stations. Such a selection is free of station selection bias.
In our preliminary analysis of these stations, we found a warming trend that is shown in the figure. It is very similar to that reported by the prior groups: a rise of about 0.7 degrees C since 1957. (Please keep in mind that the Berkeley Earth curve, in black, does not include adjustments designed to eliminate systematic bias.)
Figure: Land average temperatures from the three major programs, compared with an initial test of the Berkeley Earth dataset and analysis process. Approximately 2 percent of the available sites were chosen randomly from the complete set of 39,028 sites. The Berkeley data are marked as preliminary because they do not include treatments for the reduction of systematic bias.
The Berkeley Earth agreement with the prior analysis surprised us, since our preliminary results don't yet address many of the known biases. When they do, it is possible that the corrections could bring our current agreement into disagreement
Why such close agreement between our uncorrected data and their adjusted data? One possibility is that the systematic corrections applied by the other groups are small. We don't yet know
The main value of our preliminary result is that it demonstrates the Berkeley Earth ability to use all records, including those that are short or fragmented. When we apply our approach to the complete data collection, we will largely eliminate the station selection bias, and significantly reduce statistical uncertainties.