# Is it right practice to validate temperature data with reanalysis data?

For a project I am working on I received raw data from many meteorology stationes.
Is it right practice to test this data against air temperature from noaa's ncep/ncar data?

• What do you mean "test"? I think it can be useful to compare these sources as a sanity check.
– Abe
Jun 17, 2014 at 2:32

No. One should not attempt to validate raw in situ measurements with reanalysis data. Reanalysis data contain partial information from observations, but are mostly model fields and are provided at very coarse resolution (~1$^\circ$ longitude). In situ measurements are subject to very local, small scale effects, and have different meaning from reanalysis. One is always free to compare the two datasets, but this would not be validation in any meaningful way.

Many meteorological stations are assimilated into NOAA's reanalysis. You should be careful with this approach as the raw data you have might have been already included in the hindcast.

The reanalysis includes the data in an statistical way. In general, the reanalysis does not match the data at any point, but tries to minimize the global difference between observations and model simulation. There is no guarantee that the reanalysis is going to be close to any of the assimilated data.

• Thanks, if this stations and many others are assimilated in to the reanalysis does it not mean it will be easy to find outlires and bad data? Jun 10, 2014 at 17:59
• The reanalysis algorithm takes care of outliers and bad data.
– gerrit
Jun 10, 2014 at 18:02
• thats my point! so if i have raw data from one single station is it wrong to "clean" the data against the reanalysis Jun 10, 2014 at 18:41
• I think the point that aretxabaleta is making is that if the station that you have data from was included in the reanalysis, then finding that the two sets of data match well does not necessarily tell you much about the station's data quality - because to some degree, you have compared the station's measurements with themselves. Jun 11, 2014 at 8:42
• @SimonW: That was exactly my point. Thanks for the clarification. Jun 11, 2014 at 17:22

Official meteorological stations should have well-calibrated thermometers. In any comparison between models and measurements, the measurements would be the reference, not the models.

One way I can see to validate ground-based in-situ temperature measurements is by carrying an SI-traceable thermometer to all the sites. Unfortunately, you cannot do this retroactively. I don't know to what degree thermometers at official meteorological stations are SI-traceable.

Another way would be to look at spatio-temporal series, and search for outliers that way. This would not be a comparison between different independent methods, but would reject a particular measurement if it is too far off compared to measurements nearby in space and time. To find the right criteria to reject data is a delicate balance, and you will certainly end up either rejecting good data or accepting bad data.