The HBM-PDAF assimilation system
for operational forecasts in the North and Baltic Seas
Lars Nerger, Svetlana Losa
Alfred Wegener Institute for Polar and Marine Research Bremerhaven, Germany
Thorger Brüning, Frank Janssen
Federal Maritime and Hydrographic Agency (BSH) Hamburg, Germany
EuroGOOS 2014, Lisbon, Portugal, October 28 – 30, 2014
Lars Nerger et al. – HBM-PDAF Assimilation System
BSSC 2007, F. Janssen, S. Dick, E.
Kleine!
Grid nesting:
- 10 km grid - 5 km grid - 900 m grid
Starting point: Operational BSH Model (BSHcmod), V4
Lars Nerger et al. – HBM-PDAF Assimilation System
BSSC 2007, F. Janssen, S. Dick, E.
Kleine!
Grid nesting:
- 10 km grid - 5 km grid - 900 m grid
Starting point: Operational BSH Model (BSHcmod), V4
Goal:
• improve forecast skill of operational model using ensemble data assimilation
• started with BSHcmod
• updated to HBM for future operational use
Lars Nerger et al. – HBM-PDAF Assimilation System
single program
state time
state
observations
mesh data
Indirect exchange (module/common) Explicit interface
Model
initialization time integration post processing
Filter
Initialization analysis re-initialization
Observations
obs. vector obs. operator
obs. error
Core of PDAF
Logical separation of assimilation system
Nerger, L., Hiller, W. Software for Ensemble-based DA Systems –
Implementation and Scalability. Computers and Geosciences. 55 (2013) 110-118
Lars Nerger et al. – HBM-PDAF Assimilation System
Extending a Model for Data Assimilation
Aaaaaaaa Aaaaaaaa aaaaaaaaa
Start
Stop Do i=1, nsteps
Initialize Model
generate mesh Initialize fields
Time stepper
consider BC Consider forcing
Post-processing
Aaaaaaaa Aaaaaaaa aaaaaaaaa
Start
Stop Do i=1, nsteps
Initialize Model
generate mesh Initialize fields
Time stepper
consider BC Consider forcing
Post-processing
Model Extension for
data assimilation
Enable ensemble forecast using parallelization
Aaaaaaaa Aaaaaaaa aaaaaaaaa
Start
Stop
Initialize Model
generate mesh Initialize fields
Time stepper
consider BC Consider forcing
Post-processing init_parallel_pdaf
Do i=1, nsteps init_pdaf
assimilate_pdaf
Aaaaaaaa Aaaaaaaa aaaaaaaaa
Start
Stop
Initialize Model
generate mesh Initialize fields
Time stepper
consider BC Consider forcing
Post-processing init_parallel_pdaf
Do i=1, nsteps init_pdaf
assimilate_pdaf assimilate_pdaf
plus:
Adapt writing of sponge files
Lars Nerger et al. – HBM-PDAF Assimilation System
• Model and observation specific operations
• Elementary subroutines implemented like model routines
• Called by PDAF routines through a defined interface
Link to model
! initialize model fields from state vector
! initialize state vector from model fields Observation handling
! application of observation operator H to some vector
! initialization of vector of observations
! multiplication with observation error covariance matrix
User-supplied routines (call-back)
Lars Nerger et al. – HBM-PDAF Assimilation System
PDAF: A tool for data assimilation
PDAF - Parallel Data Assimilation Framework
" provide support for ensemble forecasts
" provide fully-implemented filter algorithms
" easily useable with (probably) any numerical model
(coupled also to NEMO, MITgcm, FESOM, ADCIRC)
" makes good use of supercomputers
" separate development of model and assimilation
algorithms
Open source:
Code and documentation available at http://pdaf.awi.de
L. Nerger, W. Hiller, Computers & Geosciences 55 (2013) 110-118
Lars Nerger et al. – HBM-PDAF Assimilation System
Assimilated Data – Satellite, MARNET and profiles
11. Oct. 2007 MARNET
Satellite data: NOAA SST Scanfish and CTD: T,S profiles MARNET: T and S time series
! Surface temperature: 12-hour composites
! Strong variation of data coverage (clouds)
Lars Nerger et al. – HBM-PDAF Assimilation System
Assimilation Methodology
! 12-hour forecast/analysis cycles
! Ensemble size 8 (sufficient for good results)
! Assumed data errors:
SST: 0.8oC (gave best results) MARNET: 0.5oC, 0.5 psu
Scanfish: 0.8oC, 0.5 psu
! Ensemble Kalman filter (local SEIK)
! Localization:
• Influence radius 100 km (tuned)
• Weight on data errors
(Exponential, e-folding at 100 km)
! Showing mainly results from BSHcmod (very similar to HBM)
Lars Nerger et al. – HBM-PDAF Assimilation System
Deviation from NOAA Satellite Data
! RMS errors for SST in 12-hour forecasts
! Average over 1 year (10/2007 – 9/2008)
! 23% overall reduction
! Mean error also reduced
No assimilation SST assimilation
Lars Nerger et al. – HBM-PDAF Assimilation System
ACCEPTED MANUSCRIPT
ACCEPTED MANUSCRIPT
16/10/07 17/10/07 18/10/07 19/10/07 20/10/07 21/10/07 22/10/07
0.4 0.5 0.6 0.7 0.8 0.9 1
date
o C
RMS error evolution
Model without DA LSEIK forecast LSEIK analysis 120h LSEIK forecast
Figure 7: RMS error temporal evolution over the period 16 October 2007 – 21 October 2007 for simulated SST without DA (black curve); LSEIK analysis (red); mean of ensemble forecast based on 12-hourly analysis (blue) and 5 days forecast (green curve) initialized with the analysis state obtained on 16 October 2007.
38
Forecast improvements
black: free model run
blue/red: 12h assimilation/
analysis cycles green: 5 day forecast
from 16/10
➜ Very stable 5-day forecasts
RMS error for SST over time
Lars Nerger et al. – HBM-PDAF Assimilation System
Red: Assimilation 12h forecasts
Validation with independent data (only SST assim.)
MARNET station data
• Reduction of
• Bias
• RMS error
Error estimates:
Bias: -0.55 -0.17 RMSe: 1.27 0.81
RMSe bias
free 0.87 0.3
satellite data 0.59 0.11 assimilation 0.55 0.08
1 year mean over 6 stations:
Lars Nerger et al. – HBM-PDAF Assimilation System
Assimilation of MARNET data
! Salinity: Significant improvement at surface and bottom
! Localization parameters influence assimilation performance
11/10/07 21/10/07 31/10/07 10/11/07 20/11/07 30/11/07 10/12/07 20/12/07 30/12/07 09/01/08 10
15 20 25
S/psu
date
Marnet station Fehmarn Belt: Surface salinity
11/10/07 21/10/07 31/10/07 10/11/07 20/11/07 30/11/07 10/12/07 20/12/07 30/12/07 09/01/08 10
15 20 25
date
S/psu
Marnet station Fehmarn Belt: Bottom salinity
Marnet data
BSHcmod without DA
exp LSEIK MAR forecast, lr = 10gp exp LSEIK MAR forecast, lr = 0/4gp
Lars Nerger et al. – HBM-PDAF Assimilation System
Summary
Ongoing work
! Include coastal mesh for assim. (900m resolution)
! Include ecosystem model ERGOM (@BSH)
! Assimilation of clorophyll data
! Switch to ESTKF filter (Nerger et al., MWR, 2012)
! HBM-PDAF system provides improved forecasts
! Very small increase of run time
! Assimilation framework PDAF to implement assimilation systems (http://pdaf.awi.de)
Lars.Nerger@awi.de
Lars Nerger et al. – HBM-PDAF Assimilation System
References
Nerger, L., Hiller, W. (2013) Software for Ensemble-based Data Assimilation Systems – Implementation and Scalability.
Computers and Geosciences. 55, 110-118
Losa, S.N. et al. (2012). Assimilating NOAA SST data into the BSH
operational circulation model for the North and Baltic Seas: Inference about the data. Journal of Marine Systems, 105-108, pp. 152-162
Losa, S. N. et al. (2014). Assimilating NOAA SST data into the BSH operational circulation model for the North and Baltic Seas: Part 2.
Sensitivity of the forecast's skill to the prior model error statistics.
Journal of Marine Systems, 129, 259-270
http://pdaf.awi.de www.demarine.de www.data-assimilation.net