Once Upon A Time at DWD The COSMO-Model - Deutscher Wetterdienst - Ulrich Schättler - Deutscher ...

Deutscher Wetterdienst

  Once Upon A Time at DWD

      The COSMO-Model

                Ulrich Schättler
            Deutscher Wetterdienst
         BU Research and Development
       Department for Numerical Modelling

Deutscher Wetterdienst

     Dawn of the COSMO-Model

25.03.2021    ICCARUS 2021     2

Deutscher Wetterdienst

First Plans for a Non-Hydrostatic Regional Model
 In the 90ies DWD had two operational NWP
  models:
     GM: a global model based on IFS from
      ECMWF.
     EM/DM: regional hydrostatic grid point
      model.
 Soon after the operationalization of the
  high-resolution „Deutschland-Modell“,
  several workshops were organized with
  participation from different modeling groups
  (universities, research institutes) to define
  requirements for a new non-hydrostatic
  regional NWP model.

     25.03.2021                    ICCARUS 2021    3

Deutscher Wetterdienst

Development of the „Lokal-Modell“
 The decision was taken to develop a non-hydrostatic model using the full
  compressible basic equations with a dynamical time stepping according to
  Skamarock and Klemp (1994) (similar to MM5).
 (Some of the) Basic Requirements:
     Regional NWP model for central Europe (but global re-locatable).
     Research Model for resolutions from 1 m to 10 km.
     Urban climate model with resolution < 50 m.
 Programming Requirements:
     To run LM as NWP model, a computer with about 0.2 TFlop/s is needed.
     Most probably with massive parallel processors (MPP) and distributed
      memory.
     Use standard FORTRAN (90) as far as possible.

    25.03.2021                    ICCARUS 2021                               4

Deutscher Wetterdienst

The First Working Version (1995/96)
 A first prototype was based on the hydrostatic EM/DM System.
 The hydrostatic dynamical core of EM/DM was replaced by the split-explicit
  time stepping for the full compressible equations.
 In the beginning, the physical parameterizations from the EM/DM were used.
 Programming Environment:
     Cray YMP with 4 processors and shared memory.
     Cray Fortran 77 (already with dynamic memory management, but no
      standard).
     Code was vectorized for YMP (inherited from EM/DM).

 Although it had a similar dynamical core, it was NOT a copy of the MM5!

    25.03.2021                    ICCARUS 2021                                 5

Deutscher Wetterdienst

Development of the Parallel Program (1996)
 The first working version was
  restructured for modern parallel
  computers with distributed memory.
 After some discussions it was decided
  to use Fortran 90 and the Message
  Passing Interface (MPI).
 The original development team:
     Günther Doms ( 20.06.2004)
     Axel Diehl
     Ulrich Schättler
    With project leader Jürgen Steppeler.

    25.03.2021                     ICCARUS 2021   6

Deutscher Wetterdienst

The First New Developments
 New Repository for LM: lm_f90 Version 1.1 on
  March 11th, 1998.
 The first new parameterization for the LM was the
  TKE based subgrid scale turbulence scheme by
  Matthias Raschendorfer (introduced June 1999).
     I wanted to insist that new modules are only
      implemented to the official version, if an
      appropriate documentation exists.
     That was my first big failure!
 The nudging approach (from Christoph Schraff)
  replaced the Optimal Interpolation used for
  EM/DM analyses.

    25.03.2021                         ICCARUS 2021   7

Deutscher Wetterdienst

Towards Operational Forecasting
 Parallel to the COSMO-Model, a new
  global model was developed at DWD,
  the GME: based on the dynamics and
  physics of EM/DM, but working on an
  icosahedral grid.

 Start of operational forecasts of GME
  and LM on December, 1st, 1999.

    25.03.2021                    ICCARUS 2021   8

Deutscher Wetterdienst

First Remarkable Forecast: Lothar
 This winter storm on December 26th, 1999
  was not properly forecasted by GME and
  LM (very polite assessment).
 There were requirements at DWD „to stop
  GME and LM immediately!“
 What happened: Between 14:15 and
  15:30 UTC a „corrected measurement“
  from a second radiosonde ascent on
  Sable Island (Canada) arrived, after the
  first ascent did not get high enough. This
  measurement has been taken, but with
  valid observation time still being 12:00
  UTC, although it was about an hour later.                    12:00 MEZ
  Back in 1999, radiosondes had no
  possibility to be given a time stamp.           Hourly lowest pressure

     25.03.2021                    ICCARUS 2021                            9

Deutscher Wetterdienst

First Remarkable Forecast: Lothar
 An early forecast („Vorlauf“) with a data cut off at 2h 15 min was better!
     48h forecast of maximal wind gust (forecast started at Dec., 24th, 12 UTC)

       Data Cut Off at: 2h 15min                    3h 30min
 In the following days (Dec. 24th-26th), forecasts drifted away from reality.
 Many correct measurements were rejected by data assimilation, because they
  did not fit to the First Guess (a 6 hour forecast from the last analysis).
 But this was not noticed because of the Christmas holidays.
 With proper initial data, GME and LM could forecast this winter storm

     25.03.2021                      ICCARUS 2021                                 10

Deutscher Wetterdienst

A Strategic Goal: Cooperation

             „Key Priority at DWD“

                  (Prof. Sarah Jones in her Welcome Address on Monday)

25.03.2021               ICCARUS 2021                              11

Deutscher Wetterdienst

First Contacts
 For the development of the LM we asked for national and international
  cooperations to share know-how and expertise of universities and research
  institutes.
 Participation to the first workshops by Universities of Bonn, Hamburg,
  Hannover, Karlsruhe, Köln, Leipzig, Mainz, AWI Bremerhaven, DLR
  Oberpfaffenhofen, HZG (former GKSS) Geesthacht, KIT (former IMK)
  Karlsruhe, PIK Potsdam, Tropos (former IfT) Leipzig, GeoPhys, MCH (former
  SMA) Zürich.
 Cooperations could be started with some institutes.
 As long as a „standard“ version of LM was not available, collaboration was
  rather theoretical.
 The University of Bonn was the first institute to install a prototype of the
  „standard“ LM in 1998.

     25.03.2021                      ICCARUS 2021                                12

Deutscher Wetterdienst

COnsortium for Small scale MOdeling
 Based on the collaboration of
  DWD and MCH, Italy (ARPA
  Bologna) and Greece were
  interested in further usage and
  development of the LM.
 This was the beginning of
  COSMO.
 A first „General Meeting“ was
  held in Bologna from 23-25
  September 1999, which
  consisted of „Work Package
  Reports“ and „Plenum
  Discussions“.
 About 10 years later LM has
  been renamed COSMO-Model.

    25.03.2021                      ICCARUS 2021   13

Deutscher Wetterdienst

Workshops and Seminars
 In the European SRNWP framework,
  DWD became the „Lead Center for
  Non-Hydrostatic Modelling“ and started
  a series of workshops in March 1996.
 As a visiting lecturer at the University
  of Bonn, Jürgen Steppeler organized a
  first „LM Nutzer Seminar“ in Langen in
  June 2000.
 This became a yearly tradition now
  known as ICCARUS.
 Contributions in the first seminar were
  mainly by „experienced scientists“, but
  soon the majority of contributions were
  by students.

     25.03.2021                    ICCARUS 2021   14

Deutscher Wetterdienst

NWP Training
 After the User Seminar in 2002, the first practical training was organized.

     25.03.2021                     ICCARUS 2021                                15

Deutscher Wetterdienst

Regional Climate Model
„The history of the CLM Community
started in 2004, when colleagues from
PIK, GKSS and BTU Cottbus presented
the first version of the Climate Local
Model (CLM) and invited all scientists to
build a network aiming to develop the
CLM systematically, to coordinate their
activities and to use the computational
resources effectively. The CLM-
community was born.“

From: „History of the CLM-Community“
https://wiki.coast.hzg.de/clmcom/history-119570580.html

     25.03.2021                        ICCARUS 2021       16

Deutscher Wetterdienst

COSMO-ART
 Over several years KIT Karlsruhe
  developed ART, the „Aerosol and
  Reactive Trace gases“.
 Initially, ART was online coupled to
  KAMM, but later the COSMO-Model
  has been used (as development for
  KAMM was stopped).
 COSMO-ART has been established
  as a widely used tool to treat
  secondary aerosols, directly emitted
  components like soot, mineral dust,
  sea salt and biological material as
  pollen.
 It is available since COSMO-Model
  version 4.9 (June 2009).                  Eruption of the unpronounceable volcano in Island.
                                            COSMO-ART simulation

    25.03.2021                    ICCARUS 2021                                            17

Deutscher Wetterdienst

COSMO Licensees
 About 20 years ago the „Workstation Version“ of the EM/DM system has been
  provided as HRM to several national weather services.
 Between 2010 and 2014 these partners migrated to the COSMO-Model.

    25.03.2021                   ICCARUS 2021                                 18

Deutscher Wetterdienst

Necessary Condition for Cooperations
                           Documentation
 The original LM development team started a
  major work to write comprehensive scientific
  documentations and user guides.
 Also the ongoing work is documented now in
  form of Newsletters and Technical Reports.
 We also made efforts to document the
  ongoing development work.
     http://www.cosmo-model.org/content/
      model/releases/developments.htm
                                                 Günther Doms:  20.06.2004
 But we realize that there is much room for     Father of the COSMO-Model
  improvement!                                   and its documentation.

    25.03.2021                    ICCARUS 2021                                19

Deutscher Wetterdienst

             More than 20 Years
              of COSMO-Model

                Are too long to talk about here!

25.03.2021              ICCARUS 2021               20

Deutscher Wetterdienst

Twilight of the COSMO-Model

25.03.2021    ICCARUS 2021    21

Deutscher Wetterdienst

Last Years Versions
 Version         Date             Contents (Highlights)            Results
                                                                   Changes
 5.06a       21.05.19 • Skin temperature formulation                 if used
                      • Latest modifications to turbulence              no
                        scheme (unification with ICON)
 5.06b       16.10.19 • Implementation of radar forward operator       no
   …            …     • Revised cloud radiation coupling               no
5.06b_4      29.11.19 • Unified surface schemes with ICON          slightly (1)
                      • Changes in data assimilation                   yes
                      • Added interfaces for GHG extensions            no
                      • Option to write restart files in NetCDF        no
                        (from CLM Community)
                      • Optimizations for GPU                         no
                      • Fix in the graupel scheme                   yes (2)
See the Release Notes for more details.
    25.03.2021                     ICCARUS 2021                                   22

Deutscher Wetterdienst

Last Years Versions
 Version         Date             Contents (Highlights)              Results
                                                                     Changes
  5.07       21.02.20 •   Fix in shallow convection                  yes (3)
                      •   GPU port of Tiedtke-Bechtold convection    if used
                      •   Run surface scheme in double precision     yes (4)
                      •   CLM Unification: Modifications for MESSy      no

 5.07a       11.05.20 •   Changes in data assimilation               possible
                      •   Improvements for NEC vectorization           no
                      •   Added a hailcast diagnostic (by crCLIM)      no
                      •   Implemented EULAG dycore as an option        no

 5.07b       19.06.20 • Changes to EMVORADO                          if used

 5.07c       03.07.20 • Updates for EMVORADO                         if used

    25.03.2021                     ICCARUS 2021                                 23

Deutscher Wetterdienst

Last Years Versions
 Version         Date             Contents (Highlights)             Results
                                                                    Changes
  5.08       23.10.20 •   Changes in data assimilation              possible
                         Reunification with CLM / crCLIM / MESSy       no
                      •   Changes to SPPT for targeted diffusion     if used
                      •   Updated CPP dycore to GridTools 2.0           no
                      •   Implemented EULAG dycore as an option
  5.09       25.02.21 • Implementation of tile-approach for urban     no
                        parameterization (TERRA-URB)
                      • Fix in seaice scheme: have to update t_s    yes (5)
                        over melted ice points
                      • Additional greenhouse-gas scenarios           no
                      • NetCDF I/O:                                   no
                          • support for NetCDF4
                          • updated NetCDF standard names

    25.03.2021                     ICCARUS 2021                                24

Deutscher Wetterdienst

Last Years Versions: Modifications of Results
1. Some fixes in TERRA have been taken from ICON (numerical stability for soil
   ice freezing; increased hydraulic conductivity for clay).
2. The terminal fall velocity of snow/rain/graupel is limited, since the particles
   have a minimal size. This fact is considered at least in the lowest level.
3. When computing the final tendencies in conv_shallow.f90, erroneously the
   index „k“ was used instead of „kdim“ to indicate the lowest level for variable
   pdtdiab_con.
4. It turned out that for small timesteps the changes to the prognostic variables
   in the surface schemes are very small and cannot be represented correctly in
   single precision. Therefore the possibility has been implemented to run the
   surface schemes (TERRA, FLake, Seaice) in double precision, even if
   running the rest of the model in single precision.
5. Over ice, the temperature can be much colder than over ice-free seapoints.
   When ice is melted, the update of t_s had been forgotten.

     25.03.2021                      ICCARUS 2021                                    25

Deutscher Wetterdienst

 Reunification with CLM / crCLIM / MESSy
 Contributions from CLM (highlights)
     Option to write restart files in NetCDF.
     New diagnostics for wind sector classes; changes to sunshine duration.
     Update of NetCDF standard names.
 Contributions from crCLIM (highlights)
     New hydrology scheme (by Linda Schlemmer, Daniel Regenass).
     New diagnostics for clear-sky radiation and vertically integrated water
      fluxes.
 Contributions from MESSy (highlights)
     Adaptations to modified COSMO Code.
     Fixes for Fortran standard (due to strict checks of NAG compiler).
     Implementation of MPI 3.0 interfaces for data types.

    25.03.2021                      ICCARUS 2021                                26

Deutscher Wetterdienst

             The Last Episode:
             COSMO-Model 6.0

25.03.2021         ICCARUS 2021   27

Deutscher Wetterdienst

Two Ongoing Developments
Pull Requests for COSMO Repository on the github-page:

 SNOWPOLINO: New multi-layer snow scheme (by Sascha Bellaire et al.,
  MCH)
 TERRA-URB: urban parameterization (by Hendrik Wouters, et al.)

Both contributions still under testing. Should be consolidated during March / April.

     25.03.2021                     ICCARUS 2021                                       28

Deutscher Wetterdienst

Test, Test, Test
 In November, COSMO-Model version 5.08 has been given to the partners to
  start testing and evaluation of the final version.
 All developments after version 5.08 are optional and should not change the
  results. As long as there are no „severe“ bug fixes.
 The fix in the sea-ice scheme (5.09) is not „severe“, as it is not used in these
  evaluation tests.
 NWP Testsuite:
     A summer and a winter period running with 7.0 and 2.8 km resolution in
      hindcast mode.
     Technical modifications to the suite were necessary (hindcast mode,
      GRIB 2)
     First forecast results are available to be evaluated now.

     25.03.2021                     ICCARUS 2021                                     29

Deutscher Wetterdienst

Test, Test, Test
 CLM technical testsuite and evaluation:
     The technical testsuite passed without problems.
     A 5-year simulation run successfully, but no full evaluation done (but:
      „sieht gut aus“).
     For further information see WG EVAL.

 crCLIM evaluation:
     Some technical issues were fixed in the last weeks.
     For further information see WGs EVAL / SupTech.

    25.03.2021                     ICCARUS 2021                                 30

Deutscher Wetterdienst

Status of COSMO-Model 6.0
 Testing of SNOWPOLINO developments is about to be finished.
 Testing of TERRA-URB implementation is about to start.
 Implementation into the master version is expected in April/May (Version 5.x)

 Depending on the results of the NWP Testsuite and the CLM / crCLIM
  evaluation, COSMO-Model 6.0 can be released afterwards.

 COSMO-Model versions 5.08 and 5.09 are not officially released, but are
  available for testing to everybody.

     25.03.2021                    ICCARUS 2021                                   31

Deutscher Wetterdienst

Further Maintenance of the COSMO-Model
 After COSMO-Model version 6.0 we will only do "perfective maintenance"
  (bug fixes, technical modifications where necessary, etc.)
 DWD already replaced the COSMO-Model on February 10th by ICON-LAM.
 But support will go on until all users have migrated to ICON-LAM

    25.03.2021                    ICCARUS 2021                             32

Deutscher Wetterdienst

             Migration to ICON

25.03.2021         ICCARUS 2021   33

Deutscher Wetterdienst

Availability of ICON
 COSMO is a main development partner of DWD. ICON is available to every
  COSMO member. The weather services already started to test ICON-LAM
  (see PP C2I).
 To institutions, ICON (for NWP applications) is available under an
  institutional license issued by DWD. This includes limited support to install
  and run ICON. (Similar to COSMO).
 To individuals, ICON is distributed under a personal, non-commercial
  research license distributed by MPI-M.

 Weather services, which are COSMO NWP license partners, will be
  migrated to ICON-LAM in the upcoming years. Brazil INMET already takes
  part in PP C2I.

     25.03.2021                     ICCARUS 2021                                  34

Deutscher Wetterdienst

Changes to the (COSMO) Workflow
 All development work is now
  organized by a common git
  repository on DKRZ‘s gitlab server.
 Below icon-nwp-dev there is the
  icon-cosmo-master, from which the
  development branches for COSMO
  and CLM start.
 All developments from COSMO or
  CLM community have to start from
  these branches.

    25.03.2021                    ICCARUS 2021   35

Deutscher Wetterdienst

             Conclusions

25.03.2021      ICCARUS 2021   36

Deutscher Wetterdienst

A SWOT Analysis for COSMO
Strengths                                    Weaknesses
 Usable community model                      Documentation of many details
                                               nevertheless is missing
 Basic model documentation
  exists                                      Many developments took too long
 Basic support available                     Feedback mechanism „forecast –
                                               model development“ could have
                                               been more efficient

Opportunities                                Threats
 Bring together users and scientists         Better models (ICON)
  from all over the world

     25.03.2021                    ICCARUS 2021                                37

Deutscher Wetterdienst

Taking the Best of Both Worlds
 Infrastructures for COSMO and ICON are rather different (groups, models,
  workflows).
 Does any group have to fear a hostile takeover? Sure NOT!
 Worlds are not clashing! There is the chance to take the best of both worlds!
  For example:
     ICON is the more powerful (and better) simulation model
     COSMO has the better documentation (model documentation, release
      notes)

                                                  And COSMO definitely has….

     25.03.2021                    ICCARUS 2021                                   38

Deutscher Wetterdienst

The COSMO Spirit

  25.03.2021       ICCARUS 2021   39