This is a follow-up on my last post, which found what appears to be a fairly robust UHI signal in the TOB dataset from USHCNv2. The same thing cannot be found in the F52 dataset. Here I look into a few things to see whether the signal in the TOB dataset may be the result of bias from changes to station location and instrumentation.
The first test here, as Zeke suggested, is to look at CRS stations only, since those likely did not undergo any instrument change during our time period. I did this by simply subbing in the ushcn-crs-stations.txt as my stations file from USHCNv2.
Ultimately, this gave me 62 stations with both valid temperature data (>=7 years per decade) and NHGIS data for my two variables. The results of method #2 (see last post for more info) were [Income Coefficient=7.752, Agriculture Coefficient= -2.235, Correl= 0.284], although by only using CRS-temperature data the dtAdj is probably not very effective because of limited data per grid cell. Here are the results of the pairwise comparison:
On the one hand, it doesn’t look like we get all that much different of an effect. But there is much less data available, and the coefficient for income drops off significantly at that 400 km threshold.
For MMTS stations, we have 296 stations available. Results of method #2 are: [Income Coefficient=3.548, Agriculture Coefficient= -0.767, Correl= 0.14] . And our pairwise comparison test:
It may be possible that the results are not simply the result of bias due to station changes, if we consider the breakdown of the stations that actually contain NHGIS data. I use the station inventory with current population data appended produced by Ron Broberg here to help create the graph below.
Note that in the subset of stations with NHGIS data available, there is a much smaller percentage of stations that might be categorized as “rural”. Since this class of stations tends to correlate with station changes, we have fewer involved that might require these type of corrections from TOB to F52.
Another thing to look at is the differing trends of just the CRS stations between those that only have temperature data vs. those that we use in our tests because they have NHGIS data as well:
The last test I attempted was to see which stations had their trends change the least between TOB and F52. I then included only stations whose trend changed by less than 0.1 degree C per decade in my analysis.
The result left me with only 114 stations that had both NHGIS and valid temperature data. Here are the results of that comparison:
Here, we note a significantly smaller effect based on the income coefficient. However, one thing worth noting is the difference in trends according to my temp anomaly calculator when only using these 114 stations.
The annual trend for TOB calculated in the last post was 0.0276. With only those 114 stations, it is 0.0275 here. So minimal change. On the other hand, F52 using all our stations with valid data yields an 0.031 trend, whereas only those stations with minimal change from TOB yields only an 0.028 trend.
There still quite a bit to investigate…