Jan Mandel/Blog/2011 Aug Sep
August 2: Data assimilation seminar: funding
August 6: Reading some papers
Data assimilation for wildfires
Generally cell-based simulations with a variant of particle filter and importance weights from distance evaluation by symmetric set difference.
- Barcelona group: Abdalhaq-2006-BCI doi ucd, Bianchini-2005-SSS doi ucd, Bianchini-2006-IPM doi ucd Rodriguez-2009-IDR doi ucd Rodriguez-2010-TPD doi ucd Cencerrado-2011-PTA doi ucd.
- Georgia State group: Gu-2008-TAP doi ucd Gu-2009-SEU Gu-2009-SEU Yan-2009-DDA doi ucd Gu-2010-DDD link Hu-2011-DDD link. Distance evaluation also by L1-norm of differences.
- See also evaluation of errors in fire spread: Fujioka-2002-NMA doi ucd by difference of ratio of distances from ignition point to the perimeter (star-shaped burn area assumed).
- Added Dec 5 2012: Symmetric difference of burned area as fitness Rodriguez-2010-TPD doi ucd or equivalently L1 difference of characteristic functions Gu-2010-DDD link.
- Kourtz-1977-FFC report an early cell model, includes wind, with pictures from a line printer! No equations but clear enough. Cited in Sullivan-2009-RWF3 doi ucd
- Alexandridis-2008-CAM doi ucd Alexandridis-2011-WFS doi ucd
- new MesoNH/ForeFire version from the Corsica group: Filippi-2011-SCF link, update on Filippi-2009-CAF doi ucd. Now couples atmosphere spread model from Balbi-2009-PMW doi ucd (which is an update on Balbi-2007-3PR doi ucd). A nice compact formulation of the spread model.
- case studies and fuels: Santoni-2011-WFB link
Convergence of the EnKF
- Thulin et al. Thuli-2011-QMC doi ucd start from our EnKF convergence paper and continue with an empirical study of stochastic convergence using the Kolmogorov–Smirnov test on multiple small ensembles. We had some related thoughts a year ago (see the picture on the right, also copy) but we had not followed through.
- Guo and Yang, Guo-2011-CRP L(p) (p > 1) Convergence results for Particle Filtering. Published by Journal of Computational Analysis and Applications which used to be Springer but now at Euxodus Press, a secretive one-man publisher - no url no doi to this paper. Copy on order from the library.
August 7: Powershow.com
- Frederica Darema DDDAS ICCS 2007
- my 2003 scattering slides
- Challenges of mesoscale data assimilation from http://hirlam.org
- Lagrangian Data Assimilation and Observing System Design from Dynamical Systems Perspective
August 9: Data assimilation seminar: DDDAS
- Craig's site Official NSF page AFOSR see also links
- Photos: long-term criticality minimization general particle filter scheme
August 12: Special Data assimilation seminar at Starbucks
Returning to last year's seminars. How to decrease the number of particles in PF?
- Vania Dukic March 12, 2010 Tracking Flu Epidemics Using Google Flu Trends and Particle Learning Dukic-2011-TFE
- Anne Sabourin August 23 and 30, 2010 elliptic KF abstract slides
- Loren's references on multimodal exponential families and sufficient statistics: Cobb-1981-MEF Cobb-1981-MEF-review, Cobb-1983-EMR Cobb-1983-EMR-updated. These have light tails. Lye-1993-REN also has heavier tails.
- scribbled notes
August 14: Fire models
- Bossert-1998-VCA comparison with experimental data
- getting a point formula from physical simulation: Reisner-1998-CDW Linn-2005-NSG, where to take the wind from? wind shear...
- using WRF with heat sources to study plume Cunningham-2005-CVS
August 21: AbTech AIM
- ITT Federal Services Corporation 1994: "AIM Abductory Inductive Mechanism (Decision-making software developed by the AbTech Corporation of Charlottesville, VA, with funding from the Strategic Defense Initiative [SDI] Small Business Innovative Research [SBIR] program. The software consists of machine learning algorithms and modeling techniques that merge the neural network concept with advanced statistical techniques to produce a tool exponentially more powerful than either precursor. AIM automatically determines the network structure, node types, and coefficients from a user-provided database. It displays a graphical representation of the completed model that illustrates the relationships among variables and gives the user insight into which variables are important. Software developers can embed AIM learning algorithms into their own software to create packages that perform automatic remodeling in situations where relationships are constantly changing. Software developers can use AIM offline to produce models that can be exported as royalty-free ANSI C source code subroutines." local copy
- SBIR grant 1995 Keith C. Drake Other SBIR grants Barron Associates AbTech
- Used in chemical engineering Sundgren-1992-CSA doi ucd, power systems Abdel-Aal-2004-SHL doi ucd,...
- Citation A tutorial on AIM and abductive modeling AI Review archive Volume 4, July/Aug. 1991 p. 7 - the journal exists at Springer but the article is not there
- Reference article Montgomery-1991-ARN doi ucd.
- Polynomial neural network course with overview and algorithms
- In 1998 AIM morphed into ModelQuest Marketminer and vanished round 2000. The company http://www.abtech.com seems to be doing something else. What outperformed this code?
- Wikipedia:Abductive reasoning Wikipedia:Group method of data handling Software KnowledgeMiner thesis about Novak-2006-GNK link reference paper Muller-2003-SOP doi ucd
August 26: Fireline intensity
- Forest Encyclopedia FireWords
- Byram-1959-CFF link pdf Alexander-1982-CIF doi ucd Smith-1978-CPB doi ucd Tangren-1976-TWF doi ucd Keeley-2009-FIF doi ucd
- Need "flame depth" or "flame residence time" per Anderson-1969-HTF link
- Following Andrews-2009-BMS link p. 110, BehavePlus uses Byram-1959-CFF link with adjustments in Albini-1976-EWB link for Rothermel's model.
- In Andrews-1986-BFB link p. 13, fireline intensity is rate of spread times heat per unit area, which is reaction intensity times flame residence time from Anderson-1969-HTF link, the latter is fuel property (Fig. 8 there).
- Again per Andrews-1986-BFB link p. 13 computation of the reaction intensity is from Rothermel-1972-MMP doi ucd (p. 6 eq. (7), p. 11 and p. 26 eq. (27)) with adjustments made in Albini-1976-EWB link (found in Appendix III p. 88; see also review of Rothermel's model from p. 80)
- Reaction intensity is tied to "fuel weight" (the time constant of burning) (see also Rothermel's eq. (7)) but this does not go into Rothermel's eq. (27)??
Also found: Forest Service wildland fire strategic plan 2006
September 1: Data assimilation seminar
- AIM demo
- Epidemic in Google Earth (Brian Smith)
September 3: Google Code suports git
- Google Code Gets Git "... the days of SVN are likely to be numbered" - it's about time
- Choose git when creating project
- Official announcement Git supportImport a Git repository to Google Code as SVN mirror
- Git for OSX
September 4: Reading some papers
September 8: Scott Lett
September 8: Fuel moisture
- Nelson's model: Carlson-2007-ANM link Nelson-2000-PDC doi ucd Nelson Fuel Moisture Model slides pdf pptx
- Comparison: Weise-2005-CTM doi ucd
- Review: Viney-1991-RFF doi ucd Papadopoulos-2011-CRW doi ucd
- Simpler: Matthews-2010-SMP doi ucd Matthews-2007-TPF doi ucd Matthews-2006-PMF doi ucd Sharples-2009-SIA doi ucd
- NFDRS: Burgan-1979-ELF link Fosberg-1971-DHT link
- See also section 4 of what is new in SFIRE paper for a very simple model.
September 12: Fire in Google Maps
- Web-based simulator of the Corsica group
- My experiments with Google API
- Visualization in Google Earth at OpenWFM
- Python scripts to create KMZ are at https://github.com/jbeezley/WRF-GoogleEarth for development use git ssh://repo.openwfm.org/home/git/WRF-GoogleEarth
- Fire code is at https://github.com/jbeezley/wrf-fire for development use git ssh://repo.openwfm.org/home/git/wrf-fire See List of WRF-Fire pages at OpenWFM for info
- Wish list
- Set up simulation with automatic online data retrieval for weather and fire.
- Click on the map and start simulation on a cluster back-end like the Corsica site
- Web portal in CrowdLabs to run the simulation and display results
- A standalone web portal with minimal dependences that anyone can install easily, all files in a public git repository
- Animation while the simulation runs - perhaps use one frame per wrfout and check for new files
- Web portal should support the Google Maps API (better for smartphones) as well as the Google Earth plugin.
- Android and/or iPhone app.