by Steve Mosher
It has been a while since we’ve done an update and there is much to report on, including an update to the web site, some additional memos/papers to discuss and an update on the papers. Let’s start with the web site.
- Updated data and code drop. The process for automatically updating our temperature record has progressed somewhat. Since we rely on 14 different datasets that each have different updating processes our updates are not to a point where new numbers can be produced on a monthly basis but that is the goal. The code and documentation has been improved so that dedicated end users can download it and get it running without too much outside help. Still it is not a beginner project. Over the course of the next few months we will be working with researchers who have expressed interest.
- Gridded data has been posted. The gridded data is in 1 degree grids and equal area grids.
- State and Province data. Since we create a temperature field we can use Shape files to extract the average temperature field for irregular areas. Data for states and provinces of the largest countries is provided. Theoretically, one could specify any arbitrary polygon and extract that from the field which may be of use for certain applications such as reconstructions.
There are four new memos that we are posting for people to comment on. Two of the memos relate to Hansens PNAS paper (H2012) on extremes. Hansen’s PNAS paper was read by some to be an argument for a more variable climate. Here is an example of how some people understood Hansen’s paper. Both of these memos have been reviewed and improved by Hansen and Ruedy so we thank them for their contributions. The 1st memo was written by Sebastian Wickenburg and the 2nd memo was written by Zeke Hausfather.
The the PNAS paper does not establish “if you put more energy into a system variability increases.” This is shown in two ways. In Wickenburg, we show that the widening of the distribution can be a mere methodological artifact. Hausfather makes the same point and illustrates a different methodology that challenges assertions of increased variability. The primary insight of the H2012 remains, in a warming climate we expect to see more warm extremes. However, H2012 did not establish or aim at establishing that the distribution of temperatures has widened. Showing a change in distribution probably requires different statistical tests than those that were applied.
The 3rd and 4th memos are extensions of our Methods paper. The 3rd memo is a simple exercise to help people visualize the difference between the Berkeley method, the CRU method and the GISS method. To illustrate the difference we use visual data rather than temperature data.
The 4th memo stems from reviews of the method paper. A reviewer of the methods paper requested that we use GCM data to establish that our method was superior to CRU’s method. As the methods paper was already rather long, we decided to write up a separate memo focused on this test. The approach is straightforward. A 1000 year GCM simulation is used as ground truth. Since this data exists for every place and time we can calculate the “true” average at any given time. This “ground truth” is then sampled by using the GHCN locations as a filter. The experiment is repeated using sub samples of the 1000 year run. The results show that if you use a limited spatial sample ( GHCN locations ) with temporal gaps ( not every station is complete ) that the Berkeley method has the lowest prediction error. This should come as no surprise. As far as I know this is the first time any rigorous pure methodological test has been performed on CRU or GISS and is one of the benefits of having code posted for the various methods.
The results paper has been published by Geostatistics and Geoinformatics, available online. The Methods paper and UHI paper are under going final review prior to being submitted.
Hausfather, Rhodes and Menne collaborated on an AGU poster comparing Berkeley “scalpel” technique to NOAA homogenization
JC moderation note: Comments on this thread will be moderated for relevance