improved initialization and prediction of clouds in ... · canada, canada, cdata assimilation and...

6th WMO Symposium on Data Assimilation -- 2013

Improved Initialization and Prediction of Clouds in Numerical Weather Prediction

Thomas Aulignéa, Gael Descombesb, and Francois Vandenbergheb

a National Center for Atmospheric Research, Boulder, Colorado, USA, [email protected],

b National Center for Atmospheric Research, Boulder, Colorado, USA.

The initialization of cloud parameters is one of the next frontiers for improving short-term prediction skills in numerical weather prediction models. We will present the latest developments in building a capability to accurately initialize cloud microphysical parameters based on satellite observations. Retrieved cloud optical properties and also all-sky satellite radiances from multiple infrared and microwave sensors are assimilated in the Weather Research and Forecasting (WRF) numerical model. The prototype uses a hybrid ensemble/variational data assimilation system with an augmented control variable for clouds, flow-dependent multivariate background errors and specific developments to address non-linearities in the observation operator and non-Gaussian error distributions related to cloud parameters. The impact of the cloud initialization on subsequent forecasts will be demonstrated.


Development and Operational Implementation of Flow-dependent Wavelet-based Correlations using the Ensemble 4D-Var at Météo-France

Loïk Berre, Gérald Desroziers, and Hubert Varella

Météo-France, France, [email protected]

A variational ensemble data assimilation system is being run at Météo-France since 2008 ([1]), in order to estimate and provide flow-dependent error variances to the deterministic 4D-Var system. It is also used in order to initialize the global ensemble prediction system. This ensemble data assimilation system has been designed to simulate the error evolution of the deterministic 4D-Var system, and it now also includes a model error estimation and representation ([2]), which relies on the comparison of ensemble- and innovation-based estimates of error variances. With respect to background error correlations, a flow-dependent wavelet approach has been developed and experimented ([3], [4]), based on sliding 4-day averages, and this is planned to become operational at Météo-France in July 2013. Wavelets are functions which vary in terms of location and scale, which means that e.g. wavelet error variances contain information on geographical variations of error correlations. A static version of wavelet-based correlations can be estimated in order to capture “climatological” heterogeneities, but this does not allow weather regime dependencies to be represented. A flow-dependent version has therefore been experimented at Météo-France. It is based on the use of sliding 4-day averages, in order to estimate associated time variations in a robust way. Such a flow-dependent wavelet approach is also a way to filter sampling noise through the use of implicit local spatial averages of ensemble correlation estimates. The foundations and results of this wavelet filtering will be shown using examples from the Météo-France ensemble data assimilation. The connection and distinction with other filtering approaches (such as the Schur filter in gridpoint space) will also be discussed. References [1] Berre, L. and G. Desroziers, 2010: Filtering of Background Error Variances and Correlations by Local Spatial Averaging: A Review. Mon. Wea. Rev., 138, 3693-3720. [2] Raynaud, L., L. Berre, and G. Desroziers, 2012 : Accounting for model error in the Meteo-France ensemble data assimilation system. Quart. J. Roy. Meteor. Soc., 138, 249-262. [3] Varella, H., Berre, L., and Desroziers, G., 2011 : Diagnostic and impact studies of a wavelet formulation of background error correlations in a global model. Quarterly Journal of the Royal Meteorological Society, Volume : 137, Issue : 658, Pages : 1369-1379. [4] Berre, L., Varella, H., and Desroziers, G., 2013 : Modelling of flow-dependent ensemble-based background error correlations using a wavelet formulation, in preparation.


Four-Dimensional Ensemble-Variational Data Assimilation for Global Deterministic Weather Prediction

Mark Buehnera, Josée Morneaub, and Cecilien Charettec

aData Assimilation and Satellite Meteorology Research Section, Environment Canada, Canada,

[email protected], bData Assimilation and Quality Control Development Section, Environment Canada, Canada, cData Assimilation and Satellite Meteorology Research Section, Environment Canada,

Canada.

The goal of this study is to evaluate a version of the ensemble-variational data assimilation approach (EnVar) for possible replacement of 4D-Var at Environment Canada for global deterministic weather prediction. This implementation of EnVar relies on 4D ensemble covariances, obtained from an ensemble Kalman filter, that are combined in a vertically dependent weighted average with simple static covariances. Verification results are presented from a set of data assimilation experiments over two separate 6-week periods that used assimilated observations and model configuration very similar to the currently operational system. To help interpret the comparison of EnVar versus 4D-Var, additional experiments using 3D-Var and a version of EnVar with only 3D ensemble covariances are also evaluated. Analyses from EnVar (with 4D ensemble covariances) nearly always produces improved, and never degraded, forecasts when compared with 3D-Var. Comparisons with 4D-Var show that forecasts from EnVar analyses have either similar or better scores in the troposphere of the tropics and the winter extra-tropical region. However, in the summer extra-tropical region the medium-range forecasts from EnVar have either similar or worse scores than 4D-Var in the troposphere. In contrast, the 6h forecasts from EnVar are significantly better than 4D-Var relative to radiosonde observations for both periods and in all regions. The use of 4D versus 3D ensemble covariances only results in small improvements in forecast quality. By contrast, the improvements from using 4D-Var versus 3D-Var are much larger. Measurement of the fit of the background and analyzed states to the observations suggests that EnVar and 4D-Var can both make better use of observations distributed over time than 3D-Var. In summary, the results from this study suggest that the EnVar approach is a viable alternative to 4D-Var, especially when the simplicity and computational efficiency of EnVar are considered. Additional research is required to understand the seasonal dependence of the difference in forecast quality between EnVar and 4D-Var in the extra-tropics.


Developing the 20th Century Reanalysis version 3 (1850-2013)

Gilbert P. Compoa,b, Jeffrey S. Whitakerb, and Prashant D. Sardeshmukha,b , Benjamin S. Giesec

a CIRES, University of Colorado, USA, [email protected], bPhysical Sciences Division, Earth System

Research Laboratory, NOAA, USA., cDepartment of Oceanography, Texas A&M University , USA

The historical reanalysis dataset generated by NOAA ESRL and the CIRES Climate Diagnostics Center, the Twentieth Century Reanalysis[1] version 2 (20CRv2), is a comprehensive global atmospheric circulation dataset spanning 1871-2011, assimilating only surface pressure and using monthly Hadley Centre SST and sea ice distributions (HadISST1.1) as boundary conditions. It has been made possible through collaboration with GCOS, WCRP, and the ACRE initiative. It is chiefly motivated by a need to provide an observational validation dataset, with quantified uncertainties, for assessments of climate model simulations of the 20th century, with emphasis on the statistics of daily weather. It uses, together with an NCEP global numerical weather prediction (NWP) land/atmosphere model to provide background "first guess" fields, an Ensemble Kalman Filter (EnKF) data assimilation method. This yields a global analysis every 6 hours as the most likely state of the atmosphere, and also yields the uncertainty of that analysis. The 20CRv2 dataset provides the first estimates of global tropospheric variability, and of the dataset's time-varying quality, spanning 1871 to the present at 2 degree spatial resolution. Intercomparisons with independent radiosonde and station temperature data indicate that the reanalyses are of high quality. Overall, the quality is approximately that of current three-day NWP forecasts. It is anticipated that the 20CRv2 will be useful to the climate research community for both diagnostic studies and model validations. Some surprising results are already evident. For instance, the long-term trends of the tropical Pacific Walker Circulation are weak or non-existent over the full period of record in this dataset. Following 20CRv2, with GCOS, WCRP, and ACRE, we are investigating 20CR version 3: an improved version of the historical reanalysis dataset. Results illustrating the effects of higher spatial resolution and increased observational density compared to 20CRv2 will be presented. 20CRv3 will have a companion ocean reanalysis generated by Texas A&M University using the Simple Ocean Data Assimilation system. The effects of using SODA boundary conditions compared to HadISST in 20CRv3 will be investigated. Together 20CRv3 and SODA will provide global states of the atmosphere, land, and ocean back to 1850. References [1] G.P. Compo, J.S. Whitaker, P.D. Sardeshmukh, N. Matsui, R.J. Allan, X. Yin, B.E. Gleason, R.S. Vose, G. Rutledge, P. Bessemoulin, S. Brönnimann, M. Brunet, R.I. Crouthamel, A.N. Grant, P.Y. Groisman, P.D. Jones, M. Kruk, A.C. Kruger, G.J. Marshall, M. Maugeri, H.Y. Mok, Ø. Nordli, T.F. Ross, R.M. Trigo, X.L. Wang, S.D. Woodruff, and S.J. Worley, 2011: The Twentieth Century Reanalysis Project. Quarterly J. Roy. Meteorol. Soc., 137, 654, 1-28. DOI: 10.1002/qj.776, February 2011. (Journal Article)

A Large Scale Host Model Constraint in a Limited Area 4D-Var

Per Dahlgrena and Nils Gustafssonb

a Research department , Swedish Meteorological and Hydrological Institute (SMHI), Sweden,

[email protected], b Research department, SMHI, Sweden.

A regional NWP model require a host model that provide lateral boundaries. We would like to present a method to include the large scales from the host model in the data assimilation process. One reason for doing this is to have the large scales in the regional model consistent with the host model already at the analysis time.

We add an extra term, Jk, to the cost function that measure the distance to the large scale vorticity field of the host model that provides lateral boundaries. The error characteristics of the Jk term are described by the horizontal spectral densities, the vertical eigenvectors and eigenvalues of the host model described in the regional model geometry. This constraint has been tested and evaluated with the HIRLAM model 4D-Var analysis on large domain covering the north Atlantic and Europe. Evaluations show clear forecast improvements up to +48h.

References[1] Dahlgren P. and Gustafsson N. “Assimilating host model information into a limited area model”, Tellus A 2012, vol. 64, 15836, DOI: 10.3402/tellusa.v64i0.15836http://www.tellusa.net/index.php/tellusa/article/view/15836


http://www.tellusa.net/index.php/tellusa/article/view/15836

The February 2013 Upgrade to the Canadian Operational Ensemble Kalman Filter

Xingxiu Denga, P. L. Houtekamerb, Herschel L. Mitchellb, Seung-Jong Baekb, and Normand Gagnona

a Canadian Meteorological Centre, Environment Canada, Canada, [email protected],

b Meteorological Research Division, Environment Canada, Canada.

The Canadian operational Ensemble Kalman Filter (EnKF) provides an ensemble of initial conditions for the Canadian Global Ensemble Prediction System (GEPS). The EnKF underwent a major upgrade in February 2013, thanks to the availability of upgraded computing facilities. The major changes include the introduction of a multi-scale algorithm and increased horizontal, vertical and temporal resolution. At the same time, a filtered topography was introduced to address an occasional instability problem seen in the previously operational GEPS. The number of assimilated radiance observations was increased via a relaxation of the data-thinning procedures. Other changes include the use of a newer version of the forecast model with improved physics, and less-biased radiance observations obtained from our center's independently improved Global Deterministic Prediction System (GDPS). With all these changes, the Canadian EnKF having 192 members is run on a 600x300 horizontal grid with 74 vertical levels. The forecast model uses a 20-minute time step. The impact of each major change was evaluated and the results will be presented. The overall impact of the upgrade will also be presented.


Application of the WRF – LETKF System over Argentina: a Case Study.

María E. Dillon (a) (b)

, Juan Ruiz (a) (c) (d)

, Estela A. Collini (b) (e)

, Yanina Garcia Skabar (a) (b) (d) (f)

(a)

Consejo Nacional de Investigaciones científicas y técnicas (CONICET), Argentina, (b)

Departamento de

Investigación y Desarrollo, Servicio Meteorológico Nacional, Argentina, [email protected] , (c)

Departamento de Ciencias de la Atmósfera y los Océanos, FCEyN, UBA, Argentina, (d)

UMI- Instituto Franco

Argentino sobre Estudios del Clima y sus Impactos, Argentina, (e)

Servicio de Hidrografía Naval, Argentina, (f)

Cátedra de Climatología y Fenología Agrícolas, Facultad de Agronomía, UBA, Argentina.

Improving the initial conditions of the short–range numerical weather prediction models is

one of the main concerns of the meteorological community. Different data assimilation

methods have been developed and are used operationally at the most important prediction

centers of the world.

The Weather and Research Forecasting Model (WRF-ARW) was implemented

experimentally at the National Meteorological Service of Argentina in 2010, and has been

run on daily basis in a quasi – operational form. The inclusion of data assimilation,

particularly using the Local Ensemble Transform Kalman Filter (LETKF) method produced

at the University of Maryland, in this forecasting system is being developed since the end

of 2012.

In this article we present a mesoscale convective system case study that occurred over the

central part of Argentina on December 6th

of 2012, which produced several damages in

various cities. The forecasts obtained from the WRF – LETKF system are evaluated. An

experiment carried out including the Atmospheric Infrared Sounder (AIRS) retrieval

products, particularly the atmospheric humidity and temperature profiles, revealed an

improvement in the forecasts, in agreement with similar experiments from other regions.

These preliminary results encourage our efforts in the development of a consistent data

assimilation system, capable of being implemented in real time at the National

Meteorological Service of Argentina.


AFES–LETKF Experimental Ensemble Reanalysis 2

Takeshi Enomotoa, b, Akira Yamazakib, T. Miyoshic, b, A. Kuwano-Yoshidab, N. Komorib, J. Inoued, e, b, M. E. Horie, b, Q. Motekie, b, M. Hattorie, b and S. Yamanef

a Disaster Prevention Research Institute, Kyoto University, Japan, [email protected],

b Earth Simulator Center, Japan Agency for Marine-Earth Science and Technology, Japan, c RIKEN Advanced Institute for Computational Science, Japan,

d National Institute for Polar Research, Research Organization of Information and Systems, Japan, e Research Institute for Global Change, Japan Agency for Marine-Earth Science and Technology, Japan,

f Department of Environmental Systems Science, Doshisha University, Japan.

An atmospheric reanalysis dataset has been constructed with the atmospheric general circulation model for the Earth Simulator (AFES) and the local ensemble transform Kalman filter (LETKF)[1]. It is a successor of AFES–LETKF experimental ensemble reanalysis (ALERA)[2], hence called ALERA2. This dataset provides six-hourly 63-member analyses and their ensemble mean and spread for five years from 1 January 2008 to 9 January 2013. ALERA2 takes advantage of improvements of the forecast model and the data assimilation code. With the updated physical packages, AFES has improved forecast skills in spite of a slightly coarser horizontal resolution of 1°. Distance-based covariance localization[3] is used in LETKF to remove inconsistencies of the analysis ensemble spread near the poles. Observations are obtained from the NCEP PREPBUFR archive of UCAR. In addition to analyzed variables, ALERA2 includes variables diagnosed in the forecast model such as precipitation and surface fluxes. ALERA2 is used as the reference of observing system experiments for field campaigns conducted by JAMSTEC. For example, OSE’s were conducted to evaluate radiosonde observations from a research vessel Mirai in the Arctic in 2010[4]. With the radiosonde observations, the tropopause folding associated with a cyclone is better represented. Radiosonde observations have not only local but also remote influences on the Northern Hemisphere mid-latitudes. References [1] T. Enomoto, T. Miyoshi, Q. Moteki, J. Inoue, M. Hattori, A. Kuwano-Yoshida, N. Komori, and S. Yamane. “Observing-system research and ensemble data assimilation at JAMSTEC.” In S. K. Park and L. Xu, editors, Data Assimilation for Atmospheric, Oceanic and Hydrological Applications (Vol. II). Springer, 2013, in press. [2] T. Miyoshi, S. Yamane, and T. Enomoto. “The AFES-LETKF experimental ensemble reanalysis: ALERA.” SOLA, 3, 45–48, April 2007. [3] T. Miyoshi, S. Yamane, and T. Enomoto. “Localizing the error covariance by physical distances within a local ensemble transform Kalman filter (LETKF).” SOLA, 3, 89–92, August 2007. [4] J. Inoue, T. Enomoto and M. E. Hori, “The impact of radiosonde data over the ice-free Arctic Ocean on the atmospheric circulation in the Northern Hemisphere.” Geophys. Res. Lett., 40 (5), 864–869, doi: 10.1002/grl.50207, March 2013.

6th

WMO Symposium on Data Assimilation -- 2013

Initial Trials of 4D-Ensemble-Var for Data Assimilation

and Ensemble Initialization

Jonathan Flowerdewa, Stephen Pring

b, Peter Jermey

b,

Andrew Lorencb, Neill Bowler

b, and Eunjoo Lee

c,b

a Weather Science, Met Office, UK, [email protected]

b Weather Science, Met Office, UK,

c Numerical Prediction, KMA, South Korea.

Hybrid data assimilation schemes augment flow-dependent ensemble covariances with extra

samples from a climatological covariance. Variational hybrids combine ensemble information

with the existing investment in methods such as 4-dimensional variational assimilation (4DVar)

[1]. However, the perturbation-forecast and adjoint models which 4DVar uses to evolve

covariances over time impose significant computational and maintenance cost, and may not scale

well on future massively parallel computer systems. 4D-Ensemble-Var (4DEnVar) offers an

alternative approach, in which the temporal correlations are taken from the ensemble [2,3].

This presentation will show results from early trials of 4DEnVar in the Met Office global

Numerical Weather Prediction (NWP) system. For initializing a deterministic forecast, we find

that 4DEnVar is superior to 3DVar and hybrid-3DVar, has similar average performance to

4DVar, but is inferior to hybrid-4DVar (the current operational system), when all are run at the

same resolution. 4DEnVar performs relatively worse in the Southern Hemisphere, where reduced

observation density places greater reliance on the accuracy of the ensemble covariances.

However, the 4DEnVar system is a factor 3-6 cheaper than hybrid-4DVar, even though the latter

has been optimized much more than the former. These cost savings could be recycled into extra

resolution, extra ensemble members, and/or use of an outer loop.

The reduced cost of 4DEnVar makes it an attractive method for ensemble initialization. It has

theoretical advantages over the Ensemble Transform Kalman Filter (ETKF) currently used by the

Met Office, in areas such as localization, re-linearization, the use of balanced variables, and

greater consistency with the way the central analysis is produced [4]. A single system serving

both purposes should also reduce maintenance costs. Better ensemble perturbations may in turn

benefit the hybrid data assimilation, further improving both the deterministic and ensemble

forecasts. We will present the results of initial trials comparing 4DEnVar with the current ETKF

in the global ensemble, including consideration of alternative inflation methods such as

relaxation-to-prior-spread and additive inflation, and re-centering of the ensemble about the high-

resolution deterministic analysis.

References

[1] A. M. Clayton, A. C. Lorenc, D. M. Barker. "Operational implementation of a hybrid

ensemble/4D-Var global data assimilation system at the Met Office," Q. J. R. Meteorol. Soc.,

DOI: 10.1002/qj.2054, November 2012.

[2] C. Liu, A. Xiao, B. Wang. "An Ensemble-Based Four-Dimensional Variational Data

Assimilation Scheme. part II: Observing System Simulation Experiments with advanced

research WRF (ARW)," Mon. Weather Rev., vol. 137, no. 5, pp. 1687–1704, May 2009.

[3] M. Buehner, P. L. Houtekamer, C. Charette, H. L. Mitchell, B. He. "Intercomparison of

variational data assimilation and the ensemble Kalman filter for global deterministic NWP.

Part II: One-month experiments with real observations," Mon. Weather Rev., vol. 138, no. 5,

pp. 1567–1586, May 2010.

[4] N. E. Bowler, J. Flowerdew, S. R. Pring. "Tests of different flavours of EnKF on a

simple model," Q. J. R. Meteorol. Soc., DOI: 10.1002/qj.2055, December 2012.


Ensemble Kalman Filter Data Assimilation for the MPAS system

So-Young Ha, Chris Snyder, Bill Skamarock, Jeffrey Anderson, Nancy Collins, Michael Duda, Laura Fowler, Tim Hoar

[email protected]

National Center for Atmospheric Research

Boulder, Colorado, U. S. A.

The Model for Prediction Across Scales (MPAS; http://mpas-dev.github.io/) is a global non-hydrostatic numerical atmospheric model based on unstructured centroidal Voronoi meshes that allow both uniform and variable resolutions. The variable resolution allows locally high-resolution meshes that transition smoothly to coarser resolution over the rest of the globe, avoiding the need to drive a limited-area model with lateral boundary conditions from a separate global model. Recently we established an interface between the MPAS and the Data Assimilation Research Testbed (DART; http://www.image.ucar.edu/DAReS/DART) system, and successfully completed analysis/forecast cycling experiments with real observations for summer months of 2008. Assimilated observations are all conventional data as well as satellite winds and GPS radio occultation refractivity data. Through these retrospective studies, we will examine issues specific to the MPAS grid, such as smoothing in the interpolation and the update of horizontal wind fields, and show their impact on the Ensemble Kalman Filter (EnKF) analysis and the following short-range forecast. Because interfaces for several other models are available within DART, we compare the MPAS results to those from cycling experiments with an identical EnKF and identical observations using either the Community Atmosphere Model (CAM) for the uniform grids or the Weather Research and Forecasting (WRF) regional model driven by boundary conditions from MPAS for the local, high-resolution grid.

Cloud Condensate Control Variable Transform and Background Errors

El í as Valur H ó lm a, Jiandong Gongb, and Philippe Lopeza

a European Centre for Medium-Range Weather Forecasts, UK, [email protected], b National

Meteorological Center, Chinese Meteorological Administration, China.

We will present progress on work to add cloud condensate control variable to a 4D-VAR analysis system at ECMWF. First, in 4D-VAR the linear physics scheme needs prognostic cloud variables to be able to use additional cloud information from observations, and in the current phase of our development the additional prognostic variable is cloud condensate (cloud liquid water and ice). Second, it is necessary to describe the relationship of cloud condensate background errors to the errors in other variables, and we have found a physical-statistical balance relationship between cloud condensate, humidity (and temperature) errors which are included as a change of variable from total to “unbalanced” cloud condensate. Third, the normalization of cloud condensate by its background error standard deviation requires particular attention because of the spatial inhomogeneity of the cloud field and the non-Gaussian errors. In this presentation we will focus on the formulation of cloud condensate background errors.

The relationship between cloud condensate and humidity/temperature background errors was investigated using forecast difference samples from ECMWF ensemble data assimilation (EDA), as was the gaussianity of different versions of the cloud condensate control variable. A global wavelet covariance matrix was derived from the EDA for the resulting cloud condensate control variable. On of the main challenges in deriving cloud condensate covariance matrices (and variances) is that clouds are not present everywhere all of the time. For variances, we have addressed this by imposing minimum climatological values of the cloud condensate errors, and similarly for the covariances nonzero values are imposed above the tropopause. The wavelet covariances are determined from a large sample of EDA forecast differences spanning different seasons, which gives a series of climatological covariances, one every few hundred kilometers. This results in different covariance matrices in different regions, with broad vertical correlations in convective regions and narrow vertical correlations in subsidence regions. Single observation experiments will be used to illustrate the different aspects of the control variable transform in clear and cloudy regions.



Improving GSI 3DVAR-Ensemble Hybrid Data Assimilation System for Mesoscale Application with the Rapid Refresh

Ming Hua, David Dowellb, Steve Weygandtc, Stan Benjamind, Jeff Whitakere, Curtis Alexanderf

aCIRES, University of Colorado and NOAA/ESRL/GSD/AMB, Boulder, CO, USA, [email protected] b,c,d NOAA/ESRL/GSD/AMB, Boulder, CO, USA

e NOAA/ESRL/PSD, Boulder, CO, USA fCIRES, University of Colorado and NOAA/ESRL/GSD/AMB, Boulder, CO, USA

NOAA’s Rapid Refresh system (RAP) is an hourly-updated regional data-assimilation and forecasting system that uses the Weather Research and Forecasting (WRF-ARW) model with 13-km horizontal grid spacing and the Gridpoint Statistical Interpolation (GSI) analysis package. The RAP version 1 has been running operationally at NCEP since May, 2012. Since then, many new advanced features have been developed and applied in a real-time experimental RAP version 2 (RAPv2) system. Among the RAPv2 enhancements, switching to a 3DVAR-Ensemble hybrid data assimilation procedure within GSI is the most important analysis upgrade. Application of the hybrid technique in the RAPv2 resulted in an immediate significant positive impact, as was reported at the 93rd AMS Annual Meeting, in January 2013.

The current RAPv2 GSI hybrid data analysis system is using coarser resolution GFS EnKF ensemble forecasts at 6 hour intervals to contribute half of the background error covariance. Most of the hybrid configurations are based on either default values or values from similar systems. The initial application of GSI hybrid within the RAPv2 has already significantly improved middle and upper level wind and water vapor forecasts, but we believe there is still some room for further improvement. We are examining a number of enhancements, including using EnKF ensemble forecasts at one or three hour intervals, tuning the ratio of static and ensemble BE, allowing BE ratio to vary vertically, and optimizing vertical and horizontal localizations for this mesoscale application. We also plan to test a RAP ensemble with GSI hybrid assimilation, which would be initialized from GFS EnKF ensemble members toward a goal of improving RAP forecasts of near surface fields and localized weather phenomena.

In the next couple of months, we will continue work on a series of experiments to improve the RAP GSI hybrid configurations as described above. This talk will share the lessons learned from these tests and report on the progress of the GSI hybrid application for the RAP system.

6th


The ECMWF Hybrid 4D-Var and Ensemble of Data Assimilations

Lars Isaksena, Massimo Bonavita

a, and Elías Hólm

a

a European Centre for Medium-Range Weather Forecasts, United Kingdom, [email protected]

The operational 4-Dimensional Variational (4D-Var) data assimilation system at the European

Centre for Medium-Range Weather Forecasts (ECMWF) was in June 2011 extended to a hybrid

system, where flow-dependent background error variances for balanced variables were provided

by an Ensemble of 4D-Var Data Assimilations (EDA) [1, 2, 3]. In addition the EDA was used to

compute new climatological background error covariances in June 2012. With the June 2013

upgrade the EDA will provide flow-dependent variance estimates for unbalanced background

error variables. At the end of 2013 we furthermore plan to extend the number of members from

10 to 25 to provide an on-line estimation of flow-dependent covariance matrices. All these

background error related upgrades of the assimilation system have resulted in significant analysis

and forecast skill improvements. The hybrid system upgrades were a major reason for ECMWF

forecast improvements during the last three years. The paper will present the EDA-based hybrid

assimilation system method, associated results, and plans for further development of the hybrid

assimilation system. We will also discuss the viability of 4D-Var for the future in face of

challenges like scalability and the need for an accurate tangent-linear approximation.

References

[1] L. Isaksen, M. Bonavita, R. Buizza, M. Fisher, J. Haseler, M. Leutbecher and L.

Raynaud. “Ensemble of data assimilations at ECMWF” ECMWF Technical Memorandum

no. 636, 45 pp., December 2010. Available from http://www.ecmwf.int/publications

[2] M. Bonavita, L. Raynaud and L. Isaksen. “Estimating background-error variances with

the ECMWF Ensemble of Data Assimilations system: some effects of ensemble size and day-

to-day variability” Quart. J. Roy. Meteor. Soc., vol. 137B, no. 655, pp. 423-434, January

2011.

[3] M. Bonavita, L. Isaksen and E. Holm. “On the use of EDA background error variances

in the ECMWF 4D-Var” Quart. J. Roy. Meteor. Soc., vol. 138B, no. 667, pp. 1540–1559, July

2012.


Local Ensemble Transform Kalman Filter Data Assimilation System Implemented to a Next-Generation Global Model of KIAPS

Ji-Sun Kanga, Jong-Im Parka, Seoleun Shina, Eugenia Kalnayb, and Takemasa Miyoshic

a Data Assimilation Team, Korea Institution of Atmospheric Prediction Systems, Republic of Korea,

[email protected], b Department of Atmospheric and Oceanic Science, University of Maryland, United States, c Data Assimilation Research Team, RIKEN Advanced Institute for Computational Science, Japan.

Korea Institute of Atmospheric Prediction Systems (KIAPS) has been developing a next-generation global numerical weather prediction (NWP) model as well as advanced data assimilation systems. As one of the most advanced data assimilation methods, Local Ensemble Transform Kalman Filter (LETKF) data assimilation [1] has been selected as the first data assimilation system implemented to the KIAPS global NWP model. Since the dynamic core of KIAPS global NWP model is very likely to have fully unstructured quadrilateral meshes based on the cubed-sphere grid, such as a spectral element dynamic core of NCAR’s High-Order Method Modeling Environment (HOMME) [2], we have first examined a modified observation operator in terms of a horizontal interpolation method. We select four points of irregular model grids not only that are closest to the observation, but also that surround the obseration in order to avoid an extrapolation. Because it is not guaranteed that those four points form a rectangle, we have attempted a different interpolation method, inverse distance weighted interpolation, from a standard bilinear interpolation. We have investigated an impact of the modified observation operator using a model with a regular grid system, by comparing its result with that of the standard observation operator. In addition, we expect to explore many interesting features introduced by the irregular grid system of the advanced KIAPS global NWP model during an analysis cycle. We have been testing the LETKF data assimilation using the simulated conventional observations and satellite data such as IASI and AMSU-A. We also plan to include an assimilation of Global Positioning System (GPS) Radio Occultation (RO) data. Progress of our LETKF data assimilation system for the KIAPS global NWP model will be presented at the symposium. References [1] Hunt, B. R., E. Kostelich, and I. Szunyogh. "Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter", Physica D, 230, 112-126, 2007. [2] Dennis, J. M., J. Edwards, K. J. Evans, O. Guba, P. H. Lauritzen, A. A. Mirin, A. St-Cyr, M. A. Taylor, and P. H. Worley. “CAM-SE: A scalable spectral element dynamic core for the Community Atmosphere Model”, The International Journal of High Performance Computing Applications, 26, 74-89, 2012.


4D-Ensemble-Var for the deterministic NCEP GFS: Dual resolution sensitivity experiments with real observations

Daryl T. Kleista

a NOAA/NWS/National Centers for Environmental Prediction/Environmental Modeling Center, USA,

[email protected]

The ability to incorporate flow-dependent, ensemble-based representations of background error covariances into variational data assimilation has recently been developed and tested for use in the NCEP data assimilation system (Gridpoint Statistical Interpolation, GSI) by utilizing the augmented control variable method. Experiments with the hybrid 3D-Ensemble-var (EnVar) system for the NCEP Global Forecast System (GFS) model have shown that the hybrid paradigm can yield substantial forecast error reduction relative to a 3DVAR-based control system in both single [1] and dual-resolution paradigms [2]. The hybrid 3D EnVar algorithm was implemented into NCEP operations for the GFS in May 2012, relying on an EnKF to update the ensemble. By taking the existing 3D EnVar algorithm in GSI and allowing for four-dimensional ensemble perturbations, coupled with the 4DVAR infrastructure already in place, a 4D EnVar capability has been developed. The 4D EnVar algorithm has a few attractive qualities relative to 4DVAR including the lack of need for tangent-linear and adjoint model as well as reduced computational cost. Results for the GFS from an observing system simulation experiment (OSSE) show that analysis error was reduced when going from 3D- to 4DEnVar [3]. Building upon the OSSE-based experimentation, low resolution experiments using real data have been carried out with the GFS comparing hybrid 3D and 4DEnVar initialization. For deterministic forecasts, it was found that those that were initialized from the 4DEnVar were generally superior to those initialized using 3DEnVar, particularly in the Southern Hemisphere. This presentation will focus on follow-on 4DEnVar sensitivity experiments, exploring potential improvements that can be gained through alternative initialization techniques (such as 4D incremental analysis update or weak constraint digital filtering), outer loops (re-running of the nonlinear model for the guess trajectory), and inclusion of temporal information within the static contribution to the solution. References [1] X. Wang, D. Parrish, D. Kleist, and J. Whitaker, “GSI 3DVar-based ensemble-variational hybrid data assimilation for the NCEP global forecast system: Single resolution experiments, Mon. Wea. Rev., accepted. [2] T. M. Hamill, J. S. Whitaker, D. T. Kleist, P. Pegion, M. Fiorino, and S. G. Benjamin, “Predictions of 2010’s tropical cyclones using the GFS and ensemble-based data assimilation methods,” Mon. Wea. Rev., 139, 3243-3247. [3] D. T. Kleist, “An evaluation of hybrid variational-ensemble data assimilation for the NCEP GFS,” Ph.D. Thesis, Dept. of Atmospheric and Oceanic Science, University of Maryland-College Park, 149 pp.


Japanese 55-year Reanalysis (JRA-55): status and plans

Shinya Kobayashia, Yukinari Otaa, Yayoi Haradaa, Ayataka Ebitaa, Masami Moriyaa, Hirokatsu Onodaa, Kazutoshi Onogia, Hirotaka Kamahorib, Chiaki Kobayashib, Hirokazu Endob,

Kiyotoshi Takahashia, Kengo Miyaokaa, and Ryoji Kumabea

a Japan Meteorological Agency (JMA), Japan, [email protected], b Meteorological Research Institute (MRI), JMA, Japan.

JMA has been conducting the second Japanese global reanalysis named JRA-55 (JRA go go) to provide a comprehensive atmospheric dataset that is suitable for studies of climate change and multi-decadal variability. It covers 55 years, extending back to 1958, when the regular radiosonde observations became operational on the global basis. The data assimilation system for JRA-55 is based on the JMA operational numerical weather prediction system as of December 2009, into which many improvements have been implemented since the time of the JRA-25 production. JRA-55 is the first global atmospheric reanalysis that applies four-dimensional variational assimilation (4D-Var) to the past half century including the pre-satellite era [1]. Improvement of long-term climate datasets is essential for the advancement of climate researches and services such as investigations of mechanisms of the climate system, studies of predictability and climate monitoring. Reanalysis of the past observations with a consistent, state-of-the-art data assimilation system has been used broadly for these purposes. It is due to the advantage that it can produce many kinds of meteorological variables, including those for which observations are rarely available, in a spatiotemporally regular manner. Reanalysis has especially been making great contributions to studies of synoptic and planetary scale phenomena such as storm tracks, blocking, MJO, ENSO and QBO. However, there remain several issues that need to be resolved in order to make a broader use of reanalysis in the climate discipline. The top issue is varying quality that originates in changes of observing systems. Many of the existing reanalyses exhibit changes, which appear to be non-climatic, around the year 1979 when the full-blown satellite observing system became operational, and the year 1998 when the ATOVS observing system was introduced. It has been pointed out that care is needed in using the reanalyses to study climatic trends and low-frequency decadal variations. In addition to addressing this issue, we also aim at overcoming deficiencies identified in JRA-25 such as a cold bias in the stratospheric temperature during the TOVS period (from the 1980s to 1990s) and the dry land surface problem in the Amazon basin. Early results of quality assessment have suggested that many of deficiencies in JRA-25 have been diminished or reduced in JRA-55. Specifically the cold bias in the lower stratospheric temperature was significantly reduced and the dry land surface problem in the Amazon basin was mitigated. Temporal consistency of the temperature analysis has also improved considerably from that of the existing reanalyses, though impact of changing observing systems is still detectable. The factors related to these quality changes will be discussed from the viewpoint of data assimilation. The production of JRA-55 completed in March 2013, and the preparation for the data release in autumn 2013 and the start of the near real-time production in early 2014 is in progress. Alongside of JRA-55, MRI is currently conducting a conventional observation only reanalysis (JRA-55C) and an AMIP type run (JRA-55AMIP) with the same data assimilation system and forecast model as used in JRA-55. We hope that inter-comparison among the “JRA-55 family” will provide an opportunity for quantitative assessment regarding impact of changing observing systems and model biases on representation of climatic trends and low-frequency variations. Reference [1] A. Ebita et al. “The Japanese 55-year Reanalysis (JRA-55): an interim report,” SOLA, vol. 7, pp. 149-152, 2011.

Sixth WMO Symposium on Data Assimilation

Improving tropical cyclone forecasts by assimilating satellite sounding measurements Jun Li1, Tim Schmit2, Mitch Goldberg3, Jinlong Li1, Pei Wang1, and John L. Beven4

1. CIMSS, University of Wisconsin – Madison, WI, U.S.A 2. Advanced Satellite Products Branch (ASPB), NOAA/NESDIS/STAR, U.S.A. 3. JPSS Program Office, NOAA/NESDIS, U.S.A 4. National Hurricane Center, NOAA/NWS, U.S.A.

Abstract Reliable and stable forecasts on tropical cyclones (TCs) such as Isaac and Sandy landed onto CONUS in 2012 are critical for decision making and better preparation. Observations of atmospheric temperature and moisture information in environment and hurricane region are very important to the prediction of the genesis, intensification, motion, rainfall potential, and landing impacts of TCs through numerical weather prediction (NWP) models. The AIRS/AMSU on Aqua, IASI/AMSU on Metop, CrIMSS (CrIS and ATMS) on Suomi NPP provide atmospheric temperature and moisture information with high vertical resolution and accuracy, which is critical for the prediction of hurricane evolution. In order to maximize the benefit of satellite microwave and advanced infrared (IR) sounder measurements for TC forecasts, real time assimilation and forecasting system is developed at CIMSS, it is called satellite Sounder Data Assimilation for TC forecasts (SDAT). The regional NWP models (WRF - Weather Research and Forecasting, and HWRF – Hurricane WRF) along with the operational Community Gridpoint Statistical Interpolation (GSI) assimilation system developed by NCEP (National Centers for Environmental Prediction) are used as the frames of SDAT, which comprises of data ingestion, processing, assimilation and forecasting. The conventional and satellite sounder observations (microwave and IR radiances, soundings, layer precipitable water - LPW etc.) are ingested into Bufr file used by GSI, then 72-hour forecasts are conducted after each assimilation. SDAT can not only assimilate radiances, but also can assimilate products such as high temporal resolution total precipitable water (TPW) from geostationary satellites such as GOES and MSG. AIRS/AMSU, IASI/AMSU and CrIMSS radiance measurements are used in assimilation and forecast experiments for hurricanes Irene (2011), Isaac (2012) and Sandy (2012). Radiance assimilation (3DVAR) and product assimilation (1DVAR/3DVAR) are compared for TC forecasts, comparable impacts were found between two approaches, although there are pros/cons of the two approaches. Overall, positive impact of assimilating microwave and advanced IR sounder measurements (both from radiances and soundings) on hurricane track and intensity forecasts is found.

Achieving Superior Tropical Cyclone Intensity Forecasts by Improving the Assimilation of High-Resolution Satellite Data into Mesoscale Prediction Models Hui Liua, Chris Veldenb, Sharan Majumdarc, Jun Lib, Tingchi Wuc, and Jeff Andersona a National Center for Atmospheric Research (NCAR), Boulder, Colorado, [email protected] b Cooperative Institute for Meteorological Satellite Studies (CIMSS) University of Wisconsin-Madison, Madison c Rosenstiel School of Marine and Atmospheric Science (RSMAS), University of Miami, Miami Forecasts of tropical cyclone intensity change have shown little improvement during the past decade, due in part to the lack of conventional observations over oceans and deficiencies in mesoscale modeling and assimilation. In this study, we build on recent advancements in mesoscale data assimilation to exploit the use of high spatial and temporal resolution satellite measurements in mesoscale model forecasts of tropical cyclones. These observations include rapid-scan Atmospheric Motion Vectors (AMVs, 15-mins), hyper-spectral AIRS advanced soundings of temperature and water vapor soundings (~15km), AMSU retrieved surface winds (25km), and AMSR total Precipitable Water (TPW, ~25km). The observations are assimilated into the Weather Research and Forecasting model (WRF) with 27km/9km nested grids. Parallel cycling assimilation experiments with and without the different types of satellite observations are performed for two TC cases, Sinlaku and Ike (2008) using the Ensemble Kalman Filter in the NCAR Data Assimilation Research Testbed (DART). The forecasts initialized from analyses that use these high-resolution satellite observations capture the intensification of the TCs during their early stages. The wind analyses of Sinlaku are found to be more consistent with independent ELDORA radar observations throughout the vertical column. In general, the impact of assimilating AMV observations dominate the impacts of other observation types assimilated in this study.

Exploring Multi-scale and Model-error Treatments in Ensemble Data Assimilation

Takemasa Miyoshia, Shigenori Otsukaa, and Keiichi Kondoa, b

a Data Assimilation Research Team, RIKEN Advanced Institute for Computational Science, Japan,

[email protected], b Graduate School of Life and Environmental Sciences, University of Tsukuba, Japan.

Ensemble data assimilation methods have been improved consistently and have become a viable choice in operational numerical weather prediction. A number of issues for further improvements have been explored, including flow-adaptive covariance localization and advanced covariance inflation methods. Dealing with multi-scale error covariance and model errors is among the unresolved issues that would play essential roles in analysis performance. In this study, a “dual-localization” approach has been developed to consider the multi-scale error covariance. With higher resolution models, generally narrower localization is required to reduce sampling errors in ensemble-based covariance between distant locations. However, such narrow localization limits the use of observations that would have larger-scale information. Previous attempts include successive covariance localization by Zhang et al. [1] who proposed applying different localization scales to different subsets of observations. The method aims at typically using sparse radiosonde observations at a larger scale, while using dense Doppler radar observations at a small scale simultaneously. The dual-localization method aims at separating scales of the analysis increments, independent of observing systems. Inspired by Buehner [2], two different localization scales are applied to find analysis increments at the two separate scales, and obtained significant improvements in simulation experiments using an intermediate AGCM known as the SPEEDY model. Another important issue is about the model errors. Among many other efforts since Dee and da Silva’s model bias estimation [3], we explore a discrete Bayesian approach to adaptively choosing model physics schemes or other parameters that produce better fit to observations. Also, the traditional state-augmentation approach to model parameter estimation has been explored. This presentation summarizes our recent progress at RIKEN on these theoretical and practical topics for further improvement of ensemble data assimilation approaches. References [1] F. Zhang, Y. Weng, J. A. Sippel, Z. Meng, and C. H. Bishop. “Cloud-Resolving Hurricane Initialization and Prediction through Assimilation of Doppler Radar Observations with an Ensemble Kalman Filter,” Monthly Weather Review, vol. 137, no. 7, 2105-2125, July 2009. doi:10.1175/2009MWR2645.1 [2] M. Buehner. “Evaluation of a Spatial/Spectral Covariance Localization Approach for Atmospheric Data Assimilation,” Monthly Weather Review, vol. 140, no. 2, 617–636, February 2012. doi:10.1175/MWR-D-10-05052.1 [3] D. P. Dee and A. M. da Silva. “Data assimilation in the presence of forecast bias,” Quart. J. Roy. Meteor. Soc., vol. 124, 269-295.


Specification of Background Errors in the ECMWF 20th Century Reanalysis Using Surface-Only Observations

(ERA-20C)

Paul Polia, Dick Dee a, David Tan a, Elias Hólm a, Massimo Bonavita a, Lars Isaksen a, and Mike Fisher a

a European Centre for Medium-Range Weather Forecasts, United Kingdom,

[email protected]

Reanalyses apply data assimilation to estimate the atmospheric state from an incomplete set of observations over long time periods. For example, Compo et al. [1] produced the NOAA 20th Century Reanalysis, assimilating surface pressure observations into an atmospheric model forced by sea-surface boundary conditions. However, the surface pressure observing system improves greatly throughout the course of the 20th century, with a near 50-fold increase in the number of observations, yet with immense disparities between Northern and Southern hemispheres. This represents a challenge to reconstruct a global weather history that is coherent in time.

Denying this observational fact by considering averages in a climatological fashion, or arbitrarily discarding observations so as to sub-sample them to a daily global constant number of observations does not really address the problem. Indeed, the global observing network is a sum of regional networks which evolved in their geographical coverages, their local reporting times, and also their instrumental qualities.

One way to envision an optimal data assimilation system for reanalysis is to extract varying amounts of information depending on the prior knowledge of the atmospheric state and to reflect this information in the uncertainty estimates that should accompany the reanalysis products. The great challenge posed by observing system changes appears then as an opportunity to develop and evaluate new data assimilation methods.

To this effect, ECMWF is producing, with funding from the European Union FP7 ERA-CLIM project, a reanalysis of the 20th century (ERA-20C). It assimilates atmospheric surface pressure observations over land and ocean, and atmospheric surface wind observations over ocean. The data assimilation system in ERA-20C is a hybrid ensemble of four-dimensional variational analyses (4D-Var). We will describe how the ensemble provides a set of background realizations that enable to update the local background error variances. This allows for geographic variations, representative of regional network disparities. We will also present how the full set of global background error covariances are updated automatically at regular intervals as the production advances. We will finally illustrate how this scheme performs in ERA-20C, based on production results, and how effectively it captures the slow improvement over time in reanalysis quality that results from an increasing number of observations.

References[1] G. P. Compo and Co-authors. “The Twentieth Century Reanalysis Project”, Q.J.R. Meteorol. Soc., vol. 137, pp. 1-28, 2011. DOI: 10.1002/qj.776



Initial Performance and Verification Results of a Cycled WRF/DART Multiscale Ensemble System at Texas Tech University

Anthony E. Reinhart and Brian Ancell

Department of Geosciences, Atmospheric Science Group, Texas Tech University, USA,

[email protected].

A real-time 50 member EnKF cycling system using DART/WRF has been in production at Texas Tech University (TTU) for several months. The system uses a nested domain configuration with an outer 36-km grid over most of the U.S. and North Pacific Ocean, and smaller 12-km (Western U.S.) and 4-km (Texas, Oklahoma, and a portion of New Mexico) nests. The system cycles every 6 hours, producing extended forecasts on the two finer nests (48-hr at 12km, 36-hr at 4km) twice daily from the 0000 and 1200 UTC analyses. Standard cloud-track wind, aircraft, radiosonde, and METAR observations are assimilated onto the 36-km and 12-km grids, with additional mesonet surface observations assimilated onto the fine-scale 4-km domain. The performance of the ensemble system will be presented here. The ensemble was initially calibrated over a limited period to possess appropriate localization and inflation over a 24-hr forecast window to produce optimal mean forecasts. Spread/skill metrics and rank histograms are evaluated here to reveal whether desirable ensemble behavior, particularly with regard to spread and bias, is maintained throughout the extended forecasts. Ensemble mean forecasts are verified against hourly surface temperature, moisture, wind, and precipitation observations, as well as temperature, pressure, moisture, and wind observations aloft every 6 hours. In addition to comparisons to observations, ensemble performance is compared to several other numerical weather prediction (NWP) models and configurations. In particular, this comparison will include a TTU 12-km/3-km real time WRF deterministic simulation initialized from the GFS, as well as the U.S. operational NAM, GFS, and RAP models. Plans for future system improvement will be discussed based on these results.

6th


Assimilating Conventional Observations for the Global Atmospheric Model

SL-AV with Local Ensemble Transform Kalman Filter

Anna Shlyaevaa,b

, Mikhail Tolstykhc,b

, Vasily Mizyakb, and Vladimir Rogutov

b

a Bauman Moscow State Technical University, Russia, [email protected],

b Hydrometeorological

Research Centre, Russia, c Institute of Numerical Mathematics, Russian Academy of Sciences, Russia.

The presentation concerns implementation of the Local Ensemble Transform Kalman Filter

scheme [1] to the global numerical weather prediction model SL-AV [2], operational in Russia.

SL-AV is a semi-Lagrangian global atmospheric model that uses semi-implicit finite-difference

dynamical core of own development and parameterizations from ALADIN/LACE model. The

experiments are carried out with the 0.9x0.72 degrees, 28 vertical sigma-levels model version.

The localization is done in the observation space, using different localization distances for

different variables and different vertical levels. We use multiplicative and additive inflation to

account for model error and the finite ensemble size. For additive inflation, a random noise is

added at every analysis step to every ensemble member. Additive inflation fields are isotropic for

every variable, use different predefined length-scales and amplitudes for different variables (the

implementation also allows them to differ in vertical).

We present the results of assimilation of conventional observations (ground stations, radiosondes,

aircraft reports and atmospheric motion vectors subset), comparing different configurations of the

assimilation system. Overall, the ensemble filter has shown stable behavior during 3 months

experiment. The implemented analysis efficiently reduces first guess root mean square errors and

removes first guess biases. The root mean square error values for temperature and wind are close

to that for the LETKF assimilation system for the NCEP global model using conventional

observations, presented in [3]. The ensemble distribution is close to Gaussian (based on the third

and fourth central distribution moments).

References

[1] B. R. Hunt, E. J. Kostelich and I. Szunyogh. "Efficient data assimilation for

spatiotemporal chaos: A local ensemble transform Kalman Filter," Physica D: Nonlinear

Phenomena, vol. 230, pp. 112-126, 2007.

[2] M.A. Tolstykh. "Variable resolution global semi-Lagrangian atmospheric model,"

Russian J. Num. An. & Math. Mod., vol. 18, no. 4, pp. 347-361, 2003.

[3] I. Szunyogh, E. J. Kostelich, G. Gyarmati, E. Kalnay, B. R. Hunt, E. Ott, E. Satterfield

and J. A. Yorke. "A local ensemble transform Kalman filter data assimilation system for

the NCEP global model," Tellus A, vol. 60A, pp. 113-130, 2007.

mailto:[email protected]

6th


The CNMCA (Italian National Meteorological Center) Operational LETKF

Data Assimilation System: recent developments

Lucio Torrisia , Francesca Marcucci

b and Valerio Cardinali

c

a Italian National Meteorological Center, Rome, Italy, [email protected],

b Italian National

Meteorological Center, Rome, Italy, c Euro-Mediterranean Center for Climate Change, Italy

The Italian National Meteorological Center has tested and implemented an ensemble data

assimilation algorithm based on the LETKF approach [1]. The CNMCA-LETKF data

assimilation system [2,3] is used operationally to initialize the deterministic COSMO-ME model

(7km) since 1 June 2011. LETKF is running with 40+1 members having a 10 km grid spacing.

Recently the change of the prognostic model (from HRM to COSMO) has been evaluated.

The observational dataset operationally ingested comprises radiosonde ascents (RAOB, surface

pressure observations from land and sea stations (SYNOP, SHIP, BUOY), manual and automatic

aircraft observations, atmospheric motion vectors from Meteosat 9, European wind profilers,

scatterometer winds from METOP and AMSU-A radiances. First results of forecast sensitivity to

observations (FSO [4] ) will be presented.

Moreover, COSMO-ME forecasts initialized by LETKF and IFS 4D-Var analysis were

objectively were compared. LETKF gives verification results slightly worse or similar to IFS 4D-

Var analysis, but it should be taken into account that LETKF runs with less observations than IFS

4DVAR (shorter cutoff time and less satellite observation types)

References

[1] Hunt, B. R., E. Kostelich, I. Szunyogh. “Efficient data assimilation for spatiotemporal

chaos: a local ensemble transform Kalman filter”, Physica D, 230, 112-126, 2007

[2] Bonavita M, Torrisi L, Marcucci F. “The ensemble Kalman filter in an operational

regional NWP system: Preliminary results with real observations”, Q. J. R. Meteorol. Soc.,

134, 1733-1744, 2008

[3] Bonavita M, Torrisi L, Marcucci F. “Ensemble data assimilation with the CNMCA

regional forecasting system”, Q. J. R. Meteorol. Soc., 136, 132-145, 2010

[4] Kalnay E., Ota Y., Miyoshi T. and Liu J. “A simpler formulation of forecast sensitivity

to observations: application to ensemble Kalman filters”, Tellus, 64A, 18462, 2012


GSI-based Four-dimensional Ensemble-variational (4DEnsVar) Data Assimilation for NCEP Global Forecast System: Single Resolution Experiments with Real Data

Xuguang Wanga and Ting Leib

abSchool of Meteorology, and Center for Analysis and Prediction of Storms, University of Oklahoma, USA

Corresponding author address: [email protected] A 3DVar-based ensemble-variational (3DEnsVar) hybrid data assimilation system was recently developed based on the Gridpoint Statistical Interpolation (GSI) data assimilation system operational at the National Centers for Environmental Prediction (NCEP), and was first tested for the Global Forecast System (GFS) [1]. It was found that the new 3DEnsVar hybrid system produced more accurate forecasts than the operational GSI 3DVar system for both the hurricane forecasts and the general global forecasts [2], [1]. This system was implemented operationally for global NWP at NCEP. The current GSI based 3DEnsVar hybrid system did not account for the temporal evolution of the error covariance within the assimilation window. While a GSI based four–dimensional variational data assimilation (DA) system where the innovation is propagated in time using a tangent linear and adjoint (TLA) of the forecast model is being developed, an alternative method to account for the temporal evolution of the error covariance within the GSI system was implemented and tested. In this method, the ensemble perturbations valid at multiple time levels within the DA window are effectively used to estimate the four-dimensional (4D) background error covariances during the variational minimization, conveniently avoiding the need of the TLA of the forecast model. This method is called “4D-Ensemble-Var (4DEnsVar)” [3]. The performance of the system was investigated with the NCEP GFS where the ensemble and control forecast were run at the same, reduced resolution. The ensemble was supplied by the EnKF. The experiments were conducted over a summer month period assimilating the NCEP operational conventional and satellite data. Both the general global forecasts and the hurricane track forecasts were verified. A series of single observation experiments revealed that the newly developed 4DEnsVar was able to reflect the temporal evolution of the background error covariance in the DA window whereas the 3DEnsVar could not. Verification of the global forecasts showed that 4DEnsVar in general improved upon 3DEnsVar. At short lead times, the improvement of 4DEnsVar relative to 3DEnsVar over Northern Hemisphere (NH) was similar to that over Southern Hemisphere (SH). At longer forecast lead times, 4DEnsVar showed more improvement in SH than NH. The improvement of 4DEnsVar over Tropics (TR) was neutral or slightly positive. Track forecasts of 16 tropical cyclones during the verification period were verified. The track forecasts initialized by the 4DEnsVar were more accurate than 3DEnsVar after the 2-day forecast lead time. The analysis generated by 4DEnsVar was more balanced than 3DEnsVar. The performance of the 4DEnsVar was degraded when less frequent ensemble perturbations were used especially at longer lead times. TLNMC showed positive impact on 4DEnsVar over SH and neutral impact over TR. For NH, slight positive impacts were found at early lead times and slight negative impacts were found after the 3-day lead time. References [1] Wang, X., D. Parrish, D. Kleist and J. Whitaker, 2013: GSI 3DVar-based Ensemble-Variational Hybrid Data Assimilation for NCEP Global Forecast System: Single Resolution Experiments, Monthly Weather Review, accepted. [2] Hamill, T., J. S. Whitaker, M. Fiorino, and S. J. Benjamin, 2011: Global ensemble predictions of 2009's tropical cyclones initialized with an ensemble Kalman filter. Mon. Wea. Rev., 139, 668-688. [3] Buehner, M., P. L. Houtekamer, C. Charette, H. L. Mitchell, and B. He, 2010: Intercomparison of Variational Data Assimilation and the Ensemble Kalman Filter for Global Deterministic NWP. Part I: Description and Single-Observation Experiments. Mon. Wea. Rev., 138, 1550-1566.


Data Assimilation for the Whole Atmosphere Model (WAM)

Houjun Wanga,b, Tim Fuller-Rowella,b, Rashid Akmaevb, Ming Hua, Daryl Kleistc, Miodrag Rancicc, Jun Wangc, Mark Iredellc, and John Derberc

a CIRES, University of Colorado, Boulder, Colorado, USA, Email: [email protected]

b NOAA, Space Weather Prediction Center, Boulder, Colorado, USA c NOAA, Environmental Modeling Center, College Park, Maryland, USA

The Gridpoint Statistical Interpolation (GSI) analysis system is used at NCEP for both regional and global operational data assimilations. It uses the 3-dimensional variational (3DVAR) analysis technique, with 4DVAR option and many more. The Developmental Testbed Center (DTC) support of the community GSI has expanded its community users, such as users of the Weather Research and Forecasting (WRF) model for regional numerical weather prediction applications. The GSI has features that make it flexible in adding state and control variables. GSI has also been used for chemical/aerosol and cloud data assimilation. Thus, GSI is a fairly versatile data assimilation system. WAM is an extension of NCEP’s Global Forecast System (GFS) model from 64 model levels (with the model top at about 60 km) to 150 model levels (with the model top at about 600 km). It covers the regions of important ionospheric processes and their variability. WAM includes basic ionospheric effects on neutral atmosphere, i.e., ion drag and Joule heating. Free annual run with WAM produced comparable climatology of tidal wave variability in the mesosphere and low thermosphere (MLT) region. GSI was extended for data assimilation for WAM. Incremental analysis update (IAU) [1] was implemented in the WAM-GSI data assimilation (DA) cycle. The first simulations with the WAM-GSI data assimilation system, or WDAS, have produced fairly realistic dynamic and electrodynamic responses to real large-scale sudden stratospheric warming (SSW) events [2]. In this talk, some on-going works with WDAS will be presented. First, the background error statistics are updated from yearlong short-term forecasts from the current DA system. Then, the effects of assimilating observations in the MLT regions, such as SABER temperature and TIDI wind data from the TIMED satellite will be assessed. Assimilation of other observational data, such as CHAMP temperature/density data, SSM/IS upper air channel radiance, will also be discussed. Assimilation and error/bias quantification of these new data sources provides new opportunities to validate WAM simulations and forecasts. Recently, a new NEMS version of WAM was implemented, and this provides new opportunity to test methodologies of data assimilation for the whole atmosphere modeling system. References [1] Bloom, S. C., L. L. Takacs, A. M. da Silva, and D. Ledvina. “Data assimilation using incremental analysis updates”. Monthly Weather Review, vol. 124, no. 6, pp 1256-1271, June 1996. [2] H. Wang, T. J. Fuller-Rowell, R. A. Akmaev, M. Hu, D. T. Kleist, and M. D. Iredell. “First simulations with a whole atmosphere data assimilation and forecast system: The January 2009 major sudden stratospheric warming”. Journal of Geophysical Research, vol. 116, A12321, doi:10.1029/2011JA017081, December 2011.

The 6th WMO Data Assimilation Symposium: Global and regional atmospheric DA

Reconnaissance Data Impact on Hurricane Intensity Prediction with the real-time WRF-EnKF System

Yonghui Weng ([email protected]) and Fuqing Zhang ([email protected])

Department of Meteorology, the Pennsylvania State University, University Park, PA 16802, USA An experimental next-generation hurricane prediction system based on the cloud-permitting Weather Research and Forecasting (WRF) model and an advanced data assimilation technique known as the ensemble Kalman filter (EnKF) has been developed at Penn State University (PSU). The system uses the EnKF to ingest high-resolution airborne Doppler radar observations of the hurricane’s inner core to provide better initial conditions to the WRF model, which in turn yields more accurate forecasts.The PSU WRF-EnKF hurricane analysis and prediction system had been operated in real-time for Atlantic storms for 5 years with airborne Doppler radar data assimilation and performed remarkably well for all landfalling hurricanes from 2008 through 2012: averaged over all 102 applicable airborne Doppler missions, errors in forecast intensity for lead times of 1 to 5 days were 15-43% less than the corresponding official forecasts issued by the National Hurricane Center. Since 1982, dropwindsondes were released from 2 WP-3D aircrafts, reconnaissance data had been become a significant data source in hurricane prediction. This study exams the impact on hurricane intensity prediction with the PSU real-time hurricane analysis and prediction system to breakthrough the limitation of the small TDR sample size. To establish a baseline level of hurricane intensity prediction accuracy, a control run without any Reconnaissance data assimilation by the cycling WRF-EnKF system is designed. Two experiments are designed: one is to assimilate flight-level and dropsonde observations, and another one is to assimilate flight-level, dropsonde and TDR observations. The preliminary result shows,

1) The control run with the cycling assimilation system largely reduces the system bias than the cold-started WRF forecasts initialized with GFS operational analysis;

2) The assimilation of fligh-level and dropsonde observations can reduce 5~15% error during the 12~96 h lead-time;

3) The assimilation with all inner-core observations including flight-level, dropsonde and TDR data can provide very realistic initial vortex, made the intensity bias very small and improve both the track and intensity prediction.

References F. Zhang, Y. Weng, J. A. Sippel, Z. Meng & C. H. Bishop, Cloud-resolving Hurricane Initialization and Prediction through Assimilation of Doppler Radar Observations with an Ensemble Kalman Filter: Humberto (2007). Mon. Wea. Rev., 137, 2105-2125 (2009). doi: 10.1175/2009MWR2645.1. F. Zhang, Y. Weng, J. F. Gamache & F. D. Marks, Performance of Convection-permitting Hurricane Initialization and Prediction during 2008-2010 with Ensemble Data Assimilation of Inner-core Airborne Doppler Radar Observations. Geophysical Research Letters, 38, L15810 (2011). doi:10.1029/2011GL048469. Y. Weng & F. Zhang, Assimilating Airborne Doppler Radar Observations with an Ensemble Kalman Filter for Convection-permitting Hurricane Initialization and Prediction: Katrina (2005). Monthly Weather Review, 140, 841-859 (2012). doi: http://dx.doi.org/10.1175/2011MWR3602.1.

Structure and Phase Error Correction of the First Guess with Hybrid 3DV-Ensemble Analysis

Wan-Shu Wua, David F. Parrisha and Mingjing Tongb

aNational Centers for Environmental Prediction, Environmental Modeling Center, Washington D.C.

[email protected], bI.M. Systems Group, Inc.

The forecast errors in the first guess include the position, the structure and the amplitude errors. Forecast error structures estimated from the model forecasts and/or previous observations are used to spread the analysis increments to the non-observed area. While this common practice has great success in analysis schemes, it is not effective in removing the structure and position errors. It is found that hybrid 3DV-ensemble analysis with differences between global ensemble mean and regional first guess as ensemble perturbation is effective in removing the structure error in the regional first guess. The same method is also used to remove the position error of the hurricanes. Extra ensemble members were added by locally shifting the first guess near the center of cyclones and the use of the hurricane central pressure observations are enhanced by adding bogus observations near the center of the cyclone.


For WMO Data Assimilation Symposium:

A Multigrid Technique in Advanced Data Assimilation

Yuanfu Xie1

Earth System Research Laboratory, NOAA, Boulder, CO , Alexander E. MacDonald, and Zoltan Toth

ABSTRACT Advanced data assimilation schemes combine information from various sources of observations, and background forecasts with knowledge about the dynamics of the systems analyzed. Modern numerical techniques are used to optimize the quality of the analysis and the efficiency of the computations. Multigrid techniques, for decades used successfully in solving numerical partial differential equations, provide an ideal platform for the advanced data assimilation systems. In this presentation, we will highlight a number of contributions multigrid techniques offer in advanced data assimilation applications, including a) use of ensemble-derived covariances in 3/4DVAR; b) improved ensemble localization; c) multiscale 4DVAR; d) computational efficiency. Some of the advantages of the multigrid techniques in data assimilation will be demonstrated through numerical experiments. A plan for a hybrid multi-scale data assimilation system for a Non-hydrostatic Icosahedral grid Model (NIM) will also be discussed.

1 Corresponding author contact information: Dr. Yuanfu Xie, NOAA/ESRL/GSD/FAB, 325 S. Broadway, CO. Email: [email protected]


A Regional GSI-Based Coupled EnKF-3DEnVAR Hybrid Data Assimilation

System for the Operational Rapid Refresh Forecasting System

Ming Xuea,b

([email protected]), Yujie Pana, Kefeng Zhu

a, Xuguang Wang

a,b, Ming Hu

c, Stanley G.

Benjaminc, Stephen S. Weygandt

c, and Jeffrey S. Whitaker

c

aCenter for Analysis and Prediction of Storms and

bSchool of Meteorology, University of

Oklahoma, Norman Oklahoma 73072, USA, [email protected] cNOAA Earth System Research Laboratory, Boulder, Colorado, USA

A coupled EnKF-3DEnVAR hybrid data assimilation system is developed for the U.S.

operational Rapid Refresh (RAP) forecasting system. The three-dimensional ensemble-

variational hybrid analysis system (3DEnVar) employs the extended control variable

method, and is built on the NCEP operational Grid-point Statistical Interpolation (GSI)

3DVar framework. It is coupled with a recently developed EnKF system for RAP [1]

which is used to update and provide to the 3DEnVar the ensemble perturbations.

Recursive filters are used to realize covariance localization in both horizontal and vertical

directions within the 3DEnVar.

The coupled hybrid system is evaluated with 3-hourly assimilation cycles over a 9-day

spring period containing active convection. All conventional observations used by

operational RAP GSI are included. To keep computational cost down for potential

operational implementation, the hybrid system is run at 1/3 of the operational RAP

resolution or about 40-km grid spacing, and its performance is compared to parallel GSI

and EnKF runs using the same data sets and resolution. Short-term forecasts initialized

from the 3-hourly analyses are verified against sounding and surface observations.

When equal weights are given to the static background error covariance and the flow-

dependent covariance derived from 40-member ensemble, the hybrid system outperforms

corresponding GSI and EnKF. When the recursive filter coefficients are tuned to achieve

a similar height dependency of vertical localization as in the EnKF, the hybrid results are

close to those of EnKF. With 20 ensemble members, EnKF, GSI and hybrid with half

ensemble covariance perform in ascending order, showing the advantage of the hybrid for

small ensembles. Two-way coupling between EnKF and hybrid did not show noticeable

improvement over one-way coupling for the testing period. Precipitation forecast skills

on the downscaled 13-km RAP grid from the hybrid analyses are better than those from

GSI analyses. Dual-resolution results where the hybrid EnVar analyses are performed on

the 13-km RAP grid will also be presented at the symposium.

References

[1] K. Zhu, Y. Pan, M. Xue, X. Wang, J. S. Whitaker, S. G. Benjamin, S. S. Weygandt, and

M. Hu, "A regional GSI-based ensemble Kalman filter data assimilation system for the

Rapid Refresh configuration: Testing at reduced resolution," Mon. Wea Rev., Accepted,

2013.


2D surface reanalysis over Europe at 5 km scale within EURO4M project: system description and validation

C.Socia, E. Bazile a, T. Landeliusb

a CNRM-GAME, Météo-France, France, [email protected]. b Swedish Meteorological and Hydrological Institute, Sweden

This poster provides a brief description of the European Reanalysis and Observations for Monitoring (EURO4M) project. It is a 4-year project which has been started in 2010 under the 7th Framework Programme, Theme 9 “Space”, and financed by the European Union. As mentioned in the Description of Work, “the overall goal of EURO4M project is to develop the capacity for, and deliver the best possible and most complete (gridded) climate change time series and monitoring services covering all of Europe.” The poster focuses on the improvements to the reanalysis of screen-level variables, and on the approach used to implement and validate an analysis of daily accumulated precipitation in the MESCAN system. The acronym MESCAN is a blending between MESAN ([1]) and CANARI (Météo-France operational surface analysis). MESCAN uses an optimal interpolation technique and refers to those parts of CANARI system which encompass the new implemented daily accumulated precipitation, more complex structure function and the error statistics for the screen-level variables from MESAN. The approach chosen to develop the precipitation analysis scheme is as in the Canadian Precipitation Analysis Project ([2]). It uses short-range forecasts as the background fields, and at this stage rain gauge data as observations. The scheme was initially developed such as to perform the precipitation analysis in the log-space on a transformed variable x=ln(RR+1), where RR is the 24-h accumulated precipitation expressed in millimetres. Nevertheless this transformed variable generates an negative bias which has more impact in hydrological application. Finally, the precipitation re-analysis over Europe will be carried out in the physical space (RR variable). The background field is a HIRLAM forecast downscaled from 22 km to 5.5km. The available surface observations over Europe have not a homogeneous distribution, especially for the 24h cumulated precipitation observation. For this parameter a common database was created by SMHI from 3 dataset provided by Météo-France, SMHI and ECA&D (KNMI). The preliminary results of the reanalysis over Europe will be shown. Validations will focus on regions with a high density of observations or complex topography (Alps, Scandinavia). Results of the comparisons of the MESCAN reanalysis with other reanalysis by MESAN, ERA-Interim will be also presented and discussed. References [1] L. Häggmark, K.-I. Ivarsson, S. Gollvik, and P-O. Olofsson. “Mesan, an operational mesoscale analysis system”, Tellus, 52A, 2-20. 2000. [2] J.-F. Mahfouf, B. Brasnett, and S. Gagnon. “A Canadian precipitation analysis (CaPA) project: Description and preliminary results”, Atm osphere-Ocean, 45:1, 1-17, 2007.

A Comparison Between 4D-Var and 4D-EnVar in the Canadian Regional Deterministic Prediction System

Jean-François Carona, Mark Buehnerb, Luc Fillionb, Thomas Milewskic and Judy St-Jamesc

a Data Assimilation and Satellite Meteorology Research Section, Environment Canada, Canada, jean-

[email protected], b Data Assimilation and Satellite Meteorology Research Section, Environment Canada, Canada, c Data Assimilation and Quality Control Development Section , Environment Canada,

Canada.

Over the recent years, Environment Canada (EC) has devoted important resources to investigate the feasibility of replacing its 4D-Var data assimilation scheme in the global deterministic prediction system (GDPS) by a computationally cheaper variational scheme (4D-EnVar) where background error covariances are represented by a blend of climatological covariances and 4D flow-dependent covariances derived from an EnKF-based global ensemble prediction system.

Following the positive results observed so far from 4D-EnVar in EC's GDPS, a similar effort was recently initiated in the regional deterministic prediction system (RDPS; limited-area domain with a 10-km grid spacing covering North America) which relies on a limited-area 4D-Var data assimilation scheme (operational since October 2012). Since there is currently no operational equivalent to the global EnKF at the regional scale at EC, we simply used the 4D ensemble covariances derived from the global EnKF as in the GDPS experiments. Results showed that a global-based 4D-EnVar scheme can provide RDPS forecasts slightly improved compared to the operational limited-area 4D-Var scheme, particularly during the first 24-h of the forecasts and in summertime convective regime where the lack of moist physical processes representation in our TL/Ad model impedes the performances of our 4D-Var scheme.


6th


The Impact of Community Radiative Transfer Model Microwave Sea Surface

Emissivity Improvements on Forecast Skill

David Neil Groffabc

, Quanhua Liucd

, Paul van Delstabc

, Andrew Collardabc

, Yanqiu Zhuabc

a NOAA National Centers for Environmental Prediction,

b I.M. Systems Group, Inc., USA,

[email protected], c Joint Center for Satellite Data Assimilation,

d University of Maryland, USA.

Observation operators are needed in data assimilation systems to map forecast model space to

observation space. In this regard, the accuracy of observation operators is important, because

aliasing of forecast model errors and errors from the observation operator will serve to degrade

analyses. For satellite-based radiance observations, the mapping to observation space requires

application of a radiative transfer model that can perform sufficiently accurate satellite-based

radiance simulations. Furthermore, the radiative transfer model must be sufficiently fast to

satisfy time constraints imposed on operational data assimilation systems. As such, the

Community Radiative Transfer Model (CRTM) is applied in the National Centers for

Environmental Prediction (NCEP) Gridpoint Statistical Interpolation (GSI) data assimilation

system.

A desirable feature that accompanies the latest CRTM, release 2.1.2, is a general improvement in

the accuracy of sea surface field of view simulations for surface sensitive microwave channels.

For example, when applied in the GSI, the variances and biases of first guess departures for

AMSUA surface sensitive channels have in general been reduced relative to the application of a

previous CRTM, release 2.0.5, over sea surfaces. This CRTM improvement is due to the

replacement of an older Fast Microwave Emissivity Model (FASTEM), FASTEM-1, with a

newer FASTEM, (FASTEM-5), in CRTM release 2.1.2. A particularly salient component of the

FASTEM-5 Geometrical Optics (GO) theory is the accounting of interactions with small-scale

sea surface waves [1]. This accounting of small-scale waves was absent in FASTEM-1. To

estimate the forecast impact of this CRTM improvement, parallel experiments were performed

using a 3dvar T254 configuration. The impact appears to be neutral for the Northern

Hemisphere, but positive in both the tropics and Southern Hemisphere. In this work, we describe

how reduction of model error aliasing in the GSI that can be attributed to application of the

aforementioned CRTM improvements, might explain the positive forecast impact.

References

[1] Q. Liu, F. Weng and S. English. "An Improved Fast Microwave Water Emissivity

Model", IEEE Transaction on Geoscience and Remote Sensing, vol. 49, no. 4, pp. 1238-1250,

April 2011.



Model Error Representation in Mesoscale WRF-DART Cycling Run

So-Young Ha, Chris Snyder, Judith Berner

National Center for Atmospheric Research Boulder, Colorado, U. S. A.

Mesoscale forecasts are strongly influenced by physical processes, from turbulence and mixing in the planetary boundary layer to moist convection and microphysics, that are either poorly resolved or must be parameterized in numerical models. Due to the model errors, mesoscale ensemble systems generally suffer from underdispersiveness that often leads to poor forecast skills. In an ensemble Kalman filter data assimilation, insufficient ensemble spread leads to poor filter performance in the analysis cycle. To alleviate the underestimate of ensemble spread, we compare two approaches to this issue: a multi-physics ensemble, in which each member's forecast is based on a distinct suite of physical parameterizations, and stochastic backscatter, in which small noise terms are included in the model equations for momentum and potential temperature. We perform our experiments in a domain over the continental U.S. using the WRF/DART system, which employs the Weather Research and Forecasting model for ensemble forecasts and the Data Assimilation Research Testbed for the ensemble Kalman filter. Verification against independent observations for a one-month summer period shows that including model-error techniques improves not only an ensemble of analyses, but also short-range forecasts started from these analyses. The stochastic backscatter scheme outperforms the multi-physics ensemble near the surface throughout the whole cycling period.

6th


Impact of Accounting for Aerosols in GEOS 3D-Var

Jong G. Kimab

, Arlindo da Silvaa, Dirceu Herdies

c, and

Ricardo Todling

a

a Global Modeling and Assimilation Office, NASA, USA, [email protected],

b Science System and

Applications, Inc., USA; c Centro de Previsao do Tempo e Estudos Climaticos, CPTEC, Brazil.

For quite sometime now the Goddard Earth Observing System (GEOS) general

circulation model has been using aerosols from the Global Ozone Chemistry Aerosol

Radiation and Transport (GOCART) to interact with its radiation physics component

(Colarco et al. 2010). More recently, a two-dimensional physical-space analysis of

aerosol optical depth assimilates AQUA and TERRA MODIS observations allowing for

real-time, three-hourly, improved GOCART aerosols. At present, the fifteen GOCART

aerosol tracers are not felt by the underlying GEOS atmospheric analysis system (the

Grid-point Statistical Analysis; GSI: Kleist et al. 2010). We have enabled GSI to take the

influence of aerosols into its observation operator, and the corresponding Community

Radiative Transfer Model (CRTM; Kleespies et al. 2004). Some preliminary studies in

the GEOS data assimilation system have shown mild, but noticeable, improvement in

temperature fields over large pockets of dust storms when the observation operator

calculations related to infrared channels, such as those from, say, HIRS and AIRS, are

allowed to feel the effect of the background aerosols.

This work will show results from expanding our preliminary studies and provide a more

complete assessment of the impact of explicitly accounting for aerosols in the

assimilation of all infrared channels currently used in GSI, including those from IASI

instruments.

References

[1] Colarco, P., A. da Silva, M. Chin and T. Diehl, 2010: Online simulations of global

aerosol distributions in the NASA GEOS-4 model and comparisons to satellite and ground-

based aerosol optical depth. J. Geophys. Res.}, 115, D14207, doi:10.1029/2009JD012820.

[2] Kleespies, T.J., van Delst, P., McMillin, L.M., and J. Derber, 2004: Atmospheric

transmittance of an absorbing gas. 6. OPTRAN status report and introduction to the

NESDIS/NCEP Community Radiative Transfer Model. Appl. Opt., 43, 3103-3109.

[2] Kleist, D. T., D. F. Parrish, J. C. Derber, R. Treadon, W.-S. Wu, and S. Lord, 2009:

Introduction of the GSI into the NCEP’s Global Data Assimilation System. {\it Wea.

Forecasting}, 24, 1691-1705..


Mesoscale data assimilation experiment with the NHM-LETKF

Masaru Kuniia

a Forecast Research Department, Meteorological Research Institute, Japan, [email protected]

The local ensemble transform Kalman filter (LETKF, [1]) is an EnKF scheme in which the algorithm achieves a high efficiency for parallel implementations. In this study, the LETKF is applied to the nonhydrostatic model of the Japan Meteorological Agency (NHM, [2]), and mesoscale data assimilation as well as ensemble forecast experiments are carried out. The LETKF was already implemented with the NHM [3], but the latest version of the LETKF is used here, including the adaptive inflation scheme [4] and the Gaussian-based localization scheme without local patches. In addition, the newly developed NHM-LETKF facilitates data assimilation experiments with one-way grid nesting in which the first guess of a coarser-resolution model is used as a boundary condition for finer-resolution model integration. Thus, the availability of practical applications for local severe weather forecasts can be enhanced by using this version of the NHM-LETKF. The NHM-LETKF is applied to a local severe rainfall event in Japan in 2012. Comparison of the root mean square errors between the model first guess and analysis reveals that the system assimilates observations appropriately. Forecasts initialized with LETKF analyses successfully capture intense rainfalls, indicating that the system works effectively for local severe weather. Investigation of probabilistic forecasts by ensemble forecasting shows this could become a reliable data source for decision making in the future. A one-way nested data assimilation scheme is also tested. The experiment results demonstrate that assimilation with a finer-resolution model provides an advantage in the quantitative forecasting of local severe weather conditions. References [1] Hunt, B. R., E. J. Kostelich, and I. Szunyogh, “Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter”, Physica D, 230, 112–126, 2007. [2] Saito, K., T. Fujita, Y. Yamada, J. Ishida, Y. Kumagai, K. Aranami, S. Ohmori, R. Nagasawa, S. Kumagai, C. Muroi, T. Kato, H. Eito and Y. Yamazaki, “The operational JMA nonhydrostatic mesoscale model”, Mon. Wea. Rev., 134, 1266-1298, 2006. [3] Miyoshi, T. and K. Aranami, “Applying a Four-dimensional Local Ensemble Transform Kalman Filter (4D-LETKF) to the JMA Nonhydrostatic Model (NHM)”, SOLA, 2, 128-131, 2006. [4] Miyoshi, T., “The Gaussian Approach to Adaptive Covariance Inflation and Its Implementation with the Local Ensemble Transform Kalman Filter”, Mon. Wea. Rev., 139, 2011.


Data Assimilation with FIM: results and prospects

Mariusz Pagowski, Jeffrey Whitaker, and Stan Benjamin

NOAA/ESRL, Boulder CO, USA, corresponding author: [email protected]

FIM (Flow-following Finite-volume Icosahedral Model, http://fim.noaa.gov) is being developed at NOAA/ESRL in Boulder. For the last four years, this global weather prediction model provided real-time forecasts allowing for an extensive evaluation and comparison with other models as FIM development has matured. Using the FIM model with existing NOAA/NCEP global data assimilation, ESRL has recently implemented a data assimilation system relies on the same components as the GFS model: the Gridpoint Statistical Interpolation (GSI, [1]) and a deterministic Ensemble Kalman Filter [2] used in a hybrid mode. In the presentation we will show performance of the system, discuss its shortcomings and outline future development. References [1] Wu, W.-S., R. J. Purser, and D. F. Parrish (2002), Three-dimensional variational analysis with spatially inhomogeneous covariances, Mon. Weather Rev., 130, 2905–2916, doi:10.1175/1520-0493(2002)130<2905:TDVAWS>2.0.CO;2. [2] Whitaker, J. S., and T. M. Hamill (2002), Ensemble data assimilation without perturbed observations, Mon. Weather Rev., 130, 1913–1924, doi:10.1175/1520-0493(2002)130<1913:EDAWPO>2.0.CO;2.

6th


Forecasting Air Temperature in Metropolitan Area of Lisbon using ARIMA models

Mónica Rodriguesa, Paula Santana

b

a Department of Geography, University of Coimbra, Portugal,

b Department of Geography, University of Coimbra, Portugal

Email address of the corresponding author:[email protected]

Air temperature, a temporal series, can be modeled using various techniques; among them the

autoregressive integrated moving average (ARIMA) models [1]. The aim of this modelling

approach is to express current time series values as a linear function of past time series values

(the autoregressive component) and current and lagged values of a white noise process (moving

average component) [2].

In this research the characteristics of temperature series will be analyzed in the Metropolitan Area

of Lisboa (MAL) and, a statistical model will be proposed by the use of using Box & Jenkins

methodology. For the temperature series the used data were the month average, from January

1950 to December 2010. For modeling by ACF and PACF methods, examination of values

relative to auto regression and moving average were made and at last, an appropriate model for

estimation of temperature values for MAL were found. The analysis of output forecast data of the

selected ARIMA (2,1,1)(0,0,0)12 model produced a highly accurate forecast. The model proposed

will be useful for future decisions.

References

[1] FC McMichael, JS Hunter. Stochastic modeling of temperature and flow in rivers,Water

Resources Research 8 (1), 87-98.

[2] G. E. Box, G. M. Jenkins. Time series analysis: forecasting and control, San Francisco,

Holden-Day, 1976.

http://scholar.google.com/citations?view_op=view_citation&hl=pt-PT&user=KMYlnhkAAAAJ&citation_for_view=KMYlnhkAAAAJ:IjCSPb-OGe4C

6th


Forecast and Analysis Verification of CPTEC/INPE G3DVAR

Fábio Luiz Rodrigues Diniza, Maurício Granzotto Mello

a, Dirceu Luis Herdies

a, Luis Gustavo Gonçalves

de Gonçalvesa, and José Antonio Aravéquia

a

a Center for Weather Forecast and Climate Studies (CPTEC), National Institute for Space Research

(INPE), Brazil, [email protected]

The Center for Weather Forecast and Climate Studies from the Brazilian National Institute for

Space Research (CPTEC/INPE) replaced its former data assimilation system, the Global

Physical-Space Statistical Analysis System (GPSAS) by the Global 3DVar (G3DVAR). The latter

is based on the Gridpoint Statistical Interpolation (GSI) jointly developed by NOAA, NASA and

NCAR, implemented on the Atmospheric Global Circulation Model developed at the

CPTEC/INPE (AGCM/CPTEC/INPE). The new system, operational at CPTEC/INPE since

January 2013, assimilates a variety of conventional and non-conventional observations every 6

hours. Since that, CPTEC/INPE trough its Group on Data Assimilation Developments (GDAD) is

evaluating the performance of the system. Currently, an effort to do this evaluation is being

conducted by forecast verification against a set of conventional observations. We will show how

G3DVAR differ from these observations by observation-minus-6hr forecast (O-F) and

observation-minus-analysis (O-A) statistics. We will also show a comparison of how these

differences are against others operational centers. The interest of this evaluation is to the

assessment of the quality improvements brought in by the recent change in data assimilation

method of the CPTEC/INPE data assimilation system.


The impacts of MODIS and SSM/I total precipitable water on Hurricane Sandy forecasts in regional NWP model

Pei Wanga,b, Jun Lia, Tim Schmitc, Jinlong Lia, Yong-Keun Leea and Wenguang Baia

a CIMSS, University of Wisconsin–Madison, WI, U.S.A. [email protected], b Department of Atmospheric and

Oceanic Sciences, University of Wisconsin-Madison, WI, U.S.A., c Advanced Satellite Products Branch (ASPB),

NOAA/NESDIS/STAR, U.S.A.

Moisture observations are important for tropical cyclone (TC) forecasts. Studies found that the hurricanes that rapidly intensified tended to exist within a moister large-scale environment than weaker storms. The impacts of total precipitable water (TPW) from Moderate Resolution Imaging Spectrometer (MODIS) infrared (IR), and Special Sensor Microwave/Imager (SSM/I) on simulations of Hurricane Sandy (2012) are assessed and compared using the regional numerical weather prediction (NWP) model - the advanced WRF (ARW) modeling system together with the Community Gridpoint Statistical Interpolation (GSI) assimilation system. The assimilation is conducted every 6 hours with conventional data, SSM/I and MODIS total precipitable water, followed by 72 hours forecasts. The comparison of MODIS and SSM/I TPW with model background shows that the SSM/I TPW is higher and moister than the first guess, while the MODIS TPW is lower and dryer. The observation minus background (O-B) and the observation minus analysis (O-A) indicate that the MODIS data fit better with the background. To verify the impacts of assimilating MODIS and SSM/I TPW, the hurricane track, minimum sea level pressure (SLP) and maximum wind speed observations from national hurricane center (NHC) are used as references for comparisons with forecasts. The 24 hour accumulated precipitation from forecasts is verified against NOAA CPC Morphing Technique (CMORPH) high resolution precipitation analysis. Both the frequency bias and the equitable threat score (ETS) are computed to show the impacts of TPW on hurricane precipitation forecasts over ocean. The GOES-13 Imager brightness temperature measurements are also compared with that of simulated from forecasts, indicating that the mesoscale features around the hurricane Sandy can be well captured by assimilating moisture information from satellite.


Hybrid ENKF Reanalysis of 1981-1982 Period for Comparison with CFSR Results

Jack Woollena, Daryl Kleistb

aIMSG/EMC/NCEP/NOAA, US, [email protected], bEMC/NCEP/NOAA, US

.

The 3DVAR GSI CFSR reanalysis encountered a number of problems in the assimilation of the relatively sparse datasets in the early 1980’s. In particular deficits were noted in the CFSR results having to do with QBO winds and tropical analyses in general, due in large part to limitations on the creation of forecast error structure functions for the GSI which would be effective for the data distribution found during that time period. Following the implementation of the hybrid-ENKF GSI analysis system in NCEP operations, some interest was developed for revisiting early 1980’s reanalysis using a low resolution version of the new operational system. A non-hybrid 3DVAR control was run at T254L64 resolution and a companion hybrid experiment run at the same resolution with 80 T126L64 members in the hybrid ensemble. These two experiments were started from CFSR initial conditions on 01July1981 and run for 10 months through 30April1982. Results from the two experiments are contrasted with those from CFSR by means of forecast and analysis fits to observations and anomaly correlations, and other features available in the NCEP VSDB diagnostic package, which will highlight the strengths and weaknesses of the three systems in the context of reanalysis in the early satellite period.


Storm Tracks and Low-Frequency Variabilities in AFES-LETKF Experimental Ensemble Reanalysis 2

Akira Yamazakia and Takeshi Enomotoa, b

a Earth Simulator Center, Japan Agency for Marine-Earth Science and Technology, Japan, [email protected], b Disaster Prevention Research Institute, Kyoto University, Japan.

An experimental ensemble atmospheric reanalysis dataset is being produced with a data assimilation system composed of the atmospheric general circulation model for the Earth Simulator (AFES) and the local ensemble transform Kalman filter (LETKF) [1]. The system has been updated from that produced ALERA (AFES-LETKF experimental ensemble reanalysis), hence the dataset is called ALERA2 [2]. The period of the ALERA2 is about five year, which is from 1 January 2008 to 31 December 2012, covered by two streams: one from 1 January 2008 and the other from 1 August 2010. Winter periods of the five-year-long dataset in Northern Hemisphere are used to investigate the relationship between storm tracks and low-frequency variabilities such as blocking. The investigation is especially focused on the behavior of the analysis ensemble spread [3]. The climatology of the ALERA2 in those periods is almost similar to an operational reanalysis dataset, the JRA-25/JCDAS [4], in the upper troposphere where the low-frequency variabilities are dominant. Also, storm tracks in these reanalysis datasets show good agreement. It is found that the large structure and time variation of the analysis ensemble spread are concentrated near the storm-track regions, especially on North and middle Pacific. Also, the time evolution of the spread is related to those of the storm track and the blocking-like pattern in the North Pacific region. References [1] T. Miyoshi, S. Yamane, and T. Enomoto. “The AFES-LETKF experimental ensemble reanalysis: ALERA.” SOLA, 3, 45-48, April 2007. [2] T. Enomoto, T. Miyoshi, Q. Moteki, J. Inoue, M. Hattori, A. Kuwano-Yoshida, N. Komori, and S. Yamane. “Observing-system research and ensemble data assimilation at JAMSTEC.” In S. K. Park and L. Xu, editors, Data Assimilation for Atmospheric, Oceanic and Hydrological Applications (Vol. II). Springer, 2013, in press. [3] T. Enomoto, M. Hattori, T. Miyoshi, and S. Yamane. “Precursory signals in analysis ensemble spread.” Geophysical Research Letters, 37, 8, doi: 10.1029/2010GL042723, April 2010. [4] K. Onogi and coauthors. “The JRA-25 Reanalysis.” Journal of the Meteorological Society of Japan, 85, 3, 369-432, August 2007.

Representing model uncertainty in data assimilation with stochastic physics

Jeffrey S. Whitakera, Philip Pegionb,c, and Thomas Hamillc

a NOAA Earth System Research Laboratory, Boulder, CO <[email protected]>, b CooperativeInstitute for Research in the Environmental Sciences, University of Colorado, Boulder, CO, c NOAA Earth

System Research Laboratory, Boulder, CO.

All data assimilation systems that use ensemble forecasts to estimate backgrounderrorcovariances require some method for representing missing sources of uncertainty, such as modelerror. The current NCEP operational hybrid 3d ensembleVar system uses adhoc inflation (bothmultiplicative and additive). In this talk I will discuss the possibility of replacing the additivecomponent with a stochastic representation of model uncertainty included in the forecast modelitself. Several different schemes, including vorticity confinement, stochastic kineticenergybackscatter, perturbed boundarylayer humidity and stochastically perturbed physics tendencieshave been tested. Results will be presented, both in the context of the 3d ensembleVar dataassimilation at reduced resolution, and in mediumrange ensemble forecasting at the currentoperational (T254L64) resolution.

6th WMO Symposium on Data Assimilation 2013


Testing and Evaluation of GSI and GSI-Hybrid Data Assimilation for

Regional Applications at the Developmental Testbed Center (DTC) Hui Shaoa, Chunhua Zhoua, Kathryn Newmana, Ming Hub,c, Mrinal K. Biswasa, Ligia Bernardetb,c,

Mingjing Tongd, John Derberd, and Jeff Whitakerb

aResearch Applications Laboratory (RAL), National Centers for Atmospheric Research (NCAR), USA, [email protected], bEarth System Research Laboratory (ESRL), National Oceanic and

Atmospheric Administration (NOAA), USA, cCooperative Institute for Research in the Environmental Sciences (CIRES), University of Colorado, USA, dEnvironmental Modeling Center

(EMC), National Centers for Environmental Prediction (NCEP), USA.

The Developmental Testbed Center (DTC) is a national distributed facility to serve as a bridge between research and operations and facilitate the activities of the Numerical Weather Prediction (NWP) community. Since 2009, the DTC has been providing the research community access to and support of the operational Gridpoint Statistical Interpolation (GSI) system and GSI based hybrid ensemble-variational system (GSI-hybrid). The DTC is also making an effort to provide objective evaluation of these systems to the research community and operational centers, focusing on the regional applications currently, though extensive testing experiments. This paper will give a summary of the latest GSI and GSI-hybrid testing activities conducted by the DTC. These tests include the pre-implementation tests and evaluation of the GSI configuration and associated tuning and the forecast impact study of regional background errors. Also included are tests conducted using the GSI-hybrid system for tropical storm forecasts. Issues uncovered and lessons learned during the testing and evaluation process will be discussed. Some system/technique comparison results (e.g., variational versus hybrid) will also be presented and discussed.

6th


Deterministic analysis in the CNMCA-LETKF system

Lucio Torrisi

a and Francesca Marcucci

b

a CNMCA-Italian National Meteorological Center, Rome, Italy, [email protected],

b CNMCA-Italian

National Meteorological Center, Rome, Italy

Many studies have clearly demonstrated the benefit of using an hybrid EnKF-variational data

assimilation system. This would imply additional costs for a small operational centre that strives

to have a single data assimilation (DA) system to develop, maintain, and run operationally.

Recently the Italian National Meteorological Center (CNMCA) has moved from a variational

(3D-VAR) to an ensemble based (LETKF [1] ) data assimilation system because this approach

provides “optimal” analysis errors estimates for ensemble forecasting and it overcomes the need

for ad hoc inverse methods. However, the use of the ensemble mean, that is the best estimate of

the true system state, as initial condition for the operational deterministic NWP system, smoothes

out the short range forecast, because the average operator tends to filter out the low-predictable

small scale features in the analysis.

The CNMCA has spent some effort in deriving a deterministic analysis from the LETKF, in order

to avoid the expensive solution of an hybrid DA system. Along with the standard CNMCA-

LETKF [2,3] state vector, a deterministic analysis is computed using the standard LETKF-

Kalman gain, the deterministic short-range forecast and the corresponding innovation vector. The

use of a deterministic state in the CNMCA system improves the first forecast hours with respect

to the use of the mean state. The deterministic forecast could also run at a higher resolution than

the LETKF one by interpolating the forecast perturbations. Different approaches to produce the

deterministic background, including also the blending technique, have been investigated.

References

[1] Hunt, B. R., E. Kostelich, I. Szunyogh. " Efficient data assimilation for spatiotemporal

chaos: a local ensemble transform Kalman filter," Physica D, 230, 112-126, 2007.

[2] Bonavita M, Torrisi L, Marcucci F. “The ensemble Kalman filter in an operational

regional NWP system: Preliminary results with real observations”, Q. J. R. Meteorol. Soc.,

134, 1733-1744, 2008.

[3] Bonavita M, Torrisi L, Marcucci F. ”Ensemble data assimilation with the CNMCA

regional forecasting system”, Q. J. R. Meteorol. Soc., 136, 132-145, 2010

Towards a Longer Assimilation Window in the IFS 4D-Var

Yannick Tr é molet a

a ECMWF, UK, [email protected]

Several studies in recent years have shown that there are theoretical benefits to the use of a weak-constraint formulation and a long assimilation window in 4D-Var. We will focus in this presentation on the strategy to increase the length of the assimilation window in ECMWF's operational forecasting system, the IFS. We will present the challenges in implementing long window weak-constraint 4D-Var, both technical and scientific.

Most operational 4D-Var implementations assume that the model being used is accurate enough that model error can be neglected over the length of the assimilation window. This assumption is one of the factors that limits the length of the assimilation window in practice, it is removed in weak-constraint 4D-Var. The full implementation of weak-constraint 4D-Var implies a four dimensional control variable and requires significant technical changes in the system. This will be achieved in a new framework: the Object-Oriented Prediction System (OOPS).

Some aspects of long-window weak constraint 4D-Var, such as convergence issues, preconditioning and nonlinearity are already being studied in the context of OOPS (see talk by M. Fisher).

At the same time, other aspects are addressed in the current system. A simpler implementation of weak-constraint 4D-Var, aimed at estimating the systematic component of model error has been implemented in operations in 2009. From this initial implementation, experience has been gained in estimating the model error covariance matrix and in increasing progressively the assimilation window length. A 24-hour assimilation window is already being used in the context of the 20 th

century reanalysis and is planned for operational implementation at ECMWF by the end of 2013. Pre-operational testing results will be shown with the full observing system and at full resolution.



GSI-based Hybrid Data Assimilation for NCEP GFS: How Is the Dual Resolution Hybrid Compared to the Single Resolution Hybrid?

Ting Leia and Xuguang Wangb

abSchool of Meteorology and Center for Analysis and Prediction of Storms, University of

Oklahoma, USA. Corresponding author email: [email protected]

A 3DVar-based ensemble-variational (3DEnsVar) hybrid data assimilation system was recently developed based on the Gridpoint Statistical Interpolation (GSI) data assimilation system and was implemented operationally for the GFS. In the operational implementation, the hybrid was run with a dual resolution configuration where the ensemble was run at a reduced resolution as compared to the control analysis and forecast. How is the performance of the hybrid run at dual resolution compared to that run at single resolution where the ensemble was run at the same resolution of the control? Experiments were conducted over a 4-week period during the 2011 summer. As a first step of answering such question, the experiments were conducted with a lower resolution than those in operations. It was found that if the static covariance was not included, the dual resolution hybrid performed significantly worse than the single resolution hybrid. However, when the static covariance was included, the degradation of the dual resolution hybrid was much reduced and nearly negligible. Series of diagnostics were conducted to further understand the performance of the dual resolution and single resolution hybrid DA systems. Diagnosing the spread-error relationship as a function of height using the innovation statistics revealed that the ensemble spread was under-dispersive, especially in low levels, and such under-dispersiveness was more severe for the dual resolution hybrid than the single resolution hybrid. Including the static B improved the spread-error relationship especially for the dual resolution hybrid. Diagnostics were also conducted in the spectral space. The first guess ensemble spreads from the single resolution and dual resolution hybrid and those inferred from the static covariance were compared. For example, for meridional wind, the vertically averaged spectra of the ensemble spreads peak at about total wavenumber 20, while those derived from the static covariance peaks at a larger total wavenumber, approximately, of 30. The spread spectra derived from the static covariance are obviously higher than the ensemble spreads for most wave numbers considered. The ensemble of the dual resolution hybrid has smaller spread than the single resolution hybrid for wavenumbers greater than about 25. After the static covariance was combined with the ensemble covariance, such differences of the first guess spreads between the single and dual resolution hybrid were reduced especially for those wavenumbers close to the truncation. The cost of the dual resolution hybrid is much cheaper compared to the single resolution hybrid. What is the tradeoff between increasing ensemble size and ensemble resolution? Would increasing the ensemble size in the dual resolution hybrid improve the performance of the hybrid? Preliminary analysis of a dual resolution experiment quadrupling the ensemble size revealed that the performance of the single resolution hybrid did not significantly improve with increasing ensemble size.

improved initialization and prediction of clouds in ... · canada, canada, cdata assimilation and...

Documents