This is a continuation of first part of the UHI analysis with the USHCNv2 temperature record from 1970-2000 using NHGIS economic and industry data. Code and intermediate results can be downloaded for this post here.
In that post, Zeke left a helpful comment mentioning that the apparent UHI signal might instead simply be the bias of comparing stations that switched from CRS to MMTS instrumentation at different times. Using the CRS-only stations here did not leave us with a lot of data to make a definitive statement.
In this analysis, I use the station history file found here to determine the month and year in which a particular station switched from CRS to MMTS, thereby allowing us to calculate the number of months between when one station and its neighbor station changed.
There are two notes I’d like to make before starting:
- The station history file ends in 1994. If there is a history file that extends beyond that, I can’t find it.
- Stations switch-overs may simply not be recorded. Therefore, I cannot simply assume that if a station does not note a switch to MMTS, that it has simply stayed as CRS. Because of this and #1, the program only classifies stations as “NO_SWITCH” IF there is no switch recorded AND the station is listed in the “ushcn-crs-stations.txt” meta file. This means that the NO_DATA category consists of stations that have either made the switch to MMTS after 1994, do not have a year or record for the switch, or have switched to ASOS instead.
Given the above notes, here is a chart showing the number of station switch-overs per year within the network for stations with valid temp data (7 or more reported years for each decade from 1970 to 2000):
And here is the subset of stations that include NHGIS data as well (which we’ll actually be using in our analysis):
As we can see, the bulk of the changes occur within the mid eighties.
Now for the results of our pairwise comparison tests. These take much of the same form as the previous tests, except we now have another threshold as well for determining inclusion: # of months between the switch-overs. A couple more notes regarding assumptions within this exercise:
- I only count the monthly difference up to the point that it will have an effect in our analysis. So since we only go up to 2000, a station which never makes the shift away from CRS will list the switch date as January 2001.
- Some switch-overs have the year listed but not the month. In these cases, I’ve set the month to “6”.
As you can see, the signal appear pretty robust and its existence is not greatly affected by the difference between the months switching to MMTS. I was hoping to do more in this post, and have begun looking into the PHA (as you may have noticed in my last post) to see if I can gauge the magnitude of the UHI effect in the F52 dataset, but have run into difficulties with this project. If anybody has experience re-creating the official F52 dataset by running the PHA, please drop me a line.