JCSDA Science Workshop on Satellite Data Assimilation June 5-7, 2013, NCWCP, College Park, MD Utility of GOES-R ABI and GLM instruments in regional data assimilation for high-impact weather Milija Zupanski Cooperative Institute for Research in the Atmosphere Colorado State University 1

Programs: GOES-R Risk Reduction JCSDA NSF Collaboration in Mathematical Geosciences (NSF-CMG) People: - Karina Apodaca (CIRA), Man Zhang (CIRA), Lous Grasso (CIRA) - Mark DeMaria (NOAA/STAR), John Knaff (NOAA/STAR) - Min-Jeong Kim (CIRA-NOAA/EMC) - Jun Li (CIMSS, Univ. Wisconsin) High-End Computing: - NOAA S4 (SSEC, Univ. Wisconsin) - NCAR CISL (Yellowstone) Acknowledgements 2

 Objectives: - Assess the benefit of future GOES-R observations (ABI and GLM) - Use realistic NOAA systems - Conduct necessary development to maximize the benefit of GOES-R  The research encompasses: - assimilation of cloudy radiances (infrared and microwave) - cloud-resolving data assimilation applications to high-impact weather - regional hybrid variational-ensemble data assimilation  The following components are incorporated:  Weather Research and Forecasting Model (WRF): - WRF-NMM, WRF-ARW, hurricane WRF (HWRF)  Observation operator: - NCEP Gridpoint Statistical Interpolation (GSI) - Community Radiative Transfer Model (CRTM)  Data assimilation: - Maximum Likelihood Ensemble Filter (MLEF) - Use of regional hybrid GSI in the future Research 3

Severe weather  Thunderstorms  Tornadoes  Rainfall  Hail  Flash floods High-impact weather and clouds 4 GOES-R Wide-spread supercell thunderstorms – indicated by clouds Moore tornado (05/20/2013)

High-impact weather and clouds 5 Tropical cyclones  High winds  Storm surge  Rainfall  Floods  Tornadoes  Rip currents Improving assimilation and prediction of clouds is critical for reducing uncertainty of high-impact weather Hurricane Sandy (2012)

Nonlinearity (and non-differentiability)  Microphysical processes  Radiative Transfer (RT) model Forecast error covariance  Flow-dependent, cross-variable correlations, microphysics, dynamics Computational limitations  High-dimensional state vector for cloud-resolving data assimilation  Required additional RT model calculations Assimilation of clouds for high-impact weather 6 GOES-R Clouds are observed by satellites  Microwave (MW), Infrared (IR)  Lightning (future GLM): Signature of high-impact weather and clouds Cloudy satellite radiances are generally not assimilated in weather operations: Potentially useful information about clouds is mostly discarded. Can we do better?

Research scope and tools 7  MW: precipitation-affected radiances (microphysics/precipitation)  IR: cloud-affected radiances (cloud top/location)  Lightning flash rate (cloud ice, storm environment)  AIRS SFOV vertical profiles of q and T (storm environment) Assimilate observations that can add new information about clouds and the storm environment  hurricane WRF (HWRF)  WRF-NMM Regional data assimilation of tropical cyclone core area and severe weather Data assimilation components  Maximum Likelihood Ensemble Filter (MLEF)  Forward GSI  Forward CRTM

 A hybrid between EnKF and variational methods -iterative minimization (variational) -multiple realizations of model and observation operators for uncertainty (ensemble)  Deterministic control forecast  Nonlinear analysis solution by an iterative minimization  Improved minimization efficiency by an implicit Hessian preconditioning Data assimilation algorithm: Maximum Likelihood Ensemble Filter (MLEF) 8 Fast convergence (1-2 iterations) from an arbitrary initial state

MLEF is modular 9 Observation module: -Transform/interpolate from model to observations -Read observations -Output observation info: increments, errors, … Forecast module: -Import the forecasting system with pre- and post-processing -Make namelist.input on-the-fly Data assimilation module: -Controls the processing of the model and observation info -Hessian preconditioning, gradient, minimization iterations -State vector and uncertainty estimates

Since the singular values  are by-product of assimilation, the flow-dependent  H and DFS can be computed In ensemble DA methods  H and DFS can be computed exactly in ensemble subspace: Gaussian pdf greatly reduces the complexity since entropy is related to covariance Change of entropy due to observations Entropy 10 Assessment: Quantifying satellite information using Shannon information measures Change of entropy Degrees of freedom for signal (DFS)

All-sky microwave radiance assimilation in Tropical Cyclone core area Model: NOAA HWRF (operational in 2011, 27km/9km) Results for TC core area (inner nest) at 9 km resolution Observations: AMSU-A all-sky radiances, Channels 1-9 and 15 assimilated Data assimilation interval: 6 hours Number of ensembles: 32 Hurricane Daniele (2010) Control variable: pd, T, q, u, v, cwm Bias correction from clear-sky GSI output From M. Zhang et al. (2013a, Mon. Wea. Rev.) 11

12 Hurricane Danielle (2010): All-sky AMSU-A information content (DFS) Cycle 1Cycle 3Cycle 5Cycle 7 ASR CSR Cloudy radiance observations add new information throughout the storm development

(a)IR imagery (b)AMSU-A retrieved precipitation rate (a)CTL – control experiment (b)ASR – all-sky experiment 13 Hurricane Danielle (2010) 6-hour forecast of total cloud condensate Data assimilation cycle 8 – valid 1200 UTC 26 Aug 2010 (a) (c) (b) (d) All-sky MW radiances improve short-range forecast of clouds in hurricane core area

Hurricane Danielle (2010): Time series of the minimum sea level pressure (hPa) for NHC best track data (thick grey line) and MLEF-HWRF experiments ASR (solid) and CSR (dashed) between 1800 UTC 24 Aug and 1800 UTC 26 Aug Hurricane Danielle (2010): Intensity

All-sky infrared radiance assimilation for Tropical Cyclone core area Model: NOAA HWRF (operational in 2011, 27km/9km) Results for TC core area (inner nest) at 9 km resolution Observations: SEVIRI all-sky radiances [10.8  m - proxy for GOES-R ABI) Data assimilation interval: 1 hour Number of ensembles: 32 Hurricane Fred (2009) No bias correction (advantage of clear-sky GSI correction not obvious) From M. Zhang et al. (2013b) 15

Hurricane Fred (2009): Analysis 16 All-sky radiance assimilation Control experiment Verification: AMSU-A NOAA-16 retrieved cloud liquid water Assimilation of all-sky infrared radiance is able to improve clouds in TC core (a) (b)(c) (a) (b)(c) Total cloud condensate (cwm) Valid 0600 UTC 9 Sep 2009

Hurricane Fred (2009): 21-hour forecast 17 (a) (b)(c) (a) (b)(c) Total cloud condensate (cwm) Valid 0300 UTC 10 Sep 2009 All-sky radiance assimilationControl experiment Verification: Seviri radiance observations Assimilation of all-sky infrared radiance improves the forecast of clouds in TC core area

-Hurricane Fred (2010) -No data thinning for SFOV profiles Degrees of freedom for signal - DFS MSG SEVIRI only MSG SEVIRI and AIRS q profile MSG SEVIRI and AIRS T profile In outer domain (with less clouds) DFS shows a benefit from AIRS SFOV specific humidity and temperature data (temperature has a stronger impact) Hurricane Fred (2009): Assimilation of all-sky SEVIRI and AIRS SFOV (q,T) in HWRF outer domain 18

Lightning data assimilation: Severe weather applications Model: NOAA WRF-NMM (27km/9km) Results for the inner nest at 9 km resolution Observations: WWLLN [proxy for GOES-R GLM) Data assimilation interval: 6 hours Number of ensembles: 32 Tornado outbreak over Southeast US in April 2011 From Apodaca et al. (2013) 19

Model domain and tornado reports for April 27, 2011 Severe weather outbreak over the southeastern US on April 25-18, D01 D02

Lightning observation operator WWLLN can observe only cloud-to-ground (C-G) flashes GLM will observe both C-G and intra-cloud flashes Regression between lightning flash rate and model variables - Best regression suggests cloud ice and vertical graupel flux (McCaul et al. 2009) WRF-NMM microphysics (Ferrier) does not predict cloud ice: - Need to rely on less accurate regression: maximum vertical updraft Present: - Use max vertical updraft and WWLLN Future: - Include more complex microphysics to improve obs operator - Use better GLM proxy observations (C-G and intra-cloud) - Increase the resolution to 1-3 km 21

 Time-dependent covariance P f shows direct relationship to obs. 2011_04_27-00:00:00 Cycle _04_27-12:00:00 Cycle _04_28-00:00:00 Cycle 5 WWLLN Observations 22 Information content of lightning observations: DFS Flow dependent information added by assimilating lightning data

Lightning data assimilation with MLEF: Single observation experiment 23 Wind increment Temp increment Psfc increment Assimilation of lightning observations impacts all model variables and improves storm environment conditions: important for maintaining impact of assimilation Analysis response to a single observation of flash rate in a 6-hour interval

Possible new research directions (1) Combine all observations in applications to TC/severe weather - All-sky infrared radiances(GOES-R ABI) - Lightning (GOES-R GLM) - All-sky microwave radiances - AIRS/IASI (sounder) - NOAA operational observations (GSI) (2) Extend the utility of GOES-R instruments to air quality - Impact of lightning on NOx and O3 (air quality) - Use ABI and GLM for high-impact weather and AQ (3) Examine the impact of WV channels for TC genesis - channel difference contains important information (J. Knaff) (4) Further development of hybrid variational-ensemble systems - hybrid forecast error covariance with optimal Hessian preconditioning 24

Hybrid data assimilation Extend MLEF to include static/variational and ensemble error covariances - Hybrid ensemble-variational error covariance with optimal Hessian preconditioning - Single DA system (no separate variational and ensemble algorithms) 25 ensstatic Potential improvements over the existing hybrid DA systems: - Optimal Hessian preconditioning (e.g., observation term included) - Considerable computational savings: adaptive ensemble forecasting - Single DA system is easier to maintain than two systems (Var and EnKF)

Potential relevance to operations (1) Utilize the development of the MLEF regional hybrid data assimilation for all-sky radiances and GLM to improve hybrid GSI - observation processing and QC is same (GSI) - evaluated in realistic situations using NOAA systems - reduced code transfer/development - explore the utility of MLEF ensembles in regional hybrid GSI (2) Add GLM assimilation capability to operational DA system - New observations accessed through GSI and CRTM - BUFR format (3) Regional assimilation of combined GLM and all-sky ABI observations for warn-on-forecast and hurricane prediction - Results show relevance of these observations for prediction of clouds - Could be used in combination with hybrid GSI or as a standalone system for high-resolution nest (e.g., TC core, WRF-NMM inner nest) 26

Future work Combine all observations in applications to TC/severe weather - All-sky infrared radiances(GOES-R ABI) - Lightning (GOES-R GLM) - All-sky microwave radiances - AIRS/IASI (sounder) - NOAA operational observations (GSI) Examine the impact of WV channels for TC genesis Impact of lightning on NOx and O3 (air quality) - Use ABI and GLM for high-impact weather and AQ Coupled models (atmosphere-chemistry-land-ocean) Further development of hybrid variational-ensemble systems - hybrid forecast error covariance with optimal Hessian preconditioning 27

Thank you! References: Apodaca, K., M. Zupanski, M. DeMaria, J. Knaff, and L. Grasso, 2013: Evaluating the potential impact of assimilating satellite lightning data utilizing hybrid (variational-ensemble) methods. Submitted to Tellus. Zhang, M., M. Zupanski, M.-J. Kim, and J. Knaff, 2013: Assimilating AMSU-A Radiances in TC Core Area with NOAA Operational HWRF (2011) and a Hybrid Data Assimilation System: Danielle (2010). Mon. Wea. Rev., in print. Zupanski M., 2013: All-sky satellite radiance data assimilation: Methodology and Challenges. Data Assimilation for Atmospheric, Oceanic, and Hydrologic Applications (Vol. II), S.-K. Park and L. Xu, Eds, Springer-Verlag Berlin Heidelberg, 730 pp. 28