Sensitivity experiments with the Runge Kutta time integration scheme Lucio TORRISI CNMCA – Pratica di Mare (Rome)

Slides:



Advertisements
Similar presentations
Introduction to data assimilation in meteorology Pierre Brousseau, Ludovic Auger ATMO 08,Alghero, september 2008.
Advertisements

Mei Xu, Jamie Wolff and Michelle Harrold National Center for Atmospheric Research (NCAR) Research Applications Laboratory (RAL) and Developmental Testbed.
1 Les règles générales WWOSC August, Montréal, Canada Didier Ricard 1, Sylvie Malardel 2, Yann Seity 1 Julien Léger 1, Mirela Pietrisi 1. CNRM-GAME,
COSMO Workpackage No First Results on Verification of LMK Test Runs Basing on SYNOP Data Lenz, Claus-Jürgen; Damrath, Ulrich
How do model errors and localization approaches affects model parameter estimation Juan Ruiz, Takemasa Miyoshi and Masaru Kunii
Günther Zängl, DWD1 Improvements for idealized simulations with the COSMO model Günther Zängl Deutscher Wetterdienst, Offenbach, Germany.
Krakow - September, 15th 2008COSMO WG 2 - Runge Kutta1 Further Developments of the Runge-Kutta Time Integration Scheme Investigation of Convergence (task.
Case 2 15 March Case 3 2 – 3 April 2003.
Lucio Torrisi WG2-COSMO GM, Bucharest 2006 Recents Improvements in LMZ Code Lucio TORRISI CNMCA – Pratica di Mare (Rome)
For the Lesson: Eta Characteristics, Biases, and Usage December 1998 ETA-32 MODEL CHARACTERISTICS.
Transitioning unique NASA data and research technologies to the NWS 1 Evaluation of WRF Using High-Resolution Soil Initial Conditions from the NASA Land.
Ensemble Post-Processing and it’s Potential Benefits for the Operational Forecaster Michael Erickson and Brian A. Colle School of Marine and Atmospheric.
Jamie Wolff Jeff Beck, Laurie Carson, Michelle Harrold, Tracy Hertneky 15 April 2015 Assessment of two microphysics schemes in the NOAA Environmental Modeling.
LAMEPS Development and Plan of ALADIN-LACE Yong Wang et al. Speaker: Harald Seidl ZAMG, Austria Thanks: Météo France, LACE, NCEP.
COSMO General Meeting Zurich, 2005 Institute of Meteorology and Water Management Warsaw, Poland- 1 - Verification of the LM at IMGW Katarzyna Starosta,
Deutscher Wetterdienst 1 Status report of WG2 - Numerics and Dynamics COSMO General Meeting , Offenbach Michael Baldauf Deutscher Wetterdienst,
Francesca Marcucci, Lucio Torrisi with the contribution of Valeria Montesarchio, ISMAR-CNR CNMCA, National Meteorological Center,Italy First experiments.
NWP at the Italian Meteorological Service: STATUS REPORT 29 th EWGLAM, Dubrovnik, 8-11 October 2007 Massimo Bonavita, Lucio Torrisi, Antonio Vocino and.
Development of WRF-CMAQ Interface Processor (WCIP)
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Parametrization of Subgrid-Scale Orographic Drag in the.
Introducing the Lokal-Modell LME at the German Weather Service Jan-Peter Schulz Deutscher Wetterdienst 27 th EWGLAM and 12 th SRNWP Meeting 2005.
In this study, HWRF model simulations for two events were evaluated by analyzing the mean sea level pressure, precipitation, wind fields and hydrometeors.
Comparison of convective boundary layer velocity spectra calculated from large eddy simulation and WRF model data Jeremy A. Gibbs and Evgeni Fedorovich.
COSMO General Meeting, Offenbach, 7 – 11 Sept Dependance of bias on initial time of forecasts 1 WG1 Overview
INSTYTUT METEOROLOGII I GOSPODARKI WODNEJ INSTITUTE OF METEOROLOGY AND WATER MANAGEMENT TITLE : IMPLEMENTATION OF MOSAIC APPROACH IN COSMO AT IMWM AUTHORS:
Ligia Bernardet 1*, E. Uhlhorn 2, S. Bao 1* & J. Cione 2 1 NOAA ESRL Global Systems Division, Boulder CO 2 NOAA AOML Hurricane Research Division, Miami.
Sensitivity Analysis of Mesoscale Forecasts from Large Ensembles of Randomly and Non-Randomly Perturbed Model Runs William Martin November 10, 2005.
Seasonal Modeling (NOAA) Jian-Wen Bao Sara Michelson Jim Wilczak Curtis Fleming Emily Piencziak.
Operational ALADIN forecast in Croatian Meteorological and Hydrological Service 26th EWGLAM & 11th SRNWP meetings 4th - 7th October 2004,Oslo, Norway Zoran.
HNMS contribution to CONSENS Petroula Louka & Flora Gofa Hellenic National Meteorological Service
Ensemble data assimilation in an operational context: the experience at the Italian Weather Service Massimo Bonavita and Lucio Torrisi CNMCA-UGM, Rome.
Overview on the Work of WG 6 Reference Version and Implementation WG Coordinator: Ulrich Schättler.
Verification Verification with SYNOP, TEMP, and GPS data P. Kaufmann, M. Arpagaus, MeteoSwiss P. Emiliani., E. Veccia., A. Galliani., UGM U. Pflüger, DWD.
10 th COSMO General Meeting, Krakow, September 2008 Recent work on pressure bias problem Lucio TORRISI Italian Met. Service CNMCA – Pratica di Mare.
INTERCOMPARISON – HIRLAM vs. ARPA-SIM CARPE DIEM AREA 1 Per Kållberg Magnus Lindskog.
Statistical Post Processing - Using Reforecast to Improve GEFS Forecast Yuejian Zhu Hong Guan and Bo Cui ECM/NCEP/NWS Dec. 3 rd 2013 Acknowledgements:
NWP at the Italian Weather Service: CURRENT STATUS AND PERSPECTIVES Massimo Bonavita, Lucio Torrisi, Fabrizio Ciciulla and Antonio Vocino UGM-CNMCA, Rome.
10 th COSMO General Meeting, Krakow, September 2008 Recent work on pressure bias problem Lucio TORRISI Italian Met. Service CNMCA – Pratica di Mare.
General Meeting Moscow, 6-10 September 2010 High-Resolution verification for Temperature ( in northern Italy) Maria Stefania Tesini COSMO General Meeting.
1 Reformulation of the LM fast- waves equation part including a radiative upper boundary condition Almut Gassmann and Hans-Joachim Herzog (Meteorological.
Eidgenössisches Departement des Innern EDI Bundesamt für Meteorologie und Klimatologie MeteoSchweiz Tuning the horizontal diffusion in the COSMO model.
Use of ALADIN for dynamical downscaling of precipitation Jure Cedilnik University of Ljubljana, Slovenia.
Overview of WG5 activities and Conditional Verification Project Adriano Raspanti - WG5 Bucharest, September 2006.
Vincent N. Sakwa RSMC, Nairobi
Evaluation of cloudy convective boundary layer forecast by ARPEGE and IFS Comparisons with observations from Cabauw, Chilbolton, and Palaiseau  Comparisons.
Group 14: East Asia Members: Gomboluudev (Mongolia), Kwon (Korea), Nguyen (Vietnam) Land Use and Cumulus Scheme Studies Using RegCM3.
Comparison of LM Verification against Multi Level Aircraft Measurements (MLAs) with LM Verification against Temps Ulrich Pflüger, Deutscher Wetterdienst.
Status Report WG2 J. Steppeler, DWD Zurich Z-Coordinate Runge Kutta and Semi-Lagrangian methods Direct implicit solvers Courant number independent.
Deutscher Wetterdienst 1FE 13 – Working group 2: Dynamics and Numerics report ‘Oct – Sept. 2008’ COSMO General Meeting, Krakau
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Status of the COSMO-1 configuration at MeteoSwiss Guy.
Global vs mesoscale ATOVS assimilation at the Met Office Global Large obs error (4 K) NESDIS 1B radiances NOAA-15 & 16 HIRS and AMSU thinned to 154 km.
OSEs with HIRLAM and HARMONIE for EUCOS Nils Gustafsson, SMHI Sigurdur Thorsteinsson, IMO John de Vries, KNMI Roger Randriamampianina, met.no.
HIRLAM coupled to the ocean wave model WAM. Verification and improvements in forecast skill. Morten Ødegaard Køltzow, Øyvind Sætra and Ana Carrasco. The.
A revised formulation of the COSMO surface-to-atmosphere transfer scheme Matthias Raschendorfer COSMO Offenbach 2009 Matthias Raschendorfer.
Introducing the Lokal-Modell LME at the German Weather Service
Operational COSMO of MeteoSwiss
Evaluation of TM5 performance for assimilation purpose
Development of nonhydrostatic models at the JMA
QPF sensitivity to Runge-Kutta and Leapfrog core
Italian Meteorological Service Pratica di Mare, Rome, Italy
Tuning the horizontal diffusion in the COSMO model
Recent developments in Latent Heat Nudging at DWD
Operational Aladin-Belgium
Rita Roberts and Jim Wilson National Center for Atmospheric Research
COSMO implementations at CNMCA: tests with different dinamycal options
Ulrich Schättler Source Code Administrator
Latest Results on Variational Soil Moisture Initialisation
Conservative Dynamical Core (CDC)
Objective verification results for RK parallel runs
The Impact of Moist Singular Vectors and Horizontal Resolution on Short-Range Limited-Area Ensemble Forecasts for Extreme Weather Events A. Walser1) M.
Presentation transcript:

Sensitivity experiments with the Runge Kutta time integration scheme Lucio TORRISI CNMCA – Pratica di Mare (Rome)

Introduction A new dynamical core has been developed in LM (Forstner and Doms, 2004). It is based on a TVD variant of 3rd-order Runge Kutta time integration scheme (RK) using a 5th-order spatial discretization of advection. The RK core should be more accurate than the standard Leap-Frog/2nd-order advection scheme (LF) and it will be used for very detailed short range forecasts. RK core needs to be tested and evaluated, before it can be operationally implemented.

Overview LM configuration and verification method. RK compared to LF. RK sensitivity to the: - integration time step (72s, 48s); - interval between two calls of some parameterizations (convection, turbulence); - turbulence parameterization scheme (diagnostic and prognostic TKE); - domain size; - moisture variables advection scheme (eulerian, semi-lagrangian); - moisture variables transport formulation. Summary and conclusion.

EXPERIMENTAL RUN ON FUJITSU (ECMWF) LM CONFIGURATION (version 2.10) Domain size465 x 385 (EuroLM) Grid spacing (7 km) Number of layers35 Time step72 sec Forecast range24 hrs Initial time of model run00 UTC Lateral boundary conditions and initial stateOp. IFS (preproc. with CNMCA-IFS2LM) L.B.C. update frequency3 hrs OrographyFiltered (eps = 0.1) Prognostic precipitationOn Rayleigh damping schemeRelaxation to LM filtered fields Interval between two calls of turbulence1 time step Turbulence parameterizationPrognostic TKE R.damping: filter iteration number10 R. damping: filter length1 LM configuration (v ) Domain size234 x 272 Grid spacing (7 km) Number of layers35 Time step and integration scheme40 sec, 3 timelevel split-explicit Forecast range48 hrs Initial time of model run00/12 UTC Lateral boundary conditionsGME L.B.C. update frequency1 hrs Initial stateGME InitializationD. F. External analysisNone StatusOperational HardwareIBM SP3 (Bologna) N° of processors used64

Objective verification method Statistical verification through comparison of LM forecasts with lowland station observations in the period 24 – 28 March 2005 (5 runs). Nearest grid point is used as LM forecast. Only land stations with h<700m and height mismatch with model topography less than 100m were used. About 3500 fc-obs pairs were used to calculate the mean error and RMSE of the surface variables forecast. They are enough to make statistical comparisons between different configurations of LM.

LF (40s) compared to RK (72s)

RK performs worse than LF for MSLP forecasts due to a large bias (also small differences in other surface variables and wind vector). RK has a slightly smaller upper level temperature RMSE. RK and LF use different time steps that determine a different time interval between two calls of physics. One experiment to find out the cause of the MSLP deficiency in RK could be to decrease the time step from 72s to 40s (time step for LF), in order to have the same parameterizations calling frequency for LF and RK.

RK (40s) compared to RK (72s)

RK with 40s time step has a slightly smaller MSLP bias than RK with 72s time step at T+18h and T+24h. This result seems to be due to the higher accuracy associated with the smaller time step. To totally exclude the influence of the interval between two calls of parameterization schemes, some experiments are useful. One experiment is to decrease the convection calling frequency nincconv from 10 to 5. Another experiment is to increase the interval between two calls of the prognostic TKE turbulence scheme ninctura from 1 to 2 (every two time steps instead of every time step).

RK (40s) with ninctura = 1, 2 or with nincconv = 10, 5

RK is not significantly sensitive to the calling frequency of the convection and prognostic TKE turbulence scheme. What is the behaviour of the diagnostic TKE turbulence scheme compared to the prognostic one? RK (40s) with ninctura = 1, 2 or with nincconv = 10, 5

RK with new and old turbulence

RK with the old turbulence scheme seems to perform better (smaller bias and standard deviation) than RK with the prognostic TKE turbulence parameterization. The improvement in the MSLP bias is related to the low level positive temperature bias of RK with old turbulence. The prognostic TKE turbulence parameterization seems to be one of the likely candidate to justify the MSLP forecast deficiency in RK. However, a large bias is still present! How does RK with old turbulence perform compared to the LF? RK with new and old turbulence

LF and RK with old turbulence

Using the old turbulence scheme the large MSLP bias difference between RK and LF is slightly reduced. A positive MSLP bias (except for T+12h) is present in RK, but a slightly smaller standard deviation than in LF is also found. Some tuning of the RK+physics core is necessary to reduce the large bias in MSLP forecast (cold bias in upper levels), but the slight improvement in the standard deviation seems to be an indication of the higher-order accuracy of the RK compared to the LF. LF and RK with old turbulence

RK with different domain sizes

The enlargement of the domain size seems to have a negative impact (larger RMSE for forecast times greater than T+6h) on the MSLP forecast. A similar result was obtained for a longer period using the LF (Torrisi, 2005). The increase of the standard deviation could be related to the improvement of the intrinsic variability of the model associated with the enlargement of the domain (BC are slightly affecting the forecast). RK with different domain sizes

RK with Eulerian and SL moisture variables advection

The SL moisture advection scheme does not show any significant difference in MSLP forecast compared to the Eulerian one (slightly larger standard deviation after T+18h balanced by a slightly smaller bias), but it seems to have a slightly better skill for 6h accumulated precipitation. RK with Eulerian and SL moisture variables advection

RK with conservation form of moisture variables transport

The conservative form of the moisture variables transport has a larger MSLP bias than the default formulation. A slight improvement after T+18h is obtained switching on the prognostic advection of density. RK with conservation form of moisture variables transport

Summary and Conclusion The comparison of LF and RK schemes was performed for a 5 days period using the EuroLM configuration. Statistical verification results showed that RK performance for surface variables was slightly better than LF one. A large MSLP bias was typical of the RK runs. Some sensitivity studies were performed on RK to determine the cause of the MSLP forecast deficiency. RK did not show any sensitive to the calls of the prognostic TKE turbulence and convection schemes. An improvement in the MSLP forecast was obtained using the old turbulent scheme, but a larger bias was found again. The impact of the domain size and different moisture variables transport formulations was also evaluated. More work, especially investigations on the numerics- physics interaction, are needed to improve the RK core.