Adding value to extended-range forecasts in northern Europe by statistical post-processing using stratospheric observations

Korhonen, Natalia; Hyvärinen, Otto; Kämäräinen, Matti; Richardson, David S.; Järvinen, Heikki; Gregow, Hilppa

doi:https://doi.org/10.5194/acp-20-8441-2020

Articles | Volume 20, issue 14

https://doi.org/10.5194/acp-20-8441-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/acp-20-8441-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 20, issue 14

Research article

|

20 Jul 2020

Research article |

| 20 Jul 2020

Adding value to extended-range forecasts in northern Europe by statistical post-processing using stratospheric observations

Natalia Korhonen, Otto Hyvärinen, Matti Kämäräinen, David S. Richardson, Heikki Järvinen, and Hilppa Gregow

Download

Final revised paper (published on 20 Jul 2020)
Preprint (discussion started on 19 Aug 2019)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Comments on the improvement in forecast skill', Anonymous Referee #1, 16 Sep 2019
RC2: 'Review comments', Anonymous Referee #2, 16 Sep 2019
RC3: 'Review comments', Anonymous Referee #3, 19 Sep 2019
AC1: 'Response to anonymous referees #1, #2, and #3 attached as supplement.', Natalia Korhonen, 25 Nov 2019

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Natalia Korhonen on behalf of the Authors (25 Nov 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (29 Nov 2019) by Peter Haynes

RR by Anonymous Referee #2 (13 Dec 2019)

Suggestions for revision or reasons for rejection

Overall the authors have addressed my comments and I think the analysis has been strengthened by incorporating all the weeks in the analysis rather than just the first week of the month. I do have a few remaining minor concerns outlined below.

Minor comments
-I still don’t find it that convincing that the QBO information adds anything. Could you compare in Figure 6, for example, what the skill is if SWIneg is only based on the ZMZW factor?

-Page 2, Line 27-I do not agree with the new description added of the AO as a “climate pattern characterized by winds circulating counterclockwise around the Arctic at around 55N latitude”. The AO characterizes fluctuations in atmospheric mass between the pole and mid-latitudes, and explains latitudinal shifts in the eddy-driven jet near 40N (see Thompson and Wallace 2000). The AO is not another word for “jet stream” or “polar front jet”. The authors should also include a brief description of how the AO from the NCEP CPC is calculated in the methodology section.

-Page 1, line 28: “lead time up to 46 days”- why 46 days instead of 42 days, which would be 6 weeks? Is this a general definition for extended-range?

-Page 3, line 9: cite Garfinkel et al. 2012

-Page 3, line 15: Is this comment on the skill of the QBO forecasts true just for ECMWF or for all ERF systems? It’s unclear.

-Page 5, Line 9: why 5 weeks? Doesn’t the analysis go to week 6?

-Throughout: AO “indexes” should either just be “index” or “indices”

-Page 6, Line 3: here I think “month” should be “week”, according to the new analysis of using all weeks, correct? Similarly, page 6, line 9-10 again refers to the “beginning of each winter month”, as does page 6, lines 14-15 and page 6, line 22. Also see caption for Figure 4.

Hide

RR by Anonymous Referee #3 (19 Dec 2019)

Suggestions for revision or reasons for rejection

I have now read the revised paper and response to reviewers. Although most of my comments have
been satisfactorily dealt with, I don't feel that my main reservation about the analysis has been
adequately addressed.

Major comment

Given that the QBO is commonly thought to influence AO via the stratospheric polar vortex, I remain
sceptical that the QBO provides additional skill in forecasting the AO index over the skill obtained
from using the polar vortex strength alone (which the authors still don't seem to have tried).

I therefore suggest the authors try using the partition based on polar vortex strength (ZMZW index)
only to forecast the AO. Unless they can demonstrate a significant improvement in forecast skill
from also including the QBO in their SWI forecast model, the QBO should not be used. Why include an
additional predictor if it isn't needed?

The authors agree that in figure 2, the partitions based on the ZMZW index (yellow and red boxes)
have a greater difference in the median values than the authors' preferred SWI index (green and
purple boxes). The p-values in both cases are now less than 0.01.

I don't really follow the authors' comment that the larger partition size for their SWI_neg index (199
members, green box in figure 2) compared to the partition based on low ZMZW (70 members, yellow box
in figure 2) gives 'more certainty' in post-processing. Surely the real test of whether to include
the QBO in the AO forecast model is whether it improves the forecast skill score?

Even if the relative lack of members in the low ZMZW partition is a problem (which isn't immediately
obvious to me) more lenient threshold values of ZMZW could be tried to rectify this (as I suggested
previously).

Minor comments

Regarding my previous major comment #2, I think it would be worth noting in the paper that looking
at winters with a strong polar vortex has been tried, even though it didn't improve the AO forecast
skill.

Regarding my minor comment on the AO index - what I wanted included in the paper was an explanation of how the AO index is calculated i.e. what procedure is applied to the 1000hPa geopotential height field
to produce the AO index?

Hide

ED: Reconsider after major revisions (22 Jan 2020) by Peter Haynes

AR by Anna Mirena Feist-Polner on behalf of the Authors (06 Mar 2020) Author's response

ED: Referee Nomination & Report Request started (19 Mar 2020) by Peter Haynes

RR by Anonymous Referee #3 (16 Apr 2020)

Suggestions for revision or reasons for rejection

Having read the latest revision of the paper, I still don't feel that my main reservation about the
analysis has been adequately addressed, and I therefore do not feel the paper is suitable for
publication in its current form. In my previous reviews, I expressed scepticism about the use of
the QBO as a predictor of the Arctic Oscillation (AO). It seemed more plausible to me that the
polar vortex strength was a stronger predictor of the AO, and that the authors should more fully
analyse this before focusing on the QBO.

In the latest draft the authors have revised their methods, but I now find the latest analysis
rather confusing and hard to follow. If I understand correctly, the authors now use the polar
vortex strength (ZMZW) as a predictor of the AO if the vortex is weak (ZMZW below its 10th
percentile) or strong (ZMZW above its 80th percentile) and demonstrate that the AO and temperature
are influenced by the polar vortex strength. For post-processing of temperature forecasts, however,
they only consider cases where ZMZW is between the 10th and 80th percentiles, and use the QBO to
categorise different forecast ensemble members, with a selection of different QBO wind thresholds
tried. This choice of method appears arbitrary and is not adequately justified in the paper. It is
hard to see a chain of reasoning that led to the final method, which seems rather complicated. Why
restrict the post-processing to those cases where the ZMZW is between the 10th and 80th percentiles,
when you've just shown that a weak or strong polar vortex has an effect on the AO and temperature?

The authors argue that the QBO is important, since they have left out the weak and strong ZMZW cases
from the post-processing, and still find improved forecast skill when using the QBO only to define
their SWI_neg and SWI_plain categories. But this misses the point, in my opinion - if the QBO is
influencing the AO via the polar vortex, then ZMZW is the only predictor needed. The SWI categories
should therefore first be defined using ZMZW only, and the forecast skill calculated. Including the
QBO in the SWI categories' definition is only then justified if it leads to a significant
improvement in skill compared to using ZMZW only.

Minor comments

p6 line 15: Where did the choice of QBO wind threshold values come from? (7, 10, 13 ms-1)

Figure 2 : the caption states that the p-values relate to pairs of distributions, but (unlike
figure 3) it's not clear what these pairs are. Also, why are there only 57 members above the 80th
percentile, but 149 members below the 20th percentile?

Figure 5,6 : The numerical values in the colour bar don't correspond to the boundaries between
colours.

Hide

ED: Reconsider after major revisions (16 Apr 2020) by Peter Haynes

AR by Svenja Lange on behalf of the Authors (04 Jun 2020) Author's response

ED: Publish as is (20 Jun 2020) by Peter Haynes

Short summary

Reanalysis data of the strength of the polar vortex is applied in the post-processing of the European Centre for Medium-Range Weather Forecasts (ECMWF) winter surface temperature forecasts for weeks 3–4 and 5–6 over northern Europe. In this way, the skill scores of these forecasts are slightly improved. It is also found that, in cases where the polar vortex was weak at the start of the forecast, the mean skill scores of these forecasts were higher than average.