220 likes | 330 Vues
This presentation explores low-dimensional modeling techniques, focusing on neutral density and the driver-response relationship using EOF (Empirical Orthogonal Function) analysis and Ensemble Kalman Filtering with TIE-GCM. Led by Tomoko Matsuo and her research team, it discusses the assimilation of ionospheric data using CHAMP satellite observations from 2001-2008. Key results include improved density modeling at 400 km altitude and implications for understanding atmospheric dynamics and forecasting through innovative statistical methodologies.
E N D
VII Driver-Response Relationships Tomoko Matsuo (CU) Low dimensional modeling of neutral density Gary Bust (ASTRA) Inference of thermospheric parameters from ionospheric assimilative maps
Lowand HighDimensionalModeling of Neutral Density PRESENTED BY: Tomoko Matsuo (CU) (a) EOF-based reduced-state modeling using CTIPe and CHAMP Suzzane Smith (REU student), Mariangel Fedrizzi (CU), Tim Fuller-Rowell (CU), Mihail Codrescu (NOAA), Jeff Forbes (CU), & Jiuhou Lei (CU) (b) Ensemble Kalman filtering with TIE-GCM Jeff Anderson (NCAR) & DART developers HAO TIEGCM developers
CTIPe and CHAMP By Courtesy of Mariangel Fedrizzi
CTIPe EOFs: 2005Singular value decomposition Diagonalize a sample covariance by SVD
Sequential non-linear regression analysis of CHAMP data Mean at 400km (2001-2008) 3-deg averaged CHAMP data normalized to 400 km using NRLMSISe00 8 years (2001-2008) precession through local time once every 133 days [Sutton et al., 2007] For pth EOF, minimize With orthonormal constraint for 2
CHAMP EOFs: 2001-2008Sequential non-linear regression [Matsuo and Forbes, 2010]
Density Modeling with CTIPe EOFs (2/3) EOF-based regression model CHAMP EOF
Driver-response relationshipin terms of EOF From CHAMP 2001-2008 [Matsuo and Forbes, 2010] EOF Modes for CIR, CME, northward IMF?
: forward model Ensemble Kalman filtering (1/3) DART Observations sparse & irregular *GCM high-dimension Data Assimilation Research Testbed http://www.image.ucar.edu/DARes/DART TIEGCM 1.93 http://www.hao.ucar.edu/modeling/tgcm/download.php
Ensemble Kalman filtering (2/3) Model Error Growth t-1 t t+1 Forecast Step Use samples!! Initial distribution forecast distribution
Ensemble Kalman filtering (3/3) t-1 t t+1 Update Step Forecast Distribution Posterior Likelihood Prior Prior
Observing System Simulation Experiment http://www.image.ucar.edu/DARes/DART • Deterministic Filter: [Anderson, 2002] • Experiment Period: Day 87-91 Year 2002 • Observation: “CHAMP density” sampled from “Truth” • with centered Gaussian random error • Assimilation cycle: ~90-min (one orbit) • Number of ensemble member: 96 • Localization: Gaspari and Cohn in horizontal • Spin-up time: 2 weeks with perturbed forcing (F10.7 & cross-polar cap potential/HPI) Strongly forced and Dissipative system stochastic forcing
Posterior Mean - Prior Mean level 22 ~ 400km Posterior Mean pressure level 22 ~ 400km level 18 ~ 300km
Posterior Mean - Prior Mean Zonal Wind Meridional Wind (level 22 ~ 400km)
Driver-response relationshipin terms of ensembles -42.5 lat & -135 lon level 22 ~ 400km F10.7 CPCP Temperature O mixing ratio O2 mixing ratio
Summary State correction via assimilation of density data Driver estimation is a key for improvement (a) EOF-based reduced-state modeling using CTIPe To-Do: Driver-Response Relationship in terms of EOFs Product: EOF-based empirical density model at 400km Real-time CTIPe + EOF-based density correction • (b)Ensemble Kalman filtering with TIE-GCM To-Do: Driver Estimation in EnKF framework, Assimilation of ground-/space-based GPS, OSSE with Champ and Grace Product: “OSEE tested” Data assimilation system using a thermosphere-ionosphere general circulation model (TIEGCM) Reanalysis DA data might be useful…
Reduced-state modeling using EOFs Reconstruction of orbit-averaged density using EOFs Champ 4EOFs+Mean 2001 2002 2003 2004 2005 2006 2007 EOF-based regression model