[NARCCAP-discuss] Another NARCCAP surface moisture problem?

[Brimelow, Julian (EC/EC)]

Hello,

Seth McGinnis suggested that I email my message I recently sent to him to this address.

Best regards,
Julian


Julian Brimelow, PhD
Physical Sciences Specialist
Applied Environmental Prediction Science - Prairie and Northern
Meteorological Service of Canada (MSC)
Environment and Climate Change Canada (ECCC)
Eastgate Offices
9250 - 49th Street
Edmonton, AB, T6B 1
Julian.Brimelow at canada.ca
Telephone 780-951-8986
Facsimile 780-495-3529
Government of Canada
Website www.ec.gc.ca<https://www.ec.gc.ca>

Julian Brimelow, PhD
Spécialiste en sciences physiques
Sciences appliquées de la prévision environnementale des Prairies et du Nord
Service météorologique du Canada (SMC)
Environnement et Changement climatique Canada (ECCC)
Eastgate Offices
9250 - 49th Street
Edmonton, AB, T6B 1K5
Julian.Brimelow at canada.ca<mailto:Julian.Brimelow at canada.ca>
Téléphone  780-951-8986
Télécopieur  780-495-3529
Gouvernement du Canada
Site Web www.ec.gc.ca<https://www.ec.gc.ca>



From: Brimelow, Julian (EC/EC)
Sent: December 8, 2015 12:02 PM
To: 'Seth McGinnis'
Cc: 'bukovsky at ucar.edu'
Subject: Another NARCCAP surface moisture problem

Hello Seth and Melissa,

Hope that this email finds you both well.  We were almost complete with the data analysis for our paper, when I came across upon something disconcerting with the 2-m specific humidity (q) in the HadCM3-MM5I profiles.

The issue is that over those regions of the NARCCAP domain where surface moisture is typically high, there are times and locations during the warm season where the 2-m specific humidity is anomalously high (I’ve seen q values of 40 g/kg!). When one uses these anomalous values to calculate the 2-m dewpoint, it exceeds the temperature, sometimes by as much as 10 C. This is obviously problematic and unphysical. The values at the first data point above the surface decrease dramatically and return to what can be considered realistic values.

We have been using the mean temperature and dewpoint in the lowest 50-mb of the profiles to lift parcels for running HAILCAST and calculating CAPE etc., so we did not spot anything obvious because mixed-layer Td is usually lower than the mixed-layer T (because of the huge drop-off in q above the surface in such instances).  Regardless, we now need to go back and identify all problem profiles and exclude them from our dataset.

The attached graphic (sorry for the quality, it is a quick screen grab) shows the problem—see the centre image in the top row. The bottom row shows some examples of the q hot spots (or speckles) on other days.  I’ve circled the outliers. There seems to be no rhyme or reason to this, other than random “speckling” of surface q. I did also identify a handful of instances in the CCSM-MM5I profiles with the same issue, but this is probably nothing to be concerned about given that it occurs extremely infrequently. The problem is much more pervasive in HadCM3-MM5I over the northeastern and southeastern Portions of the NARCCAP domain (e.g., Appalachians, Deep South, Atlantic and Great Lakes) where over 10% of the hail profiles had this issue.

The good news is that the bug appears to be transient and typically only affects one time step on a given day (we have been looking at profiles for 18, 21 00 and 03 UTC), so we can usually find another sounding for a different time at that location that is still representative of the convective conditions.

The problems with the surface moisture in the CRCM and HadCM3-MM5I data underscore the importance of also considering the performance of low-level moisture when evaluating the robustness/skill of the RCM-AOGCM pairings. One sometimes has to look very closely at the data to identify such issues. Far too often, in my opinion, the focus is only on T and precip. when evaluating the skill of the model output, this is especially problematic during the warm season when convection and the apparent temperature can be important and are very sensitive to biases in the low-level moisture. Right now, we still do not have a quantitative measure as to which of the NARCCAP RCM-AOGCM pairings show the best skill at simulating the low-level moisture during the warm season, what the regional and seasonal biases are, or how biases in temperature and surface moisture interact to either offset or exacerbate precipitation biases, for example.

Melissa, I really (!)  like the concept behind Table 6 in your recent J. Climate paper ☺.  Pity that this has only been done for the NAM region.  Thibeault and Seth (2015, JGR-A) adopted a similar approach for the northeastern USA, but they did not include low-level moisture in their list of metrics that were evaluated (see their Table 6). To date the most comprehensive evaluation of the NARCCAP model output  was undertaken by Elguindi and Grundstein (2013, Climate Change)— unfortunately their paper did not include all of the model pairings. We have used the Elguindi and Grundstein paper to identify a subset of models they considered that best simulate the current climate.

Anyhow, I just wanted to let you know about the issue we identified with the HadCM3-MM5I specific humidity, and hopefully generate some discussion about the prospect of future studies including more diverse/comprehensive verification of some of the key variables over the entire model domain.

All the best,
Julian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: huss_hadcm-mm5_speckle.png
Type: image/png
Size: 335596 bytes
Desc: huss_hadcm-mm5_speckle.png
Url : http://mailman.ucar.edu/pipermail/narccap-discuss/attachments/20151209/e8e62da3/attachment-0001.png