The scientific basis for the humidity downscaling is outlined in the following publication:
Pierce, D. W. and D. R. Cayan, 2015: Downscaling humidity with Localized Constructed Analogs (LOCA) over the conterminous United States. Climate Dynamics, DOI 10.1007/s00382-015-2845-1. Author’s copy of the paper.
What follows are notes on the particular downscaling process used to produce the available LOCA humidity data.
Not all CMIP5 global climate models saved daily near surface humidity, which is needed if the model results are to be downscaled with LOCA. The near-surface humidity variable most often saved at a daily frequency was specific humidity (“huss”, in CMIP5 parlance). A total of twenty four global models saved daily huss (compared to 32 models used for the temperature and precipitation LOCA downscaling).
In talking with our stakeholders, we found that most of them did not want specific humidity, but rather daily minimum and maximum relative humidity. Minimum humidity is important to fire weather and agricultural impacts, while maximum humidity affects human health (often described using the heat index). Since daily near surface relative humidity (RH) was rarely saved in the models, directly downscaling RH would have sharply limited the number of models we could provide data for. So, we decided to instead calculate downscaled RH from the downscaled specific humidity.
To calculate RH from specific humidity requires temperature and pressure as well as the specific humidity, which were obtained as follows.
Specific Humidity: Calculated using the full LOCA downscaling process applied to daily huss, the original GCM near surface specific humidity fields. For the training data set we used the daily specific humidity dataset from John Abatzoglou. The training data are available here.
Temperature: We used the existing LOCA-downscaled daily maximum temperature (Tmax) to calculate the daily minimum RH, and LOCA-downscaled daily minimum temperature (Tmin) to calculate the daily maximum RH.
Pressure: The RH calculation requires ambient (surface) pressure, but many models only saved sea level pressure (SLP) on a daily timescale. We therefore estimated surface pressure from SLP and temperature. To obtain surface pressure, the original model SLP fields were first bias corrected using daily North American Regional Reanalysis (NARR) data over the period 1979-2015 (37 years). The bias corrected results were then bilinearly interpolated to the 16th degree LOCA grid cells. Interpolation was used, rather then a full LOCA downscaling of SLP, because SLP generally does not have fine-scale spatial structure. The bias-corrected, interpolated SLP was then used along with the already existing LOCA-downscaled temperature and the LOCA elevation field to produce surface pressure.
Once temperature and ambient (surface) pressure were available, it was straightforward to calculate RH. For the training data sets (daily minimum and maximum RH), we used the daily RH min and max datasets from John Abatzoglou. The training data are available here.
This overall procedure is rather different from that used to create Tmin, Tmax, and Precipitation in the original LOCA data set, in that several additional variables are used in the calculation rather than just the field being downscaled. (Which is to say, downscaling Tmax requires only Tmax, but downscaling RH max requires specific humidity, Tmin, and surface pressure.) Furthermore, the final RH min and max fields were bias corrected to the Abatzoglou data, and so the final available LOCA-downscaled RH values on any particular day are not identical to the downscaled specific humidity values combined with the downscaled temperature, SLP, and elevation to produce a relative humidity value.
There are two main reasons for the difference between the available final downscaled RH values and the RH values that would be calculated with downscaled specific humidity, temperature, and SLP. One is that the RH min and max values in the training (Abatzoglou) dataset themselves are not directly calculable from the Abatzoglou specific humidity, Tmin, and Tmax because sub-daily information, not retained in the daily fields, was used to produce RH min and max (Abatzoglou, pers. comm.; see also page 123 of Abatzoglou, J., 2013, Int. J. Climatology). The second reason, which impacts RH max, is that the LOCA Tmin data were produced (trained with) Tmin data from Livneh et al. 2015, which has a non-trivial bias with respect to the Abatzoglou Tmin data.