Relative humidity from ERA5

[rmanzanas]

For ERA5, the daily relative and specific humidity (hurs and huss, respectively) data that are stored at /lustre/gmeteo/WORK/DATA/C3S-CDS/ERA5/day have been derived from other variables. In particular, hurs was calculated from daily dewpoint temperature and 2m-temperature using the tdps2hurs function from convertR (version 0.2.0) in May 2023. Likewise, huss was computed from daily hurs, pressure and 2m-temperature using the hurs2huss function from convertR (version 0.2.0) in May 2023.

This document (April 2025) shows that the approach followed for deriving hurs may lead to inaccurate results, especially for dry and semi-arid regions. However, this limitation can be improved by relying on a more sophisticated method that is currently used by ECMWF and NOAA and has been implemented in the devel branch of convertR (method = "advanced" in tdps2hurs).

Based on all this, it might be worth to:

1. directly download huss from the CDS
2. re-calculate hurs, globally, using the new method

[chantreux]

Ok so I had already downloaded:

• daily mean temperature and dew point from https://cds.climate.copernicus.eu/datasets/derived-era5-single-levels-daily-statistics?tab=overview. Is that okey for the calculations?

About the calculation:

• We then can see how to calculate hurs, I can try to reproduce the function in Python or someone can use convertR with the advanced method

For huss:

• there is no specific humidity in ERA5 single levels: https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=overview. Is there any other place to find it?

[anacv]

@chantreux to calculate hurs from dew point temperature, you can use: https://github.com/ecmwf/thermofeel/blob/master/thermofeel/thermofeel.py#L49 This is the same implementation as in the new implementation in convertR.

[chantreux]

I'm calculating the hurs using the function in : https://github.com/ecmwf/thermofeel/blob/master/thermofeel/thermofeel.py#L49 implemented in this commit (rh_from_thermofeel()) :

https://github.com/SantanderMetGroup/c3s-cds/blob/c4007434cba5cdcd1d25e0788bf36fcafd52ee97/scripts/derived/operations.py

The hurs output is in : /lustre/gmeteo/WORK/DATA/C3S-CDS/ERA5_temp/derived/derived-era5-single-levels-daily-statistics/daily/native/hurs/
How can I validate that the result is correct? Do we have the data in the document somewhere to reproduce the figures?

[anacv]

I sent you some data to do some checks.
Please note that, when this is consolidated, other datasets that relied on this need to be updated (e.g. there is a ERA5 subset in the PTI folder).

[rmanzanas]

@chantreux , it would be very useful to download surface pressure in order to derive huss from the new hurs.

[chantreux]

@chantreux , it would be very useful to download surface pressure in order to derive huss from the new hurs.

Ok! I checked and I had already downloaded it: https://github.com/SantanderMetGroup/c3s-cds/tree/main/catalogues#derived-era5-single-levels-daily-statistics-catalogue

Here are all the variables and dataset downloaded or in progess:
https://github.com/SantanderMetGroup/c3s-cds/blob/main/catalogues/catalogues/all_catalogues.csv

[chantreux]

What I did:

• Calculated hurs from ERA5 using thermofeel (from downloaded daily mean 2m temperature and daily mean dew point temperature) and compared it to the Carmen climate4R workflow data.
• Compared time series at two sample locations (Tanger and Sahara) for 2020.
• Compared spatial means (2015–2020) and plotted the difference map (Carmen - thermofeel).

Results:
Time series from both implementations overlap very closely; the difference trace is centered near zero and has very small amplitude. Spatial mean fields are visually almost identical. The difference map shows only very small regional biases (order of ~0.5–1% in a few areas). Very small differences exist; these can be caused by implementation details (thermofeel vs climate4R), numeric/rounding differences, or differences in the original datasets used by each workflow.

Conclusion:

The thermofeel implementation of hurs seems validated against the Carmen/climate4R implementation: no major discrepancies were found.

Plots:

2020 Timeseries of point centered in Sahara (dry):

2020 Timeseries of point centered in Tanger (humid):

2015-2020 Climatology comparing Carmen and thermofeel hurs:

hurs_thermofeel_2015-2020.pdf

[anacv]

Thanks @chantreux and @crodriguezrumayor. I think this looks good, I guess differences may come from the different workflows and data pre-processing.

@chantreux : after the calculation, make sure that any value below 0 is set to 0 and above 100 is set to 100. This is not done internally in thermofeel and we added it in climate4R (this is not the reason of these differences, since there are not such extreme values, but they can be found, in practice, in other locations or periods).

[chantreux]

When using the same input data we have a near exact match between both methods now (I incorporated a clip in values between 0 and 100):

2020 Timeseries of point centered in Sahara (dry):

2020 Timeseries of point centered in Tanger (humid):

2015-2020 Climatology comparing Carmen and thermofeel hurs:

[anacv]

Perfect, this looks very good now, thanks @chantreux