A 10&thinsp;km North American precipitation and land-surface  reanalysis based on the GEM atmospheric model

Gasset, Nicolas; Fortin, Vincent; Dimitrijevic, Milena; Carrera, Marco; Bilodeau, Bernard; Muncaster, Ryan; Gaborit, Étienne; Roy, Guy; Pentcheva, Nedka; Bulat, Maxim; Wang, Xihong; Pavlovic, Radenko; Lespinas, Franck; Khedhaouiria, Dikra; Mai, Juliane

doi:https://doi.org/10.5194/hess-25-4917-2021

Articles | Volume 25, issue 9

https://doi.org/10.5194/hess-25-4917-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-25-4917-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 25, issue 9

Research article

|

07 Sep 2021

Research article |

| 07 Sep 2021

A 10 km North American precipitation and land-surface reanalysis based on the GEM atmospheric model

Nicolas Gasset, Vincent Fortin, Milena Dimitrijevic, Marco Carrera, Bernard Bilodeau, Ryan Muncaster, Étienne Gaborit, Guy Roy, Nedka Pentcheva, Maxim Bulat, Xihong Wang, Radenko Pavlovic, Franck Lespinas, Dikra Khedhaouiria, and Juliane Mai

Download

Final revised paper (published on 07 Sep 2021)
Preprint (discussion started on 09 Mar 2021)

Interactive discussion

Status: closed

RC1:
'Comment on hess-2021-41', Anonymous Referee #1, 20 Apr 2021

General comments:

The paper introduces a new land surface and precipitation reanalysis for North America. The authors explain in detail different components of the model systems and how they interact with each other. Afterwards, samples of up to 7 years are examined and compared to other available datasets.

Overall the paper is well structured and very well written. I consider it relevant at least for the hydrological community and recommend to publish it following minor revisions.

My major comment is about the choice of the final setup. It was decided to produce the entire period with RDRS-10 (the Regional Deterministic Reforecast System with a horizontal resolution of 10km). However, throughput the paper no explanation is given what this choice is based on. In contrast, at various places it is noticed that RDRS-15 performs actually better than RDRS-10. A clarification for the decision seems missing and the paper would benefit from the additional information.

Specific comments:

Lines 33-37: The need for a higher spatial resolution is discussed. I suggest to add the resolution of discussed datasets, e.g. ERA5 and NARR, as well as the required resolution for land-surface and hydrological modelling applications.

Line 45: The surface reanalysis product MESCAN-SURFEX might be discussed in this section as well. For instance:

https://doi.org/10.3402/tellusa.v68.29879

http://www.uerra.eu/publications/deliverable-reports.html (deliverable 2.8)

Line 171: I am not familiar with the Yin-Yang grid. A reference would help the uninformed.

Line 198: “whole period” is not correct. Actually, ERA-interim is available 1 January 1979 – 31 August 2019.

Line 247: When is the first guess provided by GDRS? Figures 1 and 3 indicate that the first guess is based on RDRS only.

Line 531: Here is one section showing that RDRS-15 compares better to observations than RDRS-10. However, at the end it was decided to produce the final dataset with RDRS-10. The reasons remain unclear.

Lines 571-572: The given hours are hard to understand. I suggest to be explicit here:

“… are used for hours 6, 9, 12 and 15 UTC (resp. 18, 21, 0 and 3 UTC), with results shown…”

Line 582: Please rephrase the sentence.

“…, but the RDRS-10 bias value is always higher than that of the RDPS, which …”. The reader might understand RDRS-10 results are worse but that is not the case as explained afterwards. However, rephrasing this part would ease the readability.

Line 704: How can the reanalysis product, which covers only past periods, be useful for hydrological prediction?

Figure 2: The abbreviations used in the legend should be the same as in the remaining manuscript.

GEM Global should be changed to GDRS

GEM regional to RDRS

Technical corrections:

Figure 7, caption, last sentence: “Results are only based on …”

Citation: https://doi.org/10.5194/hess-2021-41-RC1
- AC1: 'Reply on RC1', Vincent Fortin, 02 Jul 2021
  
  Thanks a lot for this review and positive comments.
  Concerning your major comment, as indicated in lines 123-125 of the manuscript, “a resolution of 10km was chosen for the production of a 1980-2018 reanalysis, in order to match the current resolution of the Regional Deterministic Prediction System (RDPS) and of the Regional Ensemble Prediction System (REPS) currently in operation at CCMEP for short-term weather forecasting over North-America". Having the same resolution for the three systems (RDRS, RDPS and REPS) facilitates the computation of anomalies (by comparing RDPS or REPS to the RDRS climatology), and simplifies its application by end-users who make regular use of RDPS or REPS. The long-term plan for the RDRS is to stay as much as possible in sync with the RDPS and REPS in terms of GEM model configuration and resolution, and re-launch the reanalysis whenever major changes are made to RDPS or REPS. Finally, although some degradations are seen when increasing the resolution from 15 km to 10 km, it is shown in the paper that these differences are small. In particular, the gains obtained in terms of precipitation skill through the optimal interpolation of precipitation observations largely compensate for this small degradation (see Figure 9). It is also worth mentioning that from an hydrological application point of view, an improvement is obtained when using 10 km resolution instead of 15 km, as illustrated by Figure 10.
  During the early stage of the project, the originally targeted resolution and configuration of the system was 10 km for the very same reason as noted above. However, technical and computational resources reasons prevented from producing this preliminary 5 years sample (2010-2014) at a 10 km resolution (which was mostly dedicated to produce a proof of concept.)
  The authors agree that this should be stated more clearly and re-enforced in other sections of the manuscript, notably in Section 4.1 and 5. This will be included in a revised version of the manuscript.
  Concerning your minor comments, they will be all answered as you suggest in a revised version of the manuscript. Here we would like to further come back on some of them as we think that it can be useful and interesting for the discussion.
  L247:
  In fact, an a posteriori CaPA-24h approach was applied on both the GDRS (albeit on a subdomain covering North America only) and RDRS for the preliminary 5 years sample.
  https://collaboration.cmc.ec.gc.ca/science/outgoing/capa.grib/hindcast/capa_hindcast_rdrs_v1/Evaluation_CaPA_Reanalysis_North_America_2010-2014.pdf
  Given its usefulness, higher added value and better results, only the higher resolution CaPA-24h was produced for the final 1980-2018 reanalysis and discussed in the paper. We will remove this GDRS CaPA-24h mention from that sentence.
  Line 531:
  While it is true that RDRS-15 is slightly better than RDRS-10 for absolute and dew point temperature but not for the wind speed,it should be reinforced that despite that fact, the RDRS-10 remains better than RDPS (the operational forecasting model which is the main goal of the comparison). This is further explained around line 540.
  L704:
  Models relying on NWP outputs typically require to be calibrated in order to perform optimally, partly due to the errors/biases and shortcomings of their input datasets. This is particularly true for surface and hydrologic models.
  Such a calibration can only be performed based on archived (historical) datasets that ideally “feature” similar error climatology as the product used to drive such models in the context of forecasting. However, archived datasets from operational NWP models tend to evolve in time, and/or are usually not available for time periods long enough for such a calibration.
  Hence, the importance of having a retrospective dataset that is as close as possible to the operationally produced forecast. It thus further illustrates the importance of having a 10 km resolution along with model configuration similar to the one used to produce operational forecasts.
  Another important point worth mentioning is that hydrological models are not only used to predict future flows. For example, they can be used to predict past flows at ungauged locations, to infill missing data at gauged locations, and to perform what-if scenarios to assess the impact of climate change, land-use changes and reservoir regulation changes.
  
  Citation: https://doi.org/10.5194/hess-2021-41-AC1
RC2:
'Comment on hess-2021-41', Anonymous Referee #2, 27 May 2021
General Comments:

Overall this is a good paper which should be published. The authors develop and document an approach to produce reanalyses related to the Canadian operation NWP prediction system with an aim towards water-driven applications. The approach developed and described herein is complex and involved and sometimes leaves one with a feeling of "chewing gum and baling wire (in that a sequence of inline and offline tools are strung together is several different ways, bootstrapping down-scales, to reach the objective)." This remark is not intended as a criticism or in a negative sense: another way to state this is that this is an inventive application of available tools, that leads to some useful insights and establishes a workable system to produce reanalyses over North America.

I found the organization of the paper to be reasonable and the writing was mostly clear. Likewise, I do not have any major technical criticisms of the paper either, and found myself more or less resonating philosophically with the authors decisions, and agreeing with their logic given their objectives and circumstances. A few suggestions to improve the paper in places are offered below.

Specific Comments:

On ~line 230 + ... It is slightly disappointing, if understandable, why the authors did not employ radar data in this work. I hope that future work will allow the authors to explore the use radar data to improve precipitation analyses and assimilation. Likewise, perhaps future developments will allow them to simplify and streamline the overall process.

I was also disappointed that cloud and radiation fields were not analyzed and evaluated herein (or at least they were not presented). These could have significant impacts on the water and energy budgets, and hence on the analysis and interpretation of the other hydro-meteorological fields and their application by others. This in fact may be the most serious technical deficiency in the paper though I do not think it should prevent publication.

Lastly, in interpreting some of the biases in surface temperature and moisture in section 3, the authors may wish to consult a recent paper by Barlage et. al. in GRL (2020): Barlage, M., Chen, F., Rasmussen, R., Zhang, Z., & Miguez-Macho, G. (2021). The importance of scale-dependent groundwater processes in land-atmosphere interactions over the central United States. Geophysical Research Letters, 48, e2020GL092171. https://doi.org/10.1029/2020GL092171

(https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2020GL092171), which found both scale dependencies and a significant impact from coupling in ground water processes.

Technical Corrections:

To help readers better understand the author's work, the following suggestions are offered:

Line 278, incomplete sentence: "In order to produce a reanalysis of precipitation and land surface <***>, in addition..." <***> = a missing word: is it "land surface states" or "...fields" or "...parameters", etc.

Line 313, "mosaicked from regional multi-sensor (radar+gauges) precipitation <* analyses *> (MPEs) produced by the 12 River Forecast Centers" -> "analyses" should be "estimates".

Lines 442-443, in the sentence "...as well as observations <* across the atmosphere *> than results from...", it is not clear what is meant by "across the atmosphere". Vertically? Geographically?.

In Figure 9, the caption (at least) should define what the x-axis is in the figure (it is defined in the text, but help the reader out here...): "The x-axis is the precipitation threshold (in mm day−1)."

For Tables 4 & 5 (in the captions and where referenced in the text) present differences of RMSE, but the differences are never defined [is it 'X-Y' or 'Y-X'?]. The reader can of course 'figure it out,' but why not help the reader understand more quickly, and instead of saying (e.g.) "RDPS vs. RDRS-10", say "RDPS - RDRS-10".

Acronyms:

The authors should carefully review the use of acronyms, both to ensure that they are defined and/or are defined on first use. Some examples this reviewer found:

"ERA" (-5 and -Interim): first used on Lins 20/21; not properly defined.

"ISBA": used first on line ~143 (not defined until later in document).

"RSAS": introduced in figure in section 2.1; defined later in section 2.2.

"SYNOP, SWOB and METAR": used on line 240; defined later in section 2.5.

"COOP": line 290, not defined.
Citation: https://doi.org/10.5194/hess-2021-41-RC2
- AC2: 'Reply on RC2', Vincent Fortin, 02 Jul 2021
  
  Thank you for your review and the positive comments and suggestions. We will fully integrate them into the final version of the article.
  We recognise the complexity of the system and multiple interconnected components involved. While some streamlining is possible and will be worked on in the future, the use of existing operational system configurations was privileged for this first version of the surface and precipitation reanalysis for the North America using Global Multiscale Model (GEM), produced in Canadian Center for Meteorological and Environmental Prediction (CCMEP).
  Specific comments:
  ~L230:
  The radar data would certainly be beneficial for improving the quality and skill of precipitation analysis. They could be easily added to the offline 24 h analysis, albeit only for recent years thus affecting the time consistency of the precipitation analysis. Adding this supplementary input to the online analysis would however be much more time consuming (mostly due to technical reasons). Another possibility, even more attractive and easier to implement is to include IMERG satellite data into the final offline 24 h analysis. We are currently exploring this and it will probably be the next improvement in terms of precipitation observations assimilated.
  While cloud and radiation related fields are definitely important for surface-atmosphere interactions, they were not evaluated and analysed here, due to clarity and brevity concerns and because the focus of the paper is to introduce the surface and precipitation reanalysis and evaluate its main aspects. Those fields should be analysed in detail in a subsequent study.[Au2]
  The investigation and analysis of the soil moisture are underway and will be included in another publication.
  All technical corrections and acronyms synchronization will be incorporated into the revised version of the manuscript.
  
  Citation: https://doi.org/10.5194/hess-2021-41-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Publish subject to minor revisions (further review by editor) (09 Jul 2021) by Niko Wanders

AR by Vincent Fortin on behalf of the Authors (28 Jul 2021) Author's response Author's tracked changes Manuscript

ED: Publish as is (06 Aug 2021) by Niko Wanders

AR by Vincent Fortin on behalf of the Authors (06 Aug 2021)

Short summary

In this paper, we highlight the importance of including land-data assimilation as well as offline precipitation analysis components in a regional reanalysis system. We also document the performance of the first multidecadal 10 km reanalysis performed with the GEM atmospheric model that can be used for seamless land-surface and hydrological modelling in North America. It is of particular interest for transboundary basins, as existing datasets often show discontinuities at the border.