Tue 1/19/2016 • Discuss SCM assignment

• Some HPC cluster information

• Discuss journal articles

• WRF model overview, software

• Collect survey, project assignment (email okay)

Reminders/announcements: - WRF SCM assignment due Th 1/21 (easy – if system

cooperates)

WRF SCM • Sample sounding: I used Greensboro, NC from

yesterday just to test

• Several steps to set location/date if wish to change

• You are encouraged to try changing the input sounding, time of year, location, etc.

• Very convenient for simple physics experiments

WRF SCM Assignment • Visualize output with ncview, IDV, or NCL:

• ncview: Must (for now) run on login node from old Centos 5 • ssh –Y login51.hpc.ncsu.edu • source c5_aliases.csh (located in /gpfs_share/mea716/class ) • cd to directory where wrfout file resides • ncview wrfoutfilename

Perturbation temperature

WRF SCM Assignment • Visualize output with ncview, IDV, or NCL:

• IDV: Not installed on HPC system • sftp the wrfout file, naming differently if to a PC • Open the IDV [free from http://www.unidata.ucar.edu/software/idv/ ] • Use the “data Choosers” tab, point to wrfout.nc file

Perturbation temperature

WRF SCM Assignment • Visualize output with ncview, IDV, or NCL:

• NCL: (NCAR Command Language) • To set NCL environment,

source /gpfs_share/mea716/class/software/ncl.csh

• Create symbolic link from original wrfout to wrfout_SCM.nc • Use pre-packed ncl routine already in WRFV3/test/em_scm_xy • Just type ncl wrf_scm_1.ncl

Pre-packaged NCL routine generates 5-panel plot, feel free to modify which parameters are shown

Model Output Visualization Many options…

Vapor IDV

Vis-it

Matlab

GEMPAK

+ GrADS, RIP, Python, etc.

NCL

ncview

ParaView

NCSU HPC configuration

Helpful to have some level of familiarity with the HPC computing system

https://www.ncsu.edu/hpc/Documents/BladeCenter/hpc_cluster_config.pdf

NCSU HPC configuration See http://ncsu.edu/hpc/Hardware/henry2_architecture.php

GPFS: General Parallel File System

Shared and Distributed Memory (sm+dm)

• Single computer (node) can have more than one processor or core (the processors share memory - sm)

• Can also have several nodes, distributed memory - dm

• Most clusters have both (several cores per node and many nodes); to maximize efficiency, utilize both

• Need OpenMP for sm, MPI (MPICH2, openmpi) for dm

RAM

Proc1 Proc2

“Node 1” – shared memory Cluster of nodes - distributed memory

HPC System See: http://ncsu.edu/hpc/Documents/BladeCenter/GettingStartedbc.php Avoid running CPU or memory-intensive jobs on login nodes - Instead, use “VCL” capability, or move data to a local system for analysis

Avoid filling disks! - Use “df” and “du –sh” to monitor usage Avoid monopolizing job queues - Check using qstat716 and other variations of qstat command

Aristotle – Meteorologica, 350 BC

Newton’s Principia (1687) – calculus needed to apply math to atmosphere; Euler extended progress by 1755

Thermodynamics advanced rapidly in 1800s (see Kutzbach 1979 text – The Thermal Theory of Cyclones)

For more history: Frisinger (1977), Friedman (1989), Harper (2010), others K. Harper, Weather By the Numbers G. Kutzbach, The Thermal Theory of Cyclones R. M. Friedman, Appropriating the Weather P. Lynch, The Emergence of NWP: Richardson’s Dream

Longer-Term Historical Perspective on the Atmospheric Sciences

“meteorological journal”

Bjerknes (1904)

What was the overall science question, hypothesis, or finding presented in this paper? Is it important?

What was the basis for the authors’ claims?

What is the main result of this paper?

Did the author make a compelling case for their results?

What is the broader significance of this work? Has this paper stood the test of time?

Bjerknes (1904)

Page 1: What was status of observational network at the time of this writing?

Bjerknes, in his earlier studies of electric oscillations, actually contributed significantly to the development of “wireless telegraphy” mentioned here

Are the equations he used the same as what we use in, say WRF? Almost… but he used the 2nd law of thermo rather than a moisture continuity equation

Page 2: “But it must be admitted that we could have overlooked… “…cosmic

effects of an unknown kind may be imagined.” “… those of an electrical or optical nature…” Rainbows, electrical influence on condensation. Have these concerns been realized?

Section 3- suggests averaging of output to bring out important detail (omit noise)

Bjerknes states “The exact analytical solution, even if we could write it down,

would not give the result that we need.” –What did he mean?

Bjerknes (1904)

Page 3: end of section 4: Says problem depends on ability to divide it up. What division does he propose? (Hydro)dynamics and physics (thermodynamics)

What criticism does he direct at “theoretical hydrodynamicists”?

This relates to his discovery of the circulation theorem; separation of dynamics and thermodynamics led them to “avoid every serious contact with meteorology”

Section 5 – Here, seems to be aware of PBL complexities (“frictional stress”); what strategy does he suggest to circumvent this challenge? Bypass laboratory experiments and use empirical results on larger spatial and temporal scales

Section 6- 2nd to last paragraph: Ideas used in modern data assimilation; utilizing physical consistency between variables to derive information from observations

Section 6: Have any aspects of this discussion proven to be incorrect? “The number of individual operations does not have to be excessively large” “Even if striving for high accuracy, a one-hour interval should suffice” “…be able to work easily with much larger time intervals, such as six hours”

Bjerknes (1904)

This article is >100 years old… written long before electronic computers were invented. What aspects remain relevant?

Foresighted? Prophetic? How did it influence subsequent developments? In what ways was Bjerknes correct about challenges facing NWP?

In what ways was Bjerknes overly optimistic? Was this good or bad? “It is certain that we will not encounter insurmountable mathematical difficulties…” “… the individual operations will probably be easy to execute.” “The number of individual operations does not have to be excessively large.”

Of what relation was he not yet aware?

No knowledge of CFL condition (limiting time step), need to filter IC “Even if striving for high accuracy, a 1-hour interval should suffice.”

Bjerknes (1904)

As an aside, by 1914 Bjerknes was much more cautious about NWP 1910 went to Leipzig, began analyses for Europe for NWP initial conditions Petzold worked with Bjerknes… “confluence lines” (fronts) discovered WWI – Petzold killed (5 of 10 Ph.D. candidates from group killed) Son Jack Bjerknes took over work in Norway, frontal theory born

V. Bjerknes interacted with politicians, public… in office every day even in his 80s

The “Father” of NWP (& arguably atmospheric science):

Norwegian Vilhelm Bjerknes “If it is true, as every scientist believes, that subsequent atmospheric states develop from the preceding ones according to physical law, then it is apparent that the necessary and sufficient conditions for the rational solution of forecasting problems are the following:

1. A sufficiently accurate knowledge of the state of the atmosphere at the initial time, and

2. A sufficiently accurate knowledge of the laws according to which one state of the atmosphere develops from another.”

First Numerical Weather Forecast (without the benefit of electronic computers!) [Felix Exner also made a numerical forecast of sorts even before Richardson, but less complete] Performed this work during 1916-1919, while serving as an ambulance driver in World War I in France

Lewis Fry Richardson (1881-1953)

Lynch (1999) Richardson’s Marvelous Forecast

What was the overall science question, hypothesis, or finding presented in this paper? Is it important?

What was the basis for the authors’ claims?

What data and methods were used?

What is the main result of this paper?

Did the author make a compelling case for their results?

What is the broader significance of this work?

Almost seems to downplay V. Bjerknes’ role at first, but credits more later What was the outcome of Richardson’s first forecast?

- LFR’s forecast a “calamity” – 145 mb/ 6 h pressure tendency! Forecast initial time: 7 UTC 20 May 1910, using V. Bjerknes analysis data Why was Richardson’s initial forecast so bad?

- Biggest issue, needed to smooth initial data, particularly wind field - Also, issue with sea-level pressure correction at Strasborg - CFL condition had not yet been discovered, but here, only tendency

computation so it wasn’t cause of this bad forecast Did Richardson understand the cause of his erroneous results?

- Richardson knew there was a problem, attributed poor forecast to bad wind observations

What were the consequences of the inaccurate forecast?

Lynch (1999)

LFR (1922) published book (available in NRL) that included many advanced ideas that are still in use in modern models: • Equations in flux form • Staggered grid, grid boxes, upper BC issues • Leapfrog scheme for time (called “step-over” by Richardson)

Detailed sequence of operations outlined; Richardson even printed

“computing forms” so others could get involved Lynch shows that a reasonable tendency can be obtained from Richardson’s

initial conditions by smoothing initial data to remove high-frequency noise Lynch also ran model for 24-h forecast, but then has to discuss CFL issues; a

shorter time step was needed Lynch has made both model and initial data available on web page

Lynch (1999)

07 UTC 20 May 1910: Richardson’s Forecast

ESRL 20th Century re-analysis, Ensemble Kalman Filter data assimilation using only surface pressure observations

H

L

06 UTC 20 May 1910 (20th Century RA)

SLP, mean and spread 500 Z, mean and spread

Lynch (1994)

Richardson estimated that 64,000 human “computers” needed to keep pace with weather in “forecast factory”

Is this an accurate estimate? If not, why not? - A huge underestimate, due to lack of knowledge of CFL

http://mathsci.ucd.ie/~plynch/Dream/Dream.html

Richardson’s “night signs” are analogous to MPI software Richardson’s “carbon-based computing units” (humans) replaced by silicon chips in HPC cluster

Lynch (1999)

From Lynch, 2006: Richardson’s Dream

Also interested in turbulence and performed many terrestrial experiments. The Richardson number, a dimensionless parameter in the theory of turbulence is named after him. From Weather Prediction by Numerical Process (p 66): “Big whirls have little whirls that feed on their velocity, and little whirls have lesser whirls and so on to viscosity– in the molecular sense” [A play on Jonathan Swift's "Great fleas have little fleas upon their backs to bite 'em, and little fleas have lesser fleas, and so ad infinitum." (1733)].

Lewis Fry Richardson (1881-1953) From Wikipedia:

Pacifist Conscientious objector, drove ambulance-

WWI

After war, rejoined British Met Office

Resigned on grounds of conscience when it was merged with Air Ministry in 1920

Lewis Fry Richardson (1881-1953) From Wikipedia:

The discovery that his meteorological work was of value to chemical weapons designers led him to abandon all his efforts in this field, and destroy findings that he had yet to publish (Korner 1996).

History of NWP: Time Line • V. Bjerknes (1904): forecasting as initial-value problem; basic

system of equations already known

• L. F. Richardson (1916-19): first attempt at practical NWP

• 1928: Courant, Fredrichs, Lewy (CFL) condition for computational stability (see Charney, 1950)

• 1930’s: Advent of radiosonde, upper-air data more widely available

• 1940s: First successful numerical simulation made by Charney, Fjortoft, and von Neumann on electronic computer (filtered equations)

• First Operational NWP forecast, Sweden 1954 (Rossby’s group)