1 / 66

Multi-Scale Probabilistic M o deling in Geospace Science

Multi-Scale Probabilistic M o deling in Geospace Science. Zach Thomas The Ohio State Universi t y Ment o rs: T omo k o Matsuo, Doug Nych k a Ellen Cousins , Mi k e Wilt b erger August 1, 2014. Outline of Summer W o rk.

tirza
Download Presentation

Multi-Scale Probabilistic M o deling in Geospace Science

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Multi-ScaleProbabilisticModelinginGeospaceScience ZachThomas TheOhioStateUniversity Mentors:TomokoMatsuo,DougNychkaEllenCousins,MikeWiltberger August1,2014

  2. OutlineofSummerWork .,. Application:Modelinghigh-latitudeionosphericconvection fromsparseradarobservations

  3. OutlineofSummerWork .,. Application:Modelinghigh-latitudeionosphericconvection fromsparseradarobservations .,. StatisticalMethodology:Techniqueforspatialmodelingonthesphere

  4. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses

  5. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses .., Numericalmodeling

  6. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses .., Numericalmodeling .., Dataanalysis

  7. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses .., Numericalmodeling .., Dataanalysis .,. Practical/Societal:Understandchangesinelectromagneticenergyassociatedwithauroras

  8. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses .., Numericalmodeling .., Dataanalysis .,. Practical/Societal:Understandchangesinelectromagneticenergyassociatedwithauroras .., Disturbancesintelecommunication

  9. ScientificBackground Motivation .,. Scientific:ObtainbetterunderstandingofcomplexinteractionbetweensolarwindandEarth’smagneticfieldbystudyingvariousionosphericprocesses .., Numericalmodeling .., Dataanalysis .,. Practical/Societal:Understandchangesinelectromagneticenergyassociatedwithauroras .., Disturbancesintelecommunication .., Disturbancesinpowergrids

  10. ScientificBackground InteractionbetweenSolarWindandEarth’sMagneticField Figure:OutputfromtheLyon-Fedder-Mobarrymodelcapturingcomplexinteraction betweensolarwindand Earth’smagneticfield.Imagecourtesyofhttps://www.dartmouth.edu/physics/cism/science/lfmmodel.html

  11. ScientificBackground IonosphericElectricPotentialandPlasmaConvectionPatterns Figure:IMFeffectonionosphericconvection.ImagecourtesyofCousins etal.(2010)

  12. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. LatticeKrigisanewR packageforfast/flexiblespatialmodelingbasedonastatisticalmethodologyinNychkaetal. (2014).

  13. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. LatticeKrigisanewR packageforfast/flexiblespatialmodelingbasedonastatisticalmethodologyinNychkaetal. (2014). .,. Keyideaistoexpressthespatialprocessasamulitresolution basisfunctionexpansionwithrandomcoefficients

  14. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. LatticeKrigisanewR packageforfast/flexiblespatialmodelingbasedonastatisticalmethodologyinNychkaetal. (2014). .,. Keyideaistoexpressthespatialprocessasamulitresolution basisfunctionexpansionwithrandomcoefficients .,. RandomcoefficientsmodeledviaacertainMarkovrandom fieldmodelcalledasimultaneousautoregression(SAR)model

  15. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. Observeaprocessatnlocationswithinaspatialdomain D...callthemy(s1),...,y(sn).

  16. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. Observeaprocessatnlocationswithinaspatialdomain D...callthemy(s1),...,y(sn). .,. Foranarbitrarys∈D,wewouldliketomakeinferenceabouttheprocessy(s)fromtheobservations.

  17. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. Observeaprocessatnlocationswithinaspatialdomain D...callthemy(s1),...,y(sn). .,. Foranarbitrarys∈D,wewouldliketomakeinferenceabouttheprocessy(s)fromtheobservations. .,. Commonspatialmodel:Foranys∈D(observedornot): y(s) =m(s)+g(s)+e(s) Process=Mean+SpatialProcess+Ind.Error

  18. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. Observeaprocessatnlocationswithinaspatialdomain D...callthemy(s1),...,y(sn). .,. Foranarbitrarys∈D,wewouldliketomakeinferenceabouttheprocessy(s)fromtheobservations. .,. Commonspatialmodel:Foranys∈D(observedornot): y(s) =m(s)+g(s)+e(s) Process=Mean+SpatialProcess+Ind.Error .,. LatticeKrigdiffersfromothermethodsinitsconstructionof thespatiallydependentprocessg(s).

  19. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. CenterbasisfunctionsatgridlocationsonL gridsovertheobservationregion.

  20. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. CenterbasisfunctionsatgridlocationsonL gridsovertheobservationregion. .,. TheL gridsareobtainedbysequentiallydoublingthe resolutionofthepreviousgrid.

  21. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. CenterbasisfunctionsatgridlocationsonL gridsovertheobservationregion. .,. TheL gridsareobtainedbysequentiallydoublingthe resolutionofthepreviousgrid. .,. Theprocessg(s)isthenexpressedasabasisfunctionexpansion: L nl )) l=1j=1 g(s)= c W[d(s,s)] ∗ l,jl,jl,j

  22. ModificationofLatticeKrigfortheSphere WhatisLatticeKrig? .,. CenterbasisfunctionsatgridlocationsonL gridsovertheobservationregion. .,. TheL gridsareobtainedbysequentiallydoublingthe resolutionofthepreviousgrid. .,. Theprocessg(s)isthenexpressedasabasisfunctionexpansion: L nl )) l=1j=1 g(s)= c W[d(s,s)] ∗ l,jl,jl,j .,. Thecl,j’sarerandomcoefficientsfollowingacertainMarkovrandomfieldmodelcalledaSimultaneousAutoregression(SAR)model.

  23. ModificationofLatticeKrigfortheSphere TheGeodesicGrid .,. ModifyingLatticeKrigforuseonthespherewasdoneintwosteps:

  24. ModificationofLatticeKrigfortheSphere TheGeodesicGrid .,. ModifyingLatticeKrigforuseonthespherewasdoneintwosteps: 1. Insteadoflocatingbasisfunctionsonarectangulargridovera plane,wecenterthemonaspecializedgridonthesphere...calledthegeodesicgrid

  25. ModificationofLatticeKrigfortheSphere TheGeodesicGrid .,. ModifyingLatticeKrigforuseonthespherewasdoneintwosteps: Insteadoflocatingbasisfunctionsonarectangulargridovera plane,wecenterthemonaspecializedgridonthesphere...calledthegeodesicgrid Modifytheconstructionofthe SARmodelfortherandomcoefficients(whichnowliveonthegeodesicgrid)

  26. ModificationofLatticeKrigfortheSphere TheGeodesicGrid .,. ModifyingLatticeKrigforuseonthespherewasdoneintwosteps: Insteadoflocatingbasisfunctionsonarectangulargridovera plane,wecenterthemonaspecializedgridonthesphere...calledthegeodesicgrid Modifytheconstructionofthe SARmodelfortherandomcoefficients(whichnowliveonthegeodesicgrid) .,. Resultisatoolforspatialmodelingovergeneralregionsonthesphere...ortheentiresphere...tobeincludedinfutureversionsofLatticeKrig.

  27. ModificationofLatticeKrigfortheSphere TheGeodesicGrid .,. ModifyingLatticeKrigforuseonthespherewasdoneintwosteps: Insteadoflocatingbasisfunctionsonarectangulargridovera plane,wecenterthemonaspecializedgridonthesphere...calledthegeodesicgrid Modifytheconstructionofthe SARmodelfortherandomcoefficients(whichnowliveonthegeodesicgrid) .,. Resultisatoolforspatialmodelingovergeneralregionsonthesphere...ortheentiresphere...tobeincludedinfutureversionsofLatticeKrig. .,. Thissummer,focusonusingtheprocedureformodelingelectromagneticprocessesintheionosphere.

  28. Figure:LowResolutionBasisFunctions:CapturingLarge-ScaleDependenceFigure:LowResolutionBasisFunctions:CapturingLarge-ScaleDependence

  29. Figure:AddMediumResolutionBasisFunctions:Capturing Medium-ScaleDependence

  30. Figure:AddHighResolutionBasisFunctions:CapturingSmall-ScaleDependenceFigure:AddHighResolutionBasisFunctions:CapturingSmall-ScaleDependence

  31. TrySomeSpatialModeling SpatialInterpolationofElectricPotentialfromLFM-MIXModelOuput .,. Keymodificationforionosphereproblem:varianceofthe electricpotentialisclearlynonstationary.Wecanembedinformationfromnumericalmodeloutputintothestatisticalmodel. Figure:(Left)Regionofhighestvariability;courtesyofMinjieFan,UCDavis(Right)Weightsusedtoinducenonstationaryvarianceinspatialmodel.

  32. TrySomeSpatialModeling SpatialInterpolationofElectricPotentialfromLFM-MIXModelOuput .,. Experiment:UseLFM-MIXmodeloutputofelectricpotentialtostudypredictiveskillofourmodel

  33. TrySomeSpatialModeling SpatialInterpolationofElectricPotentialfromLFM-MIXModelOuput .,. Experiment:UseLFM-MIXmodeloutputofelectricpotentialtostudypredictiveskillofourmodel 1. Randomlyselectsampleof1000points(outof16920)uniformlyoverthepolarregion.

  34. TrySomeSpatialModeling SpatialInterpolationofElectricPotentialfromLFM-MIXModelOuput .,. Experiment:UseLFM-MIXmodeloutputofelectricpotentialtostudypredictiveskillofourmodel Randomlyselectsampleof1000points(outof16920)uniformlyoverthepolarregion. Treatthesepointsas’TheData’...trytogetbackthefull 16920pointsusingourspatialmodelonthesphere

  35. Figure:Fullelectricpotentialprocessfrommodeloutput

  36. Figure:1000randomlysampledlocationsusedtoinformtheinterpolationFigure:1000randomlysampledlocationsusedtoinformtheinterpolation

  37. "True"ElectricPotentialProcess(FromLFM-MIX) 12 40 32 24 16 40 30° 20° 8 > e- 18 0 06 -8 -16 -24 -32 -40 min:-25.37 max:41.50 00

  38. SpatialInterpolationofElectricPotential 12 40 32 24 16 40 30° 20° 8 c 0 18 0 06 B -0 -8 0. -16 -24 -32 -40 min:-25.19 max:41.55 00

  39. Error("True"ProcessMinusPredictedProcess) 12 2.0 1.6 1.2 0.8 40 30° 20° 0.4 10· "C 18 0.0-0 0. 06 -0.4 -0.8 -1.2 -1.6 -2.0 min:-2.19 max:0.76 00

  40. Error("True"Process MinusPredictedProcess) 12 .....::: 2.0 ·..·...:.···· : :.·.. ..· .,..· 1.6 ·.· ···. ..· .. ·. .... 1.2 .·--:. ·.... ::.:.... ·... ...:.·...: .:-··....... 0.8 ·:30°:..:. .,...·..·. .·: ,.'·.. '1:::·:....::.·:"·.,;;. ... ....... 0.4 ··: "' 0.0-0 0. ...·.· .·..·· :-. 18 06 o I,-,-• ...·.··....... .......·:·.· ::. :.·.· ··.:.. .: .: .... -0.4 .,.·... ·.·.. ·::.···.· ,.... .. ..... -0.8 :·. .-. .,. . .-.... -1.2 .... .:•'-• .... ...... -1.6 ....· 00 -2.0 min:-2.19 max:0.76

  41. NextStep:ModificationsforRadarObservations GettingfromObservationSpacetoElectricPotentialSpace .,. Inpractice,theelectricpotentialcanonlybeinferredfromsparseobservationsofother(related)processes.

  42. NextStep:ModificationsforRadarObservations GettingfromObservationSpacetoElectricPotentialSpace .,. Inpractice,theelectricpotentialcanonlybeinferredfromsparseobservationsofother(related)processes. .,. WeusethemethodologyinRichmondandKamide(1988)...transformbasisfunctionsintoobservationspace

  43. NextStep:ModificationsforRadarObservations GettingfromObservationSpacetoElectricPotentialSpace .,. Inpractice,theelectricpotentialcanonlybeinferredfromsparseobservationsofother(related)processes. .,. WeusethemethodologyinRichmondandKamide(1988)...transformbasisfunctionsintoobservationspace .,. Theradarsmeasureprojectionsofionosphericplasmadrift velocitiesontotheline-sight-direction:vLOS(s)=✈(s)·❛LOS.

  44. NextStep:ModificationsforRadarObservations GettingfromObservationSpacetoElectricPotentialSpace .,. Inpractice,theelectricpotentialcanonlybeinferredfromsparseobservationsofother(related)processes. .,. WeusethemethodologyinRichmondandKamide(1988)...transformbasisfunctionsintoobservationspace .,. Theradarsmeasureprojectionsofionosphericplasmadrift velocitiesontotheline-sight-direction:vLOS(s)=✈(s)·❛LOS. .,. These LOSvelocitesarerelatedtotheelectricpotentialbythefollowing: 1 ∂Φ(s) ∂Φ(s) vLOS(s)= ,− • ❛LOS · |❇(s)|∂θ ∂φ θ=Latitude,φ=Longitude,❇(s)=MagneticFieldats

  45. NextStep:ModificationsforRadarObservations TransformingBasisFunctions .,. ThisfunctionL:Φ(s)1→vLOS(s)isalinearoperator

  46. NextStep:ModificationsforRadarObservations TransformingBasisFunctions .,. ThisfunctionL:Φ(s)1→vLOS(s)isalinearoperator .,. Before: L nl Φ(s)=))ci,jWi,j[d(s,s∗j)] l=1j=1

  47. NextStep:ModificationsforRadarObservations TransformingBasisFunctions .,. ThisfunctionL:Φ(s)1→vLOS(s)isalinearoperator .,. Before: L nl Φ(s)=))ci,jWi,j[d(s,s∗j)] l=1j=1 .,. ForRadarData: L nl )) l=1j=1 { � L{Φ(s)}=v (s)= c LW[d(s,s)] ∗ i,j i,j LOS j

  48. NextStep:ModificationsforRadarObservations TransformingBasisFunctions .,. ThisfunctionL:Φ(s)1→vLOS(s)isalinearoperator .,. Before: L nl Φ(s)=))ci,jWi,j[d(s,s∗j)] l=1j=1 .,. ForRadarData: L nl )) l=1j=1 { � L{Φ(s)}=v (s)= c LW[d(s,s)] ∗ i,j i,j LOS j .,. Thecoefficientprocessisthesame...usesameestimationprocedurebutwithtransformedbasisfunctions

  49. Figure:FullelectricpotentialfieldfromLFM-MIXmodeloutput.

  50. Figure:RandomlysampledLOS directionsandcorrespondingvelocitiesusedinnumericalstudy.

More Related