From ViRBO

Jump to: navigation, search

GEM 2007 Agenda

GEM2007mini FG9 at 2007 GEM Mini-Workshop at Fall AGU

GEM 2007 Space Radiation Climatology (FG9) Combined Notes by Bob Weigel, Paul O'Brien, Tim Guild


  1. Space Radiation Climatology (FG9)
    1. Breakout 1: Thursday, 10:30-12:15
      1. Intro: O'Brien Intro to FG, discussion of focus/scope
      2. Introduction: Geoff Reeves
      3. Invited: Aaron Ridley: Space Weather Reanalysis - AMIE
      4. Invited: Bob Schunk: The Global Assimilative Ionosphere Model
      5. Paul O'Brien: Summary of Next Generation Radiation Specifications Consortium (NGRSC)
      6. Reiner Friedel: Discussion of Scope
      7. Reeves: Why Climatology (and what is it)
    2. Breakout 2: Thursday, 1:30-3:15
      1. Radiation Belt Data and Simulations
      2. Weigel: Coordination with Virtual Observatories (ViRBO) ppt
      3. Reeves
      4. Shprits
      5. Vassiliadis
      6. Reiner slides
      7. Geoff Reeves slides: Why climatology?
      8. ????????
    3. Breakout 3, Thursday, 3:45-5:30
      1. James Weygand: VMO
      2. Sasha Uhkorskiy (for Pontius Brandt) Status of IMAGE inversions.
      3. Liz Macdonald
      4. Reiner
    4. GEM Tutorial, Friday, 9:00-10:00
    5. Breakout 4, Friday, 10:30-12:15
    6. Out-Brief to GEM Plenary by O'Brien, Friday, 2:00 pm

1. Space Radiation Climatology (FG9)

1.1. Breakout 1: Thursday, 10:30-12:15

1.1.1. Intro: O'Brien Intro to FG, discussion of focus/scope

  • We want a time-dependent, “global” estimate of the state of the inner magnetosphere for a full solar cycle (reanalysis)
  • Primary targets: Particles
    • Radiation Belts (inner and outer)
    • Ring Current
    • Plasmasphere
  • Ancillary targets: Fields
    • Large-scale E&M fields
    • Wave fields (especially ULF, VLF, EMIC)
  • We will get there through data-assimilation
    • Consolidate and clean up our data
    • Augment our simulations with data-assimilation capabilities
    • Fall back on statistical models as needed

1.1.2. Introduction: Geoff Reeves

  • Noted dramatic change in GEM focus group meaning and approach.
  • Newer style is to let FG develop and evolve the issues.
  • Noted that FG will develop many of the questions to be asked and approaches to be taken.
  • They are open to "how can we develop climatology" with a focus on modeling as appropriate for GEM.
  • How will they interface with other focus groups?

1.1.3. Invited: Aaron Ridley: Space Weather Reanalysis - AMIE

  • Slide 1: Ridley has done most of the work related to AMIE.
  • Slide 1: Driven by Eric Kihn's desire to develop a sandbox for having AMIE model runs to mine.
  • Slide 1: Idea was to run AMIE for whole solar cycle and then use output to drive models that are highly dependent on it. For example, GITM.
  • Slide 1: Spent a year on automation and testing. Data quality is major factor. One year of runs is manageable without scripts. Noted difficulty with managing scripts given OS updates. Development of scripts is major time commitments.
  • Slide 1: Jay Albert noted that another difficulty is dealing with newer versions of models. Ridley noted that as a result he gutted AMIE i/o formats so that it was easier to get up and running. Albert noted another factor is change or improvement in physics. Ridley noted that physics does not really change significantly, but there is still a versioning issue (AMIE is an inversion technique, so core statistical approach does not change. Basic equations are Maxwell's equation + Kalman Filter).
  • Slide 2: Blow-ups happen. Covered methods he develped for automating re-do's of these intervals. "Gutted" AMIE to automatically deal with blow-ups.
  • Slide 2: Discussion of why blow-ups happen. Friedel noted that one could reduce resolution to deal with isolated spikes in data (real or due to bad data). Ridley noted that resolution was hard coded.
  • Slide 2: Albert asked if everything was re-run after a fix was made. Ridley noted that in principle this is done, but storage time and to a lesser extent CPU is an issue. Also, transfer time is a bottleneck and other problems were created by NOAA data transfer policies that placed a burden on making simple transfers from one floor to another.
  • Slide 3: Data processing and format issues.
  • Slide 4: Examples of problems.
  • Slide 5: Goldstein: why to these weird things happen? Ridley noted that sometimes people don't look at data.
  • Slide 6: Reeves: these problems are just part of data simulation. Sometimes you will get data that is daily-averaged and these obvious errors [e.g. sawteeth] are averaged over and the scientist does not even know it happened.
  • Slide 7: Albert: Are consistency issues flagged as part of the data output? Ridley noted IMF input is given so one can tell when problems exist.
  • Slide 7: Albert: If CPU is not issue (but disk space is), when will it make more sense to just run the model on demand? Ridley: This is true, but ...?
  • Slide 8: ?: Would be useful to have data base of information about what went into the model.
  • Slide 8: O'Brien noted that he appreciated the last slide which showed the number of publications. This is the main motivation of the FG. It is not the engineering aspect, but the fact that the work results in good science.

[More notes on Aaron Ridley’s talk]

Reasonable to run a year of a model, not really for a solar-cycle.

Observations might be the limiting factor.

Jay Albert: Do you worry about which version of the model generates the solar-cycle? The answer (I thought) was to freeze the model and clean the data.

You need to automate this somehow. Especially when the model blows up. Some logic in the script to query the average observations. If model more than 2/3 sigma away from the observations, throw away certain observations. If no better, include more observations.

DATA STORAGE IS BIGGEST ISSUE. Especially on supers.

Data processing is a big issue, too. Require standardized format for the duration of the run. Data cleaning: Spikes are easy to eliminate, but baseline shifts are very hard.

We need intelligent data cleaning algorithms, which is what the SC people called science middleware. This is the kind of thing that Sebastien did, labeling each data point with quality flags. Talk to him about automated quality controls…

Jay Albert: Did the AMIE reanalysis store the “bad” flags? When there’s not IMF, it’s easy to mask it out. Also, if storage is the issue, just distribute the code and inputs, and people could run it themselves.

Other comment: Make “stripping” scripts, so people can strip out just the parameters they’re interested in.

Aaron : we use “summary” plots for that, but (I think) not very useful.

Oh. When they tried to *USE* it, to drive other models, sometimes get unphysical results.

Everything available to the community on the website. 11 publications! This is the “voting with your feet” that Jimmy Raeder talked about.

1.1.4. Invited: Bob Schunk: The Global Assimilative Ionosphere Model

  • Slide 1: Noted that they are developing physics-based electrodynamics properties and that it will be interesting to compare results with that of AMIE.
  • Slide 2: Noted issue of bug-fixing. Most bugs are fixed within a few days, but there is a tendency for people to want to publish based on an already-fixed issue.
  • Slide 3: Data assimilated exactly as they are measured.
  • Slide 4: Gauss-Markov Kalman Filter Model
  • Slide 5: Data on web needs to have error estimates added and lots of quality control.
  • Slide 6: Ridley: Could you have used IRI? Shunk: it is too far away. Kalman filter breaks down when the data are too far away.
  • Slide 7: ?: How are errors computed? Propagating many models and ?
  • Slide 8: Data issues: Noted that in 70's data were provided with error estimates. Now everything is automated and this is not done. Noted that as a result of this work the funding agencies recognized the importance of quality control and things have been improving.
  • Slide 9: Correlation lengths - you must know in space and time. If measurements are available on 1-hour time scale, are the values valid for only 15 minutes. They did an individual study of this. Results are published in Shim et al. Found Gauss-Markov model can be improved by modifying correlation lengths that were previously estimates based on physical intuition.
  • Slide 10: Uncertain parameters in physics-based models.
  • Slide 15: Albert: In radiation belt world, not all physics is known. Some people hope to do trial-and-error tests on missing physics and see which gives a better result. Schunk: this is another approach to that they take.
  • Hudson: How sparse can your data set be? Radiation belt is quite different in that respect. Schunk: It is still worth doing. 15 years ago people did not believe it could be done. Even if data are sparse, the results may eventually drive the need for more data. Also, he found their progress resulted in funding agencies to provide more work on error estimates.
  • Shripts: Thinks it is a misconception that there are not enough data. He will show that L-shell cuts is sufficient to get results. Hudson: Very little equatorial data, however. Reeves: Kalman filtering gives you specification of where accuracy is low. Also, still need to know limitations of data. "You can't let perfection get in the way of progress". However, Kalman Filter is not a magic wand.
  • Vassiliadis: In radiation belt, correlation lengths allow one to extend to higher and lower local L-shells.
  • O'Brien: Being able to show where modeler's need missions to improve data coverage when writing proposal's is very useful.
  • Shunk: They had to do all the error estimation from web-based data sources themselves. Nice part is that even if you have 100% error, you can still assimilate the data.
  • Aaron Ridley: could you use IRI instead? I think Bob said that if your model is too far from the data, assimilation won’t work. Trevor’s comment for SIMM.
  • Question: How do you specify the error of the model? Loads of different simulations, for different conditions. That’s how you get the error covariance matrix. Otherwise, when running forward, do 30 runs for each time step, then find the covariance matrix from them.
  • Mary Hudson: Radiation belts are much data-sparser, and which lacks physics. Schunk: Worth starting, at least. Can probably get somewhere.

1.1.5. Paul O'Brien: Summary of Next Generation Radiation Specifications Consortium (NGRSC)

  • Will be moved to different section

1.1.6. Reiner Friedel: Discussion of Scope

  • Will cover a few topics related to sparseness.
  • Slide 2: Discussion of DREAM model block diagram
  • Ridley: At UMICH they have been doing something similar. Problem is with divergence of B. They have used divergence B measurements to ingest them into model. At each grid point it should be zero, but Kalman Filter must be "told" this by using it as an input.

1.1.7. Reeves: Why Climatology (and what is it)

  • Slide 1: Why climatology? This image addresses question of if we have enough data.
  • Slide 2-5: Examples of climatology. Coronal holes and strong relativitic electron appearance correlation plot of Paulikas and Blake, 1976. Plots were 6-month averages. Get a different picture if you look at 27-day averages.
  • Slide 6: How correlation studies can evolve toward climatology. Climatology can show that slope relation can change as a function of solar cycle.
  • Slide 7: How event studies can evolve toward climatology. Storms can produce enhancements, depletion, or no change.
  • Slide 8: Goldstein: Had GRL that showed correlation between plasmapause on 3-day time scale, much shorter than the Li et al. plot of 30-day averaged SAMPEX data.
  • Slide 9: Friedel: One has to be careful when using climatology data for model inputs. Often variation about climatology is quite different on short time scales.

1.2. Breakout 2: Thursday, 1:30-3:15

1.2.1. Radiation Belt Data and Simulations

  • O'Brien: Overview of data sets ppt

Terry Onsager – how does N-POESS or its absence affect our NGRSC and FG9 efforts? Reeves – if POES measurements were just a little better, they’d be really useful (for electrons). POES is high value for Protons due to long history.

1.2.2. Weigel: Coordination with Virtual Observatories (ViRBO) ppt

  • Giles: In effect, because you are storing so much data locally, you are building a resident archive + virtual observatory. You can have different funding sources for both of these. Weigel: Yes, we have that luxury, because our community is smaller and the data volume is smaller. Also, we have a need for a resident archive because so much important data is not in usable or online form.
  • Shprits: I have noticed that some of the readers linked to on the wiki don’t seem to work. Will there be a software testing phase? Weigel: Our approach has been to post things as we get them and to note if we have tested them. The idea is that someone may find use of it before we have had a chance to test it. In the case of the readers, we will be looking at them in the next six months. Shprits: One of the things that we would like is to have pre-computed PSD. Weigel: I had a conversation with Friedel last night and was surprised to learn how many people do this and how long it takes. This has been added to our “wanted” list.
  • ?: ViRBO-AGU connections for electronic supplements would be useful (of course, can’t get around subscription requirements).

1.2.3. Reeves

  • Vassiliadis- interested in observed correlation lengths as opposed to model error correlations used in enKF.
  • Hudson – How is L* of the magnetopause calculated? Trace from Shue magnetopause
  • Loto’aniu – B model is a big deal. Reeves – measure magnitude of B model effect by converting from flux to PSD in one model and back in another model.

1.2.4. Shprits

  • Hudson – where do you put plasmapause? Depends on simulation, but usually use Carpenter-Anderson (Kp based)
  • Loto’aniu – more pronounced peaks at smaller pitch angles? Shprits -- Yes. Not sure why, but pitch angle scattering might make the difference. Loto’aniu – PSD peaks can arise from non-L-diffusive ULF interactions.
  • Hudson: How are losses parameterized with respect to plasmapause location? Shprits: Yes, we use Carpenter and Anderson.
  • Loto’aniu : Why are peaks more pronounced in 3D UCLA Radiation Belt code results?
  • Extended discussion of acceleration by diffusion process that result from ULF waves

1.2.5. Vassiliadis

  • Shprits: The peaks shown are in fluxes. Vassiliadis: Tomorrow these will be discussed.
  • Konodov .. At UCLA we use Kalman filter to estimate lifetime parameters in 1-D diffusion model.
  • O’Brien: if expertise lies in analysis, climatology and reanalysis sets will help for comparison with analysis limited by available data sets. Hope is to get GEM-sponsored proposals for focus group
  • Reeves: how would we like to share data sets with each other?
  • O’Brien noted that PC index + his reanalysis would be enough to do a significant study where these data provided boundary conditions.
  • Reeves: is there a list of high priority list of runs that you think you could provide for an extended period of time or run output you want for a long period of time? Vassiliadis will provide.
  • Reeves: is there a need for a standard data product for an output? A way of comparing a data driven to data assimilation model outputs, even if the output is not something that was optimized for? O’Brien: Output must not be inarbitrary coordinate system.
  • Yuri: Misconception that we don’t have much data. Satellites actually measure over the oceans.
  • Mary: We really do lack data at equatorial plane at low L-shells. Why RBSP exists.
  • Geoff: Assimilation is a nice way to quantify where we are data-rich and data-poor.
  • Vassiliadis: We do have sparse data relative to ionosphere. But, CORRELATION lengths are important to quantify how data-poor we really are. Extend GEO observations to much lower Lshells based on those correlation lengths.
  • Paul: Quantifying that data-sparcity is good to motivate new hardware.

1.2.6. Reiner slides

  • You can actually use derived quantities for your assimilation (PSD), provided that you can provide something like data to compare it to. Maybe derived from observations, maybe just a reasonable guess.
  • Can also play with the inputs, and include derived parameters in the inputs based on a priori experience from looking at data.

1.2.7. Geoff Reeves slides: Why climatology?

  • Paulikas and Blake correlation is done for 6 month averages. If you change the averaging window, you get very different correlations ( not correlated very much) More like a lower limit. Correlation studies evolve toward climatology studies.
  • Event studies can also evolve toward climatology studies.
  • Xinlin’s SAMPEX solar-cycle plot is 30-day averages. Plasmasphere boundary nicely matches outer belt inner boundary in that plot. If you average over shorter timescales, not a very good correlation. But there’s a physical reason why you need to average longer, which is the hiss timescales of 10 days (I think?)
  • Jerry Goldstein: I did that with 3-day averages.
  • Geoff: Interesting to see the behaviour of the answer with different timescales. What happens when you average over 10 days, 27 days, 6 months, solar-cycle.

1.2.8.  ????????

  • Yuri: Are you looking at hosting fluxes AND PSD? Answer, Yes. Make sure it works.
  • Vassiliadis: Can use the correlation lengths derived from observations to spread the innovation across multiple Ls.

1.3. Breakout 3, Thursday, 3:45-5:30

Ring Current/Plasmasphere Data and Simulations

  • Identification of simulations ready for data assimilation or in progress
  • Coordination with Virtual Observatories (Including talk by James Weygand: VMOs)
  • Discussion of data sets
  • LANL GEO update (MacDonald)
  • Towards global ring current pressure - HENA inverstion status (Ukhroskiy)

Forgotten data from O’Brien’s list: GPS TEC, POES, DMSP, FAST, ground mags, IMAG & Polar waves, ground-based waves. Magnetometers from Polar, CRRES, GEO, IMAGE(?)

1.3.1. James Weygand: VMO

  • Bunch of tables, presumably at housing sim data is a big, open issue. Maybe leverage CCMC.
  • Geoff: Since some of the data are hosted at ViRBO, VMO will redirect you to ViRBO. But why even bother with ViRBO then?
  • Bob: more of a community portal, for specific groups
  • Paul: Who can host modeling run data? CCMC?
  • Bob: CCMC can’t do it. Not enough space/resources.
  • Paul: Decimate the simulation results somehow. Need to make it really easy to get, and running a simulation is prohibitive.

1.3.2. Sasha Uhkorskiy (for Pontius Brandt) Status of IMAGE inversions.

IMAGE HENA inversions good to factor of 2 but not fully automated.

  • Geoff: Is this inversion an automated procedure. At all. 50 CPU-years to invert 6 years of IMAGE data.
  • Sasha: really cool end result, since it will give you the plasma in the inner msphere, as well as the field? Is that what Sorin suggested?

1.3.3. Liz Macdonald

  • Thomsen et al, Statistics of … SW 2007. Compare with Geoplasma reanalysis.
  • Compare Thomsen SW2007 paper with Geoplasma reanalysis from O’Brien/Lemon.
  • Mark Moldwin: Factor of 2-3 variation of low energy ions throughout solar cycle. Statistical studies of O+ and He+ using GPS tomography.
  • (aside for Tim: Liz / Jim Roeder would really like O+ included in the RCM climatology).

1.3.4. Reiner

  • Goldstein: almost all RC models include a p’sphere (often Ober)
  • Run the plasmasphere simulations for a solar cycle. Then do data-assimilation. Then ask science questions about why it changes with season, solar-cycle, etc. (some pushing by Paul and Geoff on Jerry and Mike Leimohn to run their plasmasphere models on long timescales.)
  • What are the long-term effects of the solar cycle on the plasmasphere?
  • Lower pspheric density in solar max? or min?
  • Mike Liemohn: Has already ran RAM for a solar-cycle. Dipole, volland-stern fields, LANL boundary conditions. End of summer ETA.
  • Paul: We did the Geoplasma reanalysis. Use ‘em for ring current simulations.
  • Moldwin: ULF resonance from magnetometers to provide ppause location is pretty bulletproof.
  • Goldstein: funded to host plasmapause locations.
  • Michelle Thomsen: AFRL plasmapause location on nightside.
  • Reiner: Are the plasma sphere models validated? Good enough?
  • Mike Liemohn: No density along a field line measurements? Assumptions in models. No composition. Etc.
  • Richard Denton: use empirical models for a source term, then evolve according to dynamics model.
  • Goldstein: MSM is stripped down model. Could do RC, RB, maybe even plasmasphere. Has refilling algorithm.
  • Thomsen: Borovsky & Denton: plumes at GEO.
  • Geoff wants to come up with “climate” science questions that we can work toward answering, to say we did somehitng at the end of the FG. Paul thinks we should adopt the “Costner” strategy and build something, then they will come.
  • Why does the solar wind speed control relativistic electron enhancements?
  • Adding more physics to models is good, but hard. Data-assimilation could be easier.
  • More along the lines of what Paul was saying. Models might be improved a lot by improved boundary conditions, rather than inputting new physics.
  • Some models just try to predict: How many hurricanes will happen next year? Not what the wind speeds are, etc. Maybe we could fill that niche?
  • Do basic model/data comparison.

1.4. GEM Tutorial, Friday, 9:00-10:00

  • O'Brien's GEM Tutorial ppt

1.5. Breakout 4, Friday, 10:30-12:15

Strategy and planning session

  • Wanted Reanalysis product list compiled by Vassiliadis ppt
  • What can we do quickly to develop momentum?
  • What are our greatest challenges?

1.6. Out-Brief to GEM Plenary by O'Brien, Friday, 2:00 pm

  • Heard from GAIM, Space Weather Reanalysis Project, and the VxOs
  • Discussed scope – want to include ring current and plasmsphere but concerned about having enough breakouts to do this effectively
  • Already have first attempt at reanalysis of inner/out belts
  • Will likely have solar-cycle length simulation of ring current by end of summer
  • Trying to decide whether to identify specific questions we want to answer with reanalysis climatology or whether to focus just on doing the reanalysis climatology. Will create a list of specific questions to guide us and to help communicate with sponsors.
  • We ran head-on into the pattern of using GEM breakouts as an opportunity to discuss interesting recent science results, but not necessarily related to focus group. For example, everyone who wanted to present results on the radiation belts wanted to speak in our breakout. We tried to accommodate this, but not all talks were directly related to our climatology focus. We are encouraging researchers to rely more heavily on the poster sessions to present results that are not as closely tied to the focus.
Retrieved from ""
Personal tools