Caveats about Regional Climate Model outputs

Sea Ice in RegCM4 simulations

The RegCM4 regional model is unable to model sea ice. It does not contain a frozen land cover type over water in winter, nor any code describing how frozen seawater behaves. In the GCM-driven simulations, it uses the skin temperature from the atmospheric component of the GCM as the sea surface temperature (SST). This gives it a temperature that is more reflective of what the air temperature over the water would be if it were frozen, even though the water is open in the model.

We don't use the actual SST from the ocean model in the GCM because it is pegged to the freezing temperature of salt water any time there is ice, and that would lead to warm open ocean water in RegCM in the winter, and that would be sub-optimal.

However, in the simulations driven by the ERA-Int reanalysis, that was the only available option at the time. As a result, there is a notable warm bias, particularly in winter, over the oceans and nearby land in high-latitude regions in these simulations.

Calendar in HadGEM2-ES-driven WRF simulations

The HadGEM GCM uses a 360-day calendar. WRF did not have an option for that calendar when these simulations were started. Thus, for these simulations, the decision was made to modify the input data from HadGEM to match a standard calendar. To turn a 30-day month from HadGEM into a standard month, days at the end of the month were duplicated or deleted until the month had the standard number of days. This is not the case for the HadGEM-driven RegCM4 simulations, as RegCM4 handles this unusual calendar.

Interpolation Artifacts

We used the ESMF "patch recovery" algorithm to iterpolate the data from the RCM native grids onto the NAM-22i and NAM-44i common lat-lon grids. This can result in very small negative values for variables that are normally non-negative, such as specific and relative humidity (huss, hurs), incoming solar radiation (rsds), and precipitation (pr, prec).

Bias-Corrected Relative Humidity

For convenience, the archive includes some variables that are not output by the models but can be calculated from variables that are, such as wind speed (sfcWind) calculated form eastward- and northward winds (uas and vas). We calculate relative humidity from specific humidity (huss), temperature (tas), and surface pressure (ps) using the NCL function relhum. We also calculate relative humidity for the bias-corrected versions of the data because it is valuable in impacts work, but some approximations are required.

We do not bias-correct surface pressure, because there are no gridded daily observational datasets that include surface pressure. Instead, we use a static value for surface pressure based on elevation (taken from the NASA SRTM DEM, used by Daymet) and the barometric formula. This is consistent with the approach taken by the gridMET observational dataset. We use the same static pressure to convert vapor pressure to specific humidity in the Daymet observations. The effect of using static pressure instead of time-varying pressure in calculating relative humidity is expected to be very small. Static pressure files can be found here: NAM-22i static surface pressure | NAM-44i static surface pressure

For raw model output, we use daily average temperature (tas) to calculate relative humidity. However, the observational datasets don't include daily average temperature, only daily minimum and maximum temperature, so we don't have bias-corrected average temperature. To compute relative humidity, we calculate tmean, which is the average of the bias-corrected daily minimum and maximum temperatures. Although this value can in theory vary considerably from a proper time-average, the approximation has a long history of use in working with station data, and in fact many of the daily average values in the observational data products we use in bias correction are likely also tmean instead of tas.