application of a global mesoscale model for predicting the...

1

Application of a Global Mesoscale Model for Predicting the Formation of Twin Tropical Cyclones

associated with a Madden-Julian Oscillation

by

Bo-Wen Shen1,2, Wei-Kuo Tao2, Robert Atlas3

Yuh-Lang Lin4, Chris D. Peters-Lidard2, Jiundar Chern2,5, Kuo-Sen Kuo2,5

1UMCP/ESSIC; 2NASA/GSFC; 3NOAA/AOML, 4NC A&T Univ; 5UMBC/GEST;

5th COAA Taipei, Taiwan, June 28-30, 2010

2

Acknowledgements

NASA/GSFC: Oreste Reale,, Jenny Wu, Phil Webster

NASA/HQ: Tsengdar Lee

NASA/JPL: Peggy Li

Navy Research Lab: Jin Yi

NOAA/GFDL: Shian-Jiann Lin

UMCP: Antonio Busalacchi

NASA/ARC: Chris Henze, Piyush Mehrotra, Samson Cheung, Henry Jin, Johnny Chang, . C. Schulbach, R. Ciotti, C. Niggley, S. Chang, W. Thigpen, B. Hood A. Lazanoff, K. Freeman, J. Taft, control-room

Funding sources: NASA ESTO (Earth Science Technology Office) AIST (Advanced Information System Technology) Program; NASA Modeling, Analysis, and Prediction (MAP) Program; NASA Energy and Water Cycle Study (NEWS) Program; National Science Foundation (NSF) Science and Technology Center (STC)

Global Mesoscale Modeling B.-W. Shen 2009, Taiwan3

High-resolution Global Modeling at AGU 2010 WPGM

http://www.agu.org/meetings/wp10

• Meeting place: Taipei, Taiwan• Meeting time: 22-25 June 2010• Deadline: 25 Feb 2010 for abstract submissions• Session title: “High-resolution Global and Regional Modeling and Simulations of

High-impact Weather and Climate”

• Conveners: Bo-Wen Shen of UMCP/ESSIC and NASA/GSFC, Shian-Jiann Lin of NOAA/GFDL, Jin-Luen Lee of NOAA/ESRL, Masaki Satoh of CCSR at U. of Tokyo

4

Outline

Introduction

Global Mesoscale Modeling with NASA Supercomputing Technology

Numerical Results• 5-day track/intensity forecasts of Katrina (2005) and Ivan (2004) • 10-day genesis forecasts of Twin TCs (2002) • Multiscale interactions during the formation of Nargis (2008) • 15-day simulations of Madden-Julian Oscillations in 2002

Summary and Conclusions

5

Scenarios in Decadal Survey

• Extreme Event Warnings (near-term goal): Discovering predictive relationships between meteorological and climatologically events and less obvious precursor conditions from massive data set multiscale interactions; modulations and feedbacks between large/long-term scale and small/short-term scale flows

• Climate Prediction (long-term goal): Robust estimates of primary climate forcings for improved climate forecasts, including local predictions of the effects of climate change. Data fusion will enhance exploitation of the complementary Earth Science data products to improve climate model predictions.

Courtesy of the Advanced Data Processing Group, ESTO PI WorkShop, Cocoa Beach, FL

6

High-impact Tropical Weather: Hurricanes

Hurricane Katrina (2005)•Cat 5, 902 hPa, with two stages of rapid intensification

•The sixth-strongest Atlantic hurricane ever recorded.

•The third-strongest landfalling U.S. hurricane ever recorded.

•The costliest Atlantic hurricane in history! ($75 billion)

• http://en.wikipedia.org/wiki/Hurricane_Katrina

Severe Tropical Storm Nargis (2008)•Deadliest named cyclone in the North Indian Ocean Basin

•Short lifecycle: 04/27-05/03, 2008; identified as TC01B at

04/27/12Z by the Joint Typhoon Warning Center (JTWC).

•Very intense, with a Minimum Sea Level Pressure of 962 hPa

and peak winds of 135 mph (~Category 4)

•High Impact: damage ~ $10 billion; fatalities ~ 134,000

•Affected areas: Myanmar (Burma), Bangladesh, India, Srilanka

Each year tropical cyclones (TCs) cause tremendous economic losses and many fatalities throughout the world. Examples include Hurricane Katrina (2005), which is the costliest Atlantic hurricane in history, and TC Nargis, which is one of the 10 deadliest TCs of all time.

7

Progress of Hurricane Forecasts (by National Hurricane Center)

Figure: The progress of hurricane forecasts by National Hurricane Center. Horizontal axis indicates year, and vertical axis shows forecast errors. Lines with different color show different forecast intervals. During the past twenty years, track forecasts have been steadily improving (left panel), but Intensity forecasts have lagged behind (right panel).

Track Errors Intensity Errors

better

8

Multiscale Interactions of TC Formation

Global(GCMs)

dx~O(100km)

mesoscal/regional

dx~O(10km)

Clouddx~O(1km)

(convectionprocess)

Global mesoscale

Super-parameterization(multi-scale modeling framework, MMF)

MJO Equatorial Rossby Wave

vortex merger/axisymmetrization

modulation vortex dynamics

CISK/WISHE

(initial conditions, initialization) (cps, surface/boundary layer)

feedback

physicalprocesses

scaleinteraction

Model scale

CISK: conditional instability of second kind; CPs: cumulus parameterizations; MMF: multiscale modeling framework; MJO: Madden-Julian Oscillation; TC: Tropical Cyclone; WISHE: Wind induced surface heat exchange;

9

Challenges

• Satellite Data Challenges: o) Massive data storageo) Display/visualizations (~TB)

• Modeling Challenges:o) Explicit representation of (the effects of) convective-scale motionso) Verification of model simulations at high spatial and temporal resolutions o) Understanding and representation of multiscale interactions

• Computational Challenges:o) Real-time requirements with supercomputing o) Efficient data I/O (at runtime) and data access (via massive storage systems)o) Parallel computing and parallel I/O with new processors (e.g., multi-cores)o) Processing and visualizations of massive data volumes with large-scale

multiple-panel display system

Satellite, Numerical Models, and Supercomputing Technology

10

NASA Major Supercomputers

Columbia Supercomputer (ranked 2nd in late 2004)

• Based on SGI® NUMAflex™ architecture 20 SGI® Altix™ 3700 superclusters, each with 512 processors Global shared memory across 512 processors

• 10,240 Intel Itanium® 2 CPUs; Current processor speed: 1.5 gigahertz; Current cache: 6 megabytes

• 20 terabytes total memory; 1 terabyte of memory per 512 processors

Biswas, R., M.J. Aftosmis, C. Kiris, and B.-W. Shen, 2007: Petascale Computing: Impact on Future NASA Missions. Petascale Computing: Architectures and Algorithms, 29-46 (D. Bader, ed.), Chapman and Hall / CRC Press, Boca Raton, FL.

Pleiades Supercomputer (ranked 3rd in late 2008; 6th

in 2010)

• quad-core Xeon 5472 (Harpertown) CPUs, speed - 3GHz; Cache - 12MB per CPU

• 81,920 cores in total (512 cores per cabinet) • 120+ TB memory in total, 1 (8) GB memory per

core (node)• 3.1 PB disk spaces• InfiniBand, 9,216 compute nodes• 973.4 Tflops/s peak; 722.7 Tflops/s Linpack

11

Concurrent Visualization System V2.0

Figure : The newly developed Concurrent Visualization (CV) System (Version 2, top panel). Rounded rectangles indicate systems, and rectangles indicate processes. The whole system (from left to right panels) consists of a computing node ("Columbia Node"), a middle-layer system ("coalescer"), and the hyperwall-2 128-node 8-core/node rendering cluster. These systems are used for data extraction, handling, and visualization; and for MPEG image production and visualization display. The first generation CV system (Version 1) is shown in the right panel for comparison.

Shen, B.-W., W.-K. Tao, G. Bryan, C. Henze, P. Mehrotra, J.-L. F. Li, S. Cheung, 2009: High-impact Tropical Weather Prediction with the NASA CAMVis: Coupled Advanced multi-scale Modeling and concurrent Visualization Systems. Supercomputing conference 2009, Portland, Oregon, November 14-20, 2009.Green, B., C. Henze, B.-W. Shen, 2010: Development of a scalable concurrent visualization approach for high temporal- and spatial-resolution models. AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010.

12

A Science-Driven Approach

Large-scaleNumericalModeling

Supercomputing

ScientificDiscoveries

Goals: •to explore the power of supercomputing technology (e.g., supercomputers and visualization systems) on the advancement of global weather and hurricane modeling; •to discover how hurricanes form, intensify, and move with advanced numerical models; •to understand the underlining mechanisms (how realistic the model depiction of TC dynamics)•to extend the lead-time of hurricane predictions (and high-impact tropical weather predictions): from short-term (~5days) to extended-range (15~30 days) forecasts

In short, to advance numerical models and thus to extend the lead time of hurricane predictions with modern supercomputing technology

13

The Global Mesoscale Model

1. Model Dynamics and Physics:

• The finite-volume dynamical core (Lin 2004); • The NCAR physical parameterizations, and NCEP SAS as

an alternative cumulus parameterization scheme• The NCAR land surface model (CLM2, Dai et al. 2003)

2. Computational design, scalability and performance (suitable for running on clusters or multi-core systems)

14

Modeling Approach

• Given an initial mesoscale vortex (TC), we verify model’s performance in predicting its movement and intensification;

• Given a large-scale flow (e.g., MJO and/or AEWs), we investigate if modulations of TC activities (e.g., formation) can be simulated realistically;

• By performing extended-range (15-30day) simulations, we test if our modeling approach could extend the predictions of an MJO or AEWs.

• To understand if and how the lead time of “mesoscale” hurricane prediction can be extended by a global mesoscale model

15

GFS Analysis (~35km) valid at 08/29/12z 96 h Simulations with no CPS

High-resolution runs simulate realistic intensity, radius of max wind (RMW) and warm core.

Landfall errors: e32 (1/4o): 50km, g48(1/8o): 14km, g48ncps (1/8o w/o CPs): 30km

Near-eye Wind Distributions in a 2ox2o box from 96h simulations, validated at 08/29/12z.

0.25o

warm core

AOML (~5km) 0.25o (~25km)

0.125o (~12km) 0.125o (no CPS)

0.125o

Forecasts of Katrina’s Track, Intensity, Structures (Shen et al., 2006b, Geophys. Res. Lett., L13813, doi:10.1029/2006GL026143)

Selected as Journal Highlight by American Geophysical Union Featured in Science magazine (August, 2006)Featured in the 2006 Annual report of SAIC (Science Application International Corp.)

pres

sure

RMW

16

Exploring No-Man’s Land

(Shen et al., 2006c)

Black: best track;Each color line: a 5-day track forcast(13 Forecasts of Ivan, 2004)

NASA high-resolution GCMNHC official forecasts

The official track forecasts relied heavily on the global model forecasts, which prematurely eroded the large and strong subtropical ridge to the north of Ivan that extended well westward across the Bahamas, Florida, and into the Gulf of Mexico. In fact, several of the GFS model forecast cycles consistently eroded the ridge across Bahamas and took Ivan well to the east of Florida, even as the hurricane was approaching Jamaica. (Stewart, 2004, NHC TC report)

17

Intensity Forecasts of Ivan (2004)

Realistic intensification rate, suggesting that given an initial (basic) state, the model is capable of determining the growth rate of the unstable mode (?)

Realistic intensity evolution at landfall, suggesting remarkable model performance in simulating the effects of surface fluxes during the transition of a model storm from Ocean to Land

18

Evaluation for Ivan’s Forecasts

Forecast Evaluation for Hurrican Ivan (2004)

0

20

40

60

80

100

120

140

160

180

Forecast Period (h)

1/8 degree 1/4 degree

1/8 degree 27.77 41.27 63.69 77.01 110.1 134.4 151

1/4 degree 45.16 57.06 81.2 90.16 112 154.2 149.8

12 24 36 48 72 96 120

Preliminary evaluation for thirteen 5-day forecasts of Hurricane Ivan (2004) initialized at 0000 UTC September 3-15, 2004. The corresponding tracks are shown in Figure 4. The average official track errors (with the number of cases in parentheses) by NHC were 24 (63), 47 (61), 79 (59), 108 (56), 161 (52), 222 (48), and 289 (44) n mi for 12, 24, 36, 48, 72, 96, 120 h forecasts. The corresponding track errors with FSU Superensemble model are 21 (53), 38 (51), 58 (51), 81 (51), 126 (47), 171 (43), and 199 (38) nautical mi.

19

MJO and TC Genesis

• The MJO, also referred to as the 30-60 day or 40-50 day oscillation, turns out to be the main intra-annual fluctuation that explains weather variations in the tropics. The MJO affects the entire tropical troposphere but is most evident in the Indian and western Pacific Oceans.

• The modulation of TC activity by the MJO in different regions was documented by Liebmann et al. (1994), Maloney and Hartmann (2000).

• Twin TCs, straddling the equator at low latitudes, occasionally may occur in the Indian Ocean and West Pacific Ocean (e.g., Lander 1990).

20

MJO Dynamics

Current understanding (including theory and hypotheses) indicates that

1. moisture convergence (e.g., Lau and Peng 1987; Wang 1988), 2. surface heat and moisture fluxes (e.g., Emanuel 1987; Neelin et al.

1987), 3. cloud-radiation feedback (e.g., Hu and Randall 1994, 1995),4. convection-water vapor feedback (e.g., Woolnough et al. 2000;

Tompkins 2001), and 5. “discharge-recharge” associated with moist static energy build-up

and release (e.g., Blade and Hartmann 1993)

are important for the MJO’s initiation, intensification, and propagation (see a review by Zhang 2005).

21

Simulation and Visualization of Twin Tropical Cyclones

TC01A TC01A

Kesiny Kesiny

TC02B

Errol

Figure : Predictions regarding the formation of twin tropical cyclones in the Indian Ocean: (a) MJO-organized convection over the Indian Ocean at 0630 UTC 1 May 2002. When the MJO moved eastward, two pairs of twin TCs appeared sequentially on 6 May (b) and 9 May (c), including TC 01A, Kesiny, TC 02B and Errol. Two TCs (01A and 02B) with anti-clockwise circulation appeared in the Northern Hemisphere, while two TCs (Kesiny and Errol) with clockwise circulation in the Southern Hemisphere; (d) Four-day forecasts of total precipitable water, showing realistic simulations of TC’s formation and movement (see Shen et al., 2010b for details).

0630 UTC 1 May 2002 0000 UTC 6 May 2002 0000 UTC 9 May 2002

Shen, B.-W., W.-K. Tao, R. Atlas, Y.-L. Lin, C.-D. Peters-Lidard, J.-D. Chern, K.-S. Kuo, 2010b: Forecasting Tropical Cyclogenesis with a Global Mesoscale Model: Preliminary Results for Twin Tropical Cyclones in May 2002. (submitted)

Previous studies suggest that twin tropical cyclones (TCs), symmetric with respect to the equator, may occur associated with a large-scale Madden-Julian Oscillation (MJO). Here, it is shown that high-resolution simulations of twin TCs associated with the MJO in 2002 are in good agreement with the satellite observations.

22

Formation of 6 TCs in four 10-day Simulations

Init at 05/01/00z

Init at 05/06/00z

Init at 05/11/00z

Init at 05/22/00z

Best tracksindicated byblue lines:


23

Forecasts of Twin TCs:

averaged precipitation over May 8 – 11, 2002 NASA TRMM CNTL (no CPs)

EXP-A EXP-B

All of three runs (CNTL, EXP-A and EXP-B) are initialized at 0000 UTC May 6, 2002 with different moist physical processes.

CNTL: No CPs

EXP-A: with Zhang and McFarlane (1995) and Hack (1994) schemes for deep and shallow- and-midlevelconvection, respectively.

EXP-B: with NCEP SAS (simplified Arakawa and Schubert) scheme (Pan and Wu 1995)

(i)The false-alarm event appears in all of three runs, suggesting missing physical processes in the model or imperfect ICs?(ii)Can column-based physics (CPs) simulate spatial distribution of precipitation better?

TC 01A

TC 02B

KesinyErrol


24

15-day Simulations of an MJO in May 2002

05/02

05/07

05/12

05/17

Accurate prediction of tropical activity at sub-seasonal scales is crucial for extending numerical weather prediction beyond 2 weeks. Among the challenges of this goal is accurate forecasting of a Madden-Julian Oscillation (MJO). This figure is to show that both the fvGCM and fvMMF can realistically simulate the MJO up to 15 days.

Figure. Velocity potential at 200 hPa every 5 days in May 2002 from NCEP analysis (left panels), 15-day fvGCM model predictions at 1/8 degree resolution (middle panels), and 15-day fvMMF model predictions (right panel). The velocity potential plots showed the observed and predicted patterns and propagations of Madden-Julian Oscillation (MJO). The high-resolution fvGCM is able to reproduce realistically the observed patters and intensity as NCEP analysis. The fvMMF even with coarse-resolution (2x 2.5 degree) is also able to predict the large-scale MJO event, except its intensity is somewhat overestimated .

NCEP Reanalysis Global Mesoscale Model Multiscale Modeling Framework

25

Very Severe Tropical Storm Nargis (2008)

00Z Apr 22 12z Apr 27 11z May 2

Durga(22-24 Apr)

Rosie(21-24 Apr) Formation of Nargis Landfall in Myanmar

Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2010a: Predicting Tropical Cyclogenesis with a Global Mesoscale Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008) . J. Geophys. Res., doi:10.1029/2009JD013140.

26

Cited as one of four “Research Breakthroughs” by UMCP

27

Controlling Factors in Nargis Formation

7-day global multiscale simulations suggest the following favorite factors for the formation and initial intensification of tropical cyclone Nargis:

Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2010a: Predicting Tropical Cyclogenesis with a Global Mesoscale Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008) . J. Geophys. Res., doi:10.1029/2009JD013140.

28

3D Visualization of Multiscale Processes associated with the Formation of Nargis (2008)

Figure : Realistic 7-day simulations of the formation and initial intensification of TC Nargis (2008) initialized at 0000 UTC April 22, 2008, showing streamlines at different levels. Low-level winds are in blue and upper-level winds in red: (a) formation of a pair of low-level mesoscale vortices (labeled in ‘V’) at 84h simulation. (b) intensification of the northern vortex (to the left) (c); formation of TC Nargis associated with the enhancement of the northern vortex; (d) intensification of TC Nargis associated with upper-level outflow and moist processes, indicated by the enhanced upper-level outflow circulation. Approaching easterly upper-level winds (labeled in ‘E’) increase the vertical wind shear, suppressing the enhancement of the southern vortex (to the right) in panel (b).

VV

W

E

Nargis

The 3D visualization is to illustrate (1) the formation of a pair of low-level vortices; (2) the transformation of the northern vortex into the TC Nargis; (3) the suppression of the southern vortex.

This visualization is produced by the NASA CAMVis information system with the following configurations: •Model: NASA fvGCM;•Visualization: CVs v2.0•Model spatial resolution: 1/4 degree;•time step: 900 seconds in physics,

45 seconds in dynamics; •# of horizontal grid points: 1000x721;•simulation length: 7 days;•I/O: every physics time step; •file size: 716 GB with 61 vertical levels•wall-time: 2 hours with 240 CPUs

29

Large-scaleNumericalModeling

Supercomputing

ScientificDiscoveries

Summary

“The Cycle”

NASA Global Mesoscale Model: one of the first ultra-high resolution GCMs

NASA Multi-scale Model Framework: consisting of the NASA global model and tens of thousands of copies of NASA cloud resolving model (GCE)

Approaches with explicitly-resolved convection and/or its effects to reduce the uncertainties of cumulus parameterizations

Model Validations with mesoscale weather systems such as the Catalina Eddy, Hawaiian Wake, Mei-Yu front etc

Columbia: SGI Altix, 14,336 cores (Itanium II)Pleiades: SGI Altix ICE, 81,920 cores (Xeon)Hyperwall-2: 128 panels

A unified view on TC formation, including modulation by large-scale flows and interaction between mesoscale vortices, surface fluxes and convection. hierarchical multiscale interactions

Future work: extending the current approach to study hurricane climate and impact of global change on hurricane climate.

Improved forecasts of TC track, intensity and formation

Improved extended-range (15~30 –days) simulations of MJOs and AEWs.

30

Computational performance

• 10 days/70 mins with 240 cpus on columbia • 90 days/75 mins with 720 cpus on pleiades

• 5 days/3 hours with 60 cores, suitable for running on clusters and/or multi-core system

31

Publications (03/15/2009-06/15/2010)

Major publications:• 28. Shen, B.-W. et al., 2009: African Easterly Waves and African Easterly Jet in 30-days High-resolution Global Simulations. A

Case Study during the 2006 NAMMA period in 2006. (Submitted.)• 27. Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2009: Improving Tropical Cyclogenesis Prediction with a Global Mesoscale

Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008). (submitted to JGR- Atmosphere; in press).

• 26. Shen, B.-W., W.-K. Tao, R. Atlas, Y.-L. Lin, C.-D. Peters-Lidard, J.-D. Chern, K.-S. Kuo, 2009: Forecasting Tropical Cyclogenesis with a Global Mesoscale Model: Preliminary Results for Twin Tropical Cyclones in May 2002. (Submitted).

• 25. Shen, B.-W., W.-K. Tao, G. Bryan, C. Henze, P. Mehrotra, J.-L. F. Li, S. Cheung, 2009: High-impact Tropical Weather Prediction with the NASA CAMVis: Coupled Advanced multi-scale Modeling and concurrent Visualization Systems. Supercomputing conference 2009, Portland, Oregon, November 14-20, 2009.

• 24. W.-K. Tao, D. Anderson, J. Chern, J. Entin, A. Hou, P. Houser, R. Kakar, S. Lang, W. Lau, C. Peters-Lidard, X. Li, T. Matsui, M. Rienecker, M. R. Schoeberl, B.-W. Shen, J. J. Shi, and X. Zeng, 2008: A Goddard Multi-Scale Modeling System with Unified Physics. Submitted to Annales Geophysicae Special Issue dedicated to The 1st International Conference on From Desert to Monsoons (in press).

Conference Presentations/Reprint/Seminars:• 23. Shen, B.-W., W.-K. Tao, W. K. Lau, J. Chern, J.-L. F. Li, B. Green, R. Atlas, 2010: Global Multiscale Modeling with NASA

Supercomputing Technology: Mutiscale Simulations of Tropical Cyclogenesis associated with an MJO or AEW. AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (submitted)

• 22. Li, J.-L. F, D. E. Waliser, J. Chern, B.-W. Shen, W.-T. A. Chen, J. Teixeira, W.-K. Tao, T. L. Kubar, 2010: Cloud Vertical Structure in the simulations with Conventional GCMs and a Multi-scale Modeling Framework GCM, Global Analyses and CloudSat observations. AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (submitted)

• 21. Green, B., C. Henze, B.-W. Shen, 2010: Development of a scalable concurrent visualization approach for high temporal- and spatial-resolution models. AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (submitted)

32

Publications (03/15/2009-06/15/2010)

• 20. Chern, J.-D., W.-K. Tao, B.-W. Shen, T. Matsui, J.-L. F Li, D. E Waliser, 2010: Evaluations of the Global and Regional Energy and Water Cycles in a Multiscale Modeling Framework System. AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (submitted)

• 19. Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2009: Predicting Tropical Cyclogenesis with a Global Mesoscale Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008). AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (submitted)

• 18. Shen, B.-W., W.-K. Tao, G. Bryan, C. Henze, P. Mehrotra, J.-L. F. Li, 2009: High-impact Tropical Weather Prediction with the NASA Coupled Advanced multi-scale Modeling and concurrent Visualization Systems (CAMVis). The NASA ESTO AIST 2010 Workshop: Decadal Survey Era Capabilities and Technology Roadmap, Cocoa Beach, FL, February 8-11 2010.

• 17. Shen, B.-W., S.-J. Lin, J.-L. Lee, M. Satoh, and R. Atlas, 2010: "High-resolution Global and Regional Modeling and Simulations of High-impact Weather and Climate", AGU 2010 Western Pacific Geophysics Meeting, Taipei, Taiwan, June 22-25, 2010. (session proposal; accepted)

• 16 Shen, B.-W. and K.-S. Kuo, 2009: Twin Tropical Cyclones associated with Madden-Julian Oscillation in the Indian Ocean, Journal of Earth Science Phenomena, 2009, 13.

• 15. Shen, B.-W., 2009: Global Mesoscale Modeling with NASA Supercomputing Technology: Hierarchical Multi-scale Interactions during the Formation of Tropical Cyclone Nargis (2008). National Central University, Chung-Li, Taiwan, January 5, 2010. (invited talk)

• 14. Shen, B.-W., 2009: Application of a NASA Global Mesoscale Model to Study Tropical Cyclogenesis associated with an MJO or AEW. National Taiwan Normal University, Taipei, Taiwan December 22, 2009. (invited talk)

• 13. Shen, B.-W., W.-K. Tao, W. K. Lau and R. Atlas, 2009: Global Multiscale Modeling on NASA Supercomputers: Extended- Range Simulations of Madden-Julian Oscillations and African Easterly Waves. Fifth International Ocean-Atmosphere Conference (COAA2010), Taipei, Taiwan, June 28-30, 2010. (an abstract submitted)

• 12. Shen, B.-W. and W.-K. Tao, 2009: Application of a Global Mesoscale Model to Study Hierarchical Multiscale Interactions during the Tropical Cyclogenesis associated with African Easterly Waves in 2006. AGU 2009 Fall Meeting, San Francisco, December 14-18, 2009. (poster presentation)

• 11. Shen, B.-W., 2009: Extending the Lead Time of Tropical Cyclogenesis Prediction with a Global Mesoscale Model. Supercomputing Conference 2009, Portland, Oregon, November 14-20, 2009. (selected as one of top NASA project demonstrations at SC09). (booth demonstration; oral presentations)

33

Publications (03/15/2009-06/15/2010)

• 10. Shen, B.-W., 2009: Global Mesoscale Modeling with NASA Supercomputing Technology: Hierarchical Multi-scale Interactions during Tropical Cyclogenesis associated with an MJO or AEW. M Square Office Building of UMCP/ESSIC, November 9, 2009. (an invited talk)

• 9. Shen, B.- W., 2009c: In Support of Hurricane Forecasting Research at CWB in Taiwan. October, 7, 2009 (in Chinese). (an invited news article)

• 8. Shen, B.- W., 2009b: Supercomputer and Typhoon Prediction. Want daily forum. September 8, 2009. (in Chinese). (an invited news article)

• 7. Shen, B.-W., 2009a: Current Challenges of Hurricane Intensity Predictions in Taiwan. Want daily forum and China Times. August 14, 2009. (in Chinese).

• 6. Shen, B.-W., 2009: Global Mesoscale Modeling with NASA Supercomputing Technology: Hierarchical Multi-scale Interactions during Tropical Cyclogenesis associated with an MJO or AEW. The Atlantic Oceanographic and Meteorological Laboratory (AOML) / Hurricane Research Division (HRD) , Miami, FL, August 17-21, 2009. (an invited talk).

• 5. Shen, B.-W. and W.-K. Tao, 2009: Global Multiscale Modeling on NASA Supercomputers: Extended-Range Simulations of MJOs and AEWs. 7th CMMAP Bi-annual team meeting, Fort Collins, CO, July 27-30, 2009. (poster presentation)

• 4. Shen, B.-W. and W.-K. Tao, W. K. Lau, R. Atlas, and J. Chern, 2009: Hierarchical Multi-scale Interactions during Tropical Cyclone Formation associated with an MJO or AEW. 7th CMMAP Bi-annual team meeting, Fort Collins, CO, July 27-30, 2009. (oral presentation)

• 3. Chern, J., W.-K. Tao, T. Matsui, and B.-W. Shen, 2009: Evaluating Goddard Multi-scale MOdeling Framework Through the TRMM Triple-Sensor Three-Step Evaluation Framework (T3EF). 7th CMMAP Bi-annual team meeting, Fort Collins, CO, July 27- 30, 2009. (poster presentation)

• 2. Shen, B.-W. and W.-K. Tao, 2009: Global Multiscale Modeling on NASA Supercomputers: Extended-Range Simulations of MJOs and AEWs. The EarthCARE Workshop 2009, Kyoto, Japan, June 10-12, 2009. (session chair; poster presentation)

• 1. Shen, B.-W., 2009: Global Multi-scale Modeling on NASA Supercomputers: Application to Hurricane Forecasts. The first project meeting of AIST Project CAMVis. NASA/ARC/NAS, Moffett Field, CA, May 11, 2009. (oral presentation)

34

Acronym

• 3D-Winds 3D Tropospheric Winds from Space-based Lidar (DS mission)• ACE Aerosol-Cloud-Ecosystem (DS mission)• CAMVis Coupled NASA Advanced Multiscale Model and Concurrent Visualization Systems• CISK: Conditional Instability of second kind• Columbia: supercomputer at NASA/ARC, operating in late 2004• CPs: Cumulus (or Cloud) parameterizations• CVs: concurrent visualization (CV) system• fvGCM: finite-volume General Circulation Model (global model)• GCE: Goddard Ensemble Model (cloud model)• GCM: General Circulation Model • HW-2: hyperwall-2• Ifort: Intel Fortran Compiler (version 10, 11)• IPoIB: Internet Protocol over InifiBand• MJO Madden-Julian Oscillation (large-scale tropical weahter)• MMF (fvMMF) fvGCM + tens of thousands of GCEs• MPI: message passing interface, widely used in distributed-memory system• NHC: National Hurricane Center• OBS: observation• OMP: OpenMP, multi-threaded programming paradigm,

suitable for shared-memory machines• PATH Precipitation and All-weather Temperature and Humidity (DS mission)• Pleiades: supercomputer at NASA/ARC, operating in late 2008• Procs: processes• RDMA: remote direct memory access• SMAP Soil Moisture Active-Passive (DS mission)• TC: tropical cyclone, a generic term for hurricanes, typhoons, cyclones• vis: visualization• WCRP: World Climate Research Programme • WISHE: wind induced surface heat exchange • WWRP/THORPEX: World Weather Research Programme/ \The Observing System Research and Predictability Experiment • XOVWM Extended Ocean Vector Winds Mission (DS mission)• YOTC: Year of Tropical Convection

35

Backup Slides

36

Visualizations of Scale Interactions

Blue: low-level windsRed: upper-level windsA 5-days Forecast of Hurricane Ivan (2004), starting at 1200 UTC September 11

12 h 48 h

76 h 120 h

LL

L

LEnhancement of upper-level winds, as shown by pink color lines

IVAN

Before Ivan made landfall, scale interaction between its outflow and an upper-level trough (labeled as “L”) might have been contributed to Ivan’s intensification, as indicated by purple lines in panel (d).

(a) (b)

(c) (d)

37

• Project demonstration on magic planet at Maryland day, April 25, 2009.

• Bo-Wen Shen was interviewed by Earth Magazine of American Geological Institute to provide comments on the recent article entitled “Maximum hurricane intensity preceded by increase in lightning frequency” by Price et al. (2009), which was published by Nature Geoscience. As the current progress of hurricane intensity prediction is slow, and accurate intensity prediction with a numerical model is still very challenging, Shen agrees that correlation between hurricane intensity and lightning frequency could be a useful indicator to predict hurricane intensity.

• Bo-Wen Shen (UMCP/ESSIC, 613.1) published three news articles (in Chinese) to discuss the challenges of intensity and rainfall predictions for landfalling Typhoons in Taiwan, including deadly Typhoon Morakot, and the recent advance in high-resolution global models and supercomputer technology at NASA that show a potential for improving intensity predictions. Morakot (2009) devastated Taiwan and Eastern China, and caused tremendous damage by dumping a total of about 2.5 meters (100 inches) of rain in Taiwan in early August, 2009.

Education and Public Outreach

38

Visualizations on the Magic Planet at Maryland Day (April 25, 2009)

39

WCRP (World Climate Research Progreamme)

• The World Climate Research Programme, sponsored by the International Council for Science (ICSU), the World Meteorological Organization (WMO) and the Intergovernmental Oceanographic Commission (IOC) of UNESCO, is uniquely positioned to draw on the totality of climate-related systems, facilities and intellectual capabilities of more than 185 countries. Integrating new observations, research facilities and scientific breakthroughs is essential to progress in the inherently global task of advancing understanding of the processes that determine our climate.

• The two overarching objectives of the WCRP are: to determine the predictability of climate; and to determine the effect of human activities on climate

• To achieve its objectives, the WCRP adopts a multi-disciplinary approach, organizes large-scale observational and modelling projects and facilitates focus on aspects of climate too large and complex to be addressed by any one nation or single scientific discipline.

• Scientific guidance for the WCRP is provided by the Joint Scientific Committee (JSC), consisting of 18 scientists selected by mutual agreement between the three sponsoring organizations and representing climate-related disciplines in atmospheric, oceanic, hydrological and cryospheric sciences. JSC for WCRP (as from 1 January 2010): Chair Professor A. Busalacchi ESSIC, University of Maryland, US

40

WWRP/THORPEX

• WWRP: World Weather Research Programme • THORPEX: The Observing System Research and Predictability Experiment • THORPEX is a 10-year international research and development programme to

accelerate improvements in the accuracy of one-day to two-week high impact weather forecasts for the benefit of society, the economy and the environment.

• THORPEX establishes an organizational framework that addresses weather research and forecast problems whose solutions will be accelerated through international collaboration among academic institutions, operational forecast centres and users of forecast products

• WMO AREPWWRP• The Atmospheric Research and Environment Programme (AREP) co-ordinates and

stimulates research on the composition of the atmosphere and weather forecasting, focusing on extreme weather events and socio-economic impacts. AREP supports global research initiatives under programmatic guidance from the Commission for Atmospheric Sciences (CAS). It does this through two programmes: Global Atmosphere Watch (GAW) and the World Weather Research Programme (WWRP) including THORPEX.

41

Year of Tropical Convection

• The Year of Tropical Convection (YOTC), which commenced on 1 August 2008, is a joint World Climate Research Programme (WCRP) – World Weather Research Programme (WWRP)/THORPEX initiative. YOTC is a year of coordinated observing, modelling, and forecasting with a focus on organized tropical convection, its prediction, and predictability. The realistic representation of tropical convection in global models is a long-standing, grand challenge for both numerical weather prediction and climate prediction. Incomplete knowledge and practical issues in this area disadvantage the modelling and prediction of prominent phenomena of the tropical atmosphere across a wide range of scales. Scientific challenges encompass the prediction of El Niño-Southern Oscillation (ENSO), monsoons’ active/break periods, tropical cyclones and many other tropical (and extratropical) features. During YOTC the intent is to exploit the vast amounts of existing and emerging observations, the expanding computational resources and the development of new, high-resolution modelling frameworks. This focussed activity and its ultimate success, will benefit from the coordination of a wide range of ongoing and planned international programmatic activities and collaboration among the operational prediction, research laboratory and academic communities.

42

Tropical Cyclones and Climate Change Knutson et al., 2010, NGEO779

• For future projections of tropical cyclone activity, the challenge is to develop both a reliable projection of changes in the various factors influencing TCs, both local and remote, and a means of simulating the effect of these climate changes, such as storm frequency, intensity and track distribution. environmental factors and their impact on TC activities,

• Uncertainties in model projections of future TC activitiy arise owing to both uncertainties in how the large-scale tropical climate will change and uncertainties in the implications of these changes for TC activity.

• Although each of global climate models project a substantial increase in tropical SSTs during the twenty-first century, important differences exist in the regional-scale details of their projects, which lead to marked differences in the downscaled regional projections of TC activity.

• We cannot at this time conclusively identify anthropogenic signals in past TC area.

• - obervational, theoretical and modeling studies, ….. but still have many limitations.

43

• For all cyclone parameters, projected changes for individual basins show large variations between different modeling studies. - “predictability” at regional scales is not very good,

• “we have very low confidence in projected changes in individual basins” physical parameterizations are not universal????

• “we have low confidence in projected changes on TC genesis-location, tracks, duration and areas of impacts. Existing model projection do no show dramatic large-scale changes in these features. large-scale flows are ok, but their modulations on TC activities are different???

• High-resolution dynamical models consistently indicate that greehouse warming will cause the globally averaged intensity of TCs to shift towards stronger storms (with intensity increase of 2-11% by 2100), … decreases in the globally averaged frequency of TCs, by 6-34%.

• Rainfall rates are likely to increase

44

The need of 3D Visualizations for the representation of multiscale interactions

Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2009: Improving Tropical Cyclogenesis Prediction with a Global Mesoscale Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008). (submitted to JGR, in press).

High-resolution simulations with this model show that the formation of TC Nargis can be realistically predicted up to 5 days in advance. It is suggested the improved representation of the environmental conditions (listed in the figure) and their hierarchical multiscale interactions were the key to achieving this lead time.

Though these multiple processes and their scale interactions can be realistically simulated with the advanced global mesoscale model, presentation of these processes in a simple form for education and outreach purposes becomes very challenging. To achieve this goal, 3D visualizations will be illustrated in the next slide.

45

Coupling NASA Advanced Multi-Scale Modeling and Concurrent Visualization Systems for Improving Predictions of Tropical High-Impact Weather (CAMVis)

Key Milestones

TRLin = 3, TRLcurrent

= 4

• Implement (update) model components and CV on the Pleiades (Columbia) supercomputer;

• Conduct initial benchmarks 09/2009• Improve parallel scalability of model components;• Integrate the NASA fvGCM and CV; Develop the super-

component mgGCE 03/2010

• Couple the NASA mgGCE and CV;• Implement and test an I/O module

09/2010• Integrate the fvMMF (fvGCM+mgGCE) and CV 03/2011• Streamline data flow for production runs 09/2011• Test the CAMVis system; Produce results 03/2012

Approach1.

Improve parallel scalability of the multi-scale modeling system to take full advantage of the next-generation peta-

scale supercomputers (e.g., NASA Pleiades)

2.

Integrate NASA multi-

scale model system, including Goddard Cloud Ensemble model (GCE) and the finite-volume General Circulation Model (fvGCM), and the concurrent visualization (CV) system

3.

Significantly streamline data flow for fast processing and visualizations ; Conduct high-resolution numerical runs

4.

Test coupled systems

•

Co-I’s: Wei-Kuo Tao (GSFC, Co-PI), Bryan Green (Co-PI), Chris Henze, Piyush Mehrotra, (ARC), Jui-Lin Li (JPL)

•

Partners: Antonio Busalacchi (UMD), Peggy Li (JPL)

PI: Bo-Wen Shen (UMD/ESSIC)

Co-I’s/Partners

(a,) fvGCM, showing global simulations of tropical cyclone Nargis (2008) and westerly wind belt in the Indian Ocean, (b) mgGCE with multiple copies of

GCEs, simulating cloud motions, (c) Concurrent visualization (CV) system on the 128-panel , and (d) Pleiades supercomputer with 51,200 cores

ObjectiveCAMVis weather prediction tool will be developed to achieve the following goals by seamlessly integrating NASA technologies (including advanced multiscale modeling visualizations and supercomputing):• to inter-compare satellite “observations”

and model simulations at fine resolution, aimed at improving understanding of consistency of satellite-derived fields;•to improve the insightful understanding of the roles of atmospheric moist thermodynamic processes and cloud-

radiation-aerosol interactions with 3D visualizations;•to improve real-time prediction of high-impact tropical weather at different scales.Project CAMVis has the potential for supporting the following NRC Decadal Survey Earth Science missions, including CLARREO, ACE, PATH, 3D-Winds.

(a) (b)

(c) (d)

46

Hierarchical Multiscale Interactions

Meaning: • multiscale but “asymmetric” , e.g., large scale vs. small scale (e.g., organization vs. individuals)• dependence of predictability at different scales (“P” may not be totally independent)• (potential) existence of control-feedback-response relationship (at different stages) that can be

simulated with numerical models (e.g., Q-E for CPs?)• a different range of “involved” scales at a different stage of a (TC) lifecycleImplications: • From a modeling perspective, clarifications of the following terms might be helpful to understand the

hierarchical (two-way) multiscale interactions. 1. lifetime of the large-scale flow (forcing duration): the period that the large-scale flow could act as a

forcing, (e.g., barotropic or baroclinic instatilbity); 2. lifetime of the small scale flow;3. response time: the time for small scale flow to respond and adjust (e.g., time for the initiation of

convection); 4. feedback duration: the period that small scale flow can provide feedbacks to large-scale flows (which

should be less than its lifetime)

• For two-way scale interactions, accurate representation of “forcing” and its “control”, and simulations of “response” and its “feedback” are equally important!

how quickly a small-scale flow can respond in reality and in “models”.

how long it will take for small-scale flows to have significant impact on the large-scale flow

47

Scale Interactions

1. Is the initiation of small-scale weather system an IVP?2. Is the initiation of any consecutive small-scale system an IVP? Does the initiation of the new

system depend on the previous small-scale system or the large-scale (environmental) system? 3. If depending on the large-scale system, to what extent the previous small-scale system could

impact/change the large-scale system? Phases (timing) and magnitudes (intensity). what’s impact scale in timing and spacing (on 2nd, 3rd, 4th small-scale system?)

4. From a modeling perspective, spin-run time and initialization time are equally important.

48

Scale Interactions

1. Lifetime of a small-scale weather system (SWS): initiation, intensification, decaying etc.Q: what’s the feedback over a lifecycle? How about aggregate feedbacks over multiple lifecycles1. If a SWS is stochastic: is initiation or evolution stochastic? Is its feedback stochastic? What

happens if its feedback weaker than other forcing? 2. Timing of initiation depends on (i) physical processes and (ii) model spin-run proceeses3. GCM vs. mg-GCE (3D, IVP?) AGCM vs. OGCM (“strong” 2D surface condition, BVP?)

Deterministic

Stochastic

49

Modeling Approach

• Given an initial mesoscale vortex (TC), we verify model’s performance in predicting its movement and intensification;

• Given a large-scale flow (e.g., MJO and/or AEWs), we investigate if modulations of TC activities (e.g., formation) can be simulated realistically;

• By performing extended-range (15-30day) simulations, we test if our modeling approach could extend the predictions of an MJO or AEWs.

• To understand if and how the lead time of “mesoscale” hurricane prediction can be extended by a global mesoscale model

This work is to illustrate the “potential predictability” of high-impact tropical weather with a global meososcale model, which has shown the potential of simulating more realistic hierarchical multiscale interactions (control/response/feedbacks relationship). To achieve our goals of improving our understanding of and confidence in model’s performance in hurricane short-term and climate simulations, we have done the following:

Model’s predictive ability != Predictability (of a weather system)

50

Physics Parameterizations

• Moist physics:-Deep convections: Zhang and McFarlane (1995);

Pan and Wu (1995, aka NCEP/SAS)-Shallow convection: Hack (1994)-large-scale condensation (Sundqvist 1988)-rain evaporation

• Boundary Layer- first order closure scheme- local and non-local transport (Holtslag and Boville 1992)

• Surface Exchange

- Bryan et al. (1996)

Pan, H.-L., and W.-S. Wu, 1995: Implementing a mass flux convection parameterization package for the NMC medium-range forecast model. NMC office note, No. 409, 40pp. [Available from NCEP].

51

Scientific Visualizations with the NASA CAMVis

• Computational Science (CS) is defined as an inter- disciplinary filed with the goals of understanding and solving complex problems using high-end computing facilities

• The next-generation visualizations should help provide the insights of the complicated physical processes in these complex problems (e.g., hurricane prediction) and thus improve our understanding of these processes at different scales and their interactions.

• A central question to be addressed: “Is new science being produced (with high-resolution modeling) or just really cool pictures?”, which was raised by Mahlman and others who have reservations (Science, p1040, 2006 August).

52

Formation of 6 TCs in four 10-day Simulations

Init at 05/01/00z

Init at 05/06/00z

Init at 05/11/00z

Init at 05/22/00z

Best tracksindicated byblue lines:


53

15-day Simulations of an MJO in 2002

Semidiurnal (?)

Time

54

Westerly Wind Bursts, Low-level Convergence and a pre-TC mesoscale vortex

in 7-day Simulations

04/22 04/29time

5oS

20oN

Two phases of enhanced convection

850-hPa U winds along lon 89oE

850-hPa U winds along lon 88oE

04/22 04/26

Low-level convergence

W

E

Formation/intensification of a pre-TC mesoscale vortex is related to (1) the occurrence of westerly wind burst and (2) peak of low-level convergence.

55

Averaged precip and 850-hPa winds

Averaged preccip and 850-hPa windsfrom 04/27 (day 5 ) to 04/29 (day 7)

Averaged NASA TRMM precip andNCEP Analysis winds

• Suggesting that the aggregate effects of moist processes (latent heating release, surface fluxes and water vapor transport) are simulated reasonably well ?

• Upscaling interactions associated with moist covection?Shen, B.-W., W.-K. Tao, W. K. Lau, R. Atlas, 2010a: Predicting Tropical Cyclogenesis with a Global Mesoscale Model: Hierarchical Multiscale Interactions During the Formation of Tropical Cyclone Nargis (2008) . J. Geophys. Res., doi:10.1029/2009JD013140.

56

Outline

Introduction

Global Mesoscale Modeling with NASA Supercomputing Technology

Simulations of High-impact Tropical Weather• 5-day track/intensity forecasts of Katrina (2005) and Ivan (2004) • 10-day genesis forecasts of Twin TCs (2002) • Multiscal interactions during the formation of Nargis (2008) • 15/30-day simulations of Madden-Julian Oscillations in 2002/2006• Five AEWs and genesis of Hurricane Helene (2006) in 30-day runs

Global Mesoscale Model as a New Research Tool

Summary and Conclusions

57

African Easterly Jet (AEJ)

Thorncroft and Blackburn, 1999

• AEJ is a midtropospheric jet located over tropical north Africa during the northern hemisphere summer

• AEJ is seen as a prominent feature in the zonal wind, with a maximum of around 12.5 m/s at 600-700 hPa and 15oN.

• Its vertical shears are crucial in organizing moist convection and generation of squall lines; • Its horizontal and vertical shears are important for the growth of easterly waves;• Below the AEJ, the easterly wind shear is in thermal wind balance with the surface

temperature gradient (BWS)• One may view the AEJ as resulting from the combination of diabatically forced meridional

circulations which maintain it and easterly waves which weaken it. As the nature of diabatic forcing (e.g., moist or dry convection) differs between models, simulated AEJ is likely to be different (e.g., different heights and/or strengths, which will subsequently affect the easterly waves - model climate

• The rate at which the AEJ is maintained is likely to be particularly sensitive to the boundary layer parameterization

• Other factors should be included: radiation; the establishment of the surface temperature and humidity gradients themselves; etc

58

Map of North Africa

Berry and Thorncroft ( 2005, MWR)

10oN

20oN

0o10oW 10oE 20oE 30oE 40oE

59

30-day Averaged U Winds and Temp

(init at 08/22/00z)NCEP Reanalysis Model Simulations

Shen, B.-W. et al., 2010c: African Easterly Waves and African Easterly Jet in 30-day High- resolution Global Simulations. A Case Study during the 2006 NAMMA period. Submitted.

60

30-day Averaged Precip and 850 hPa Winds (init at 08/22/00z)

TRMM Precip and NCEP Aeanalysis Winds

Model Simulations

Shen, B.-W. et al., 2010c: African Easterly Waves and African Easterly Jet in 30-day High- resolution Global Simulations. A Case Study during the 2006 NAMMA period. Submitted.

61

Sensitivity Experiments

62

Predictability at Different Scales

synoptic scalemesoscale cloud scale

(convectionprocess)

Global mesoscale

Super-parameterization(multi-scale modeling framework, MMF)

• Predictability at different scales may not be totally independent.• With regional models, researchers previously focused on improving convective-scale flows and

their upscaling/feedback (timing and location) processes, and thus mesoscale flows.• Here we try to point out the importance of improving two-way interactions between these synoptic-

and meso-scale using a global mesoscale model, in particular, if the above approach may have a “theoretical limit” on the improvement of the mesoscale predictability.

• In addition, the aforementioned multiscale interaciotns with resolved convection seem to be more realistic, after the uncertainties of CPs are removed.

application of a global mesoscale model for predicting the...

Documents