Numerical Weather Prediction Modeling for MURI Atmospheric Parameter Retrievals

Numerical Weather Prediction Modeling for MURI/Atmospheric Parameter Retrievals John R. Mecikalski, Derek J. Posselt CIMSS Co-Investigators 1. 2. 3. 4. OUTLINE Overview of NWP support GIFTS Simulated Data for Algorithm & Product Development Computational Requirements Atmospheric Parameter Retrievals Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

NWP Infrastructure at CIMSS • PSU/NCAR MM 5 • UW-Nonhydrostatic Modeling System (UW-NMS) • Weather Research & Forecasting (WRF) • Rapid Update Cycle-2 (RUC 2) The capabilities of numerically simulating the atmosphere over a wide range of meteorological scales, over large geographical domains, and for realtime numerical weather prediction (NWP) are rapidly increasing at CIMSS. Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

NWP support for Ongoing Projects • Cloud-Radiative modeling for instrument validation, radiance and retrieval algorithm development (MURI PI & Co-I’s) • Generate “truth” atmosphere for satellitebased estimates of PBL stability, and convective initiation studies (MURI Co-I’s & MURI PM) • Assessment of turbulence with NAST-I data (MURI PM) Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

The UW-NMS and PSU/NCAR MM 5 • A robust NWP system: – – scaleable explicit physics at 4 km-resolution explicit microphysics for accurate clouds variably-stepped topography • Excellent model for “Cloud-Radiative” experiments: Independent of other NWP systems • Developed for multi-processor, distributed memory computational environments (Fortran-90 with MPI). Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

NWP for Retrieval & Simulation • NWP can provide atmospheric retrieval algorithms critical “first guess” information. • NWP for direct GIFTS data and instrument simulation: – A numerically simulated atmosphere is considered “nature” and is assumed to very accurately represent the true state. – Requires a sophisticated numerical model and is therefore computationally very expensive (1 Gflops per data “cube”) Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Improving Retrieval First Guess AERI (red) versus Radiosonde (black) Temperature Dew Point Temperature First Guess Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

NWP in Atmospheric Retrievals: AERI+Model/Radiosonde Temperature Comparison • April 1998 -October 1999, 463 AERI+Model/ radiosonde profiles Altitude (km) • Differences are less than 1 deg K GIFTS ETA MODEL • AERI+Model every 10 minutes, sondes every 3 -12 hours AERI/MODEL - RADIOSONDE (K) Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin NWP MODEL

Simulating GIFTS Data High-resolution numerical simulations are used to provide the following atmospheric parameters to the GIFTS radiative-transfer model: – Temperature – Water vapor mixing ratio – Mixing ratios and mean particle diameters of cloud and ice liquid water – Liquid and ice water path – Cloud-top height with respect to both liquid and ice cloud Goal: To provide investigators with simulated (interferogram) data that accurately represents what will eventually come from the GIFTS instrument. Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Numerical Model Output to Simulated Radiances NWP Data Cube Simulated GIFTS Data Cube Forward Model: Dave Tobin Clouds from UW-NMS Simulated GIFTS Brightness Temperatures Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Derived GIFTS Data GIFTS SIMULATED VERTICAL TEMPERATURE PROFILE Pressure (mb) • Vertical Temperature Profiles • 1 sounding per 1 scene pixel • 128 x 128 = 16384 scene pixels • 4 km pixel spatial resolution (nadir) GIFTS SIMULATED TEMPERATURE DATA CUBE 950 MB LAYER TEMPERATURE e (d eg) itud Lat “Regional Sounding” Product: • 100 vertical layers • Retrieved values/cube = 128 x 100 • 1. 6 million retrieved values/cube • 10 second dwell time Temperature (K) Air Temperature (K) Longitu de (deg ) Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Modeling Infrastructure for Large Scale GIFTS-IOMI data processing • Need to simulate GIFTS data at 4 x 4 km resolution over large domain: – Distributed memory, massively parallel computer code and computer system – Must be accomplished in a timely manner [O(few days)] – Demands a sophisticated atmospheric model (UWNMS) • Combination of UW-NMS and MM 5 allows us to simulate “large” regional domains. Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Necessary Computational Infrastructure for MURI: SMP & Cluster Systems • Large-Scale Symmetric Multi-Processor (SMP) – – Integrated unit with high-bandwidth backplane Shared RAM, multiple CPU, single OS kernel Communication: shared memory & semaphores Examples: SGI Origin, IBM RS/6000 • Linux Cluster – – Network of inexpensive COTS computers Multiple RAM, multiple CPU, multiple OS kernel Communication: TCP/IP, sockets & datagrams Examples: Sandia ‘CPlant’, Forecast Systems Laboratory ‘Jet’ • 64 -bit Linux Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Hardware Diagram: Linux Cluster Network Switch UPS Rack and CPUs 30% of the initial cost Disk Fileserver Array Computer Tape Archive 70% of the initial cost Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Current Computational Limits 3 x 3 Cube Domain: • 4 km resolution • Number of grid points: 400 x 40 = 6400000 • Approximate memory use: 15 -20 Gb of RAM • Total Cluster memory: 24 Gb of RAM Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Near-Term Limits 5 x 5 Cube Domain: • 4 km resolution • Number of grid points: 640 x 40 = 16384000 • Anticipated memory use: 45 -50 Gb of RAM • 16 more processors (8 more nodes) • 32 -bit limit for domain set-up Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Future Needs 25 x 25 Cube Domain: • 4 km resolution • Number of grid points: 3200 x 40 = 4. 1 x 108 • Approximate memory use: 1170 Gb of RAM • Doable? Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Minimum Calculation Times for 3 by 3 Simulation (3 time steps) * Not parallelized yet Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Atmospheric Parameter Retrievals Progress to Date: June 2002–May 2003 • Temperature and Water Vapor – Jun Li • First Winds – Chris Velden, Gail Dengel • Stability Indices – Wayne Feltz, John Mecikalski • Atmospheric (PBL) Turbulence – John Mecikalski, Ryan Torn, Wayne Feltz • Visibility: “GVision” – Derek Posselt, Wayne Feltz, John Mecikalski, Tom Rink Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Atmospheric Temperature, Moisture, Ozone Clear Sky retrieval of T 1. The Physical retrieval model has been developed 2. Two fast ways for Jacobian calculation are developed 3. Contrast between surface skin temperature and surface air temperature on boundary layer moisture 4. Simulation studies using cube data from MM 5 Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

First Winds Issues for Winds … Critical need for LARGE “cube” simulation data sets; without such large domains, useful wind sets from simulated GIFTS data are not possible – First “GIFTS” winds have been produced. – Issues of cube numbers when retrieving winds (2 x 2 or larger) – Issues of concatenating cubes Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Stability Indices “Truth” LCL Retrieved LCL Lower Stability Higher Stability First Retrieval of Stability Example: Lifted Condensation Level (LCL) LCL may be used as a measure of boundary layer depth, and/or the depth of the inversion atop the boundary layer (e. g. , marine inversion). Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

PBL Turbulence Evaluating Turbulence from Hyperspectral Measurements Example: Shear-Driven Instabilities: AERI-derived Boundary Layer Depths CBL Waves & Rolls Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

PBL Turbulence August 8, 2001 qe autocorrelation starting 300 minutes after sunrise Boundary Layer Cumulus Correlation June 23, 2001 q autocorrelation starting 320 minutes after sunrise Clear Day (no clouds) Time Difference (min) Preliminary Findings: We appear to have identified boundary layer “roll” turbulent features that produce qe variations at this AERI site at periodic intervals. Ongoing work will evaluate the horizontal scales of these roll structures Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond based on bulk stability parameters (e. g. , Ri). 14– 15 May 2002, Madison, Wisconsin

“GVision” – A Tool for Coalescing GIFTS-IOMI Data “Vis. AD” Capabilities: Java-based application, developed at UW, that is designed for the optimal manipulation and display of large meteorological data sets. Purpose: – Draw together disparate data for viewing atmospheric parameters – Capitalize on in-house expertise for visualizing GIFTS-IOMI data – Test and validate all models using within UW-MURI (NWP, RTE, etc. ) – Develop 3 rd-order fields (i. e. slantwise visibility) Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

Goal for Indian Ocean METOC Imager IOMI-GIFTS 4 km “Cube” Point Weather Data • Marine Inversion • Aerosol & dust detection • Flight-level & directional visibility • Flight-level turbulence • SST for engine efficiency • Surface characterization Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin

In Conclusion. . . • We need to be thinking “Big” – NWP to support IOMI-GIFTS for large domains – IOMI-GIFTS data flow within an Modeling system • NWP for next generation IOMI – IOMI-GIFTS validation experiments: NWP support – NWP to develop “turn-key” fleet-ready data system • First Progress of Atmospheric Parameters – T, q, winds, stability, turbulence & – Continue from here. . . Workshop on Hyperspectal Meteorological Science of UW-MURI and Beyond 14– 15 May 2002, Madison, Wisconsin