open problems in light scattering by ice particles chris westbrook department of meteorology
TRANSCRIPT
Open problems in light scattering by ice
particles
Chris Westbrook
www.met.reading.ac.uk/radarDepartment ofMeteorology
Overview of ice cloud microphysics:
CirrusParticles nucleated
at cloud top
Growth by diffusion of vapour onto ice surface – pristine crystals
Aggregation of crystals
Evaporation of ice particles (ie. no rain/snow at ground)
• Redistributes water vapour in troposphere• Covers ~ 30% of earth, warming depends on microphysics• Badly understood & modelled
verticallypointing
radar
=8.6mm
Overview of ice cloud microphysics:
thick stratiform cloud
Nucleation & growth of pristine crystals(different temp & humidity -> different habits)
Aggregation
Melting layer- snow melts into rain (most rain in the UK starts as snow)- if T<0°C at ground then will precipitate as snow- if the air near the ground is dry then may evaporate on the way
ICE
RAIN
If supercooled water droplets are present may get ‘riming’:
• Important for precipitation forecasts
Need for good scattering models
• Need models to predict scattering from non-spherical ice crystals if we want to interpret radar/lidar data, particularly:
– Dual wavelength ratios size ice content
– Depolarisation ratio LDR– Differential reflectivity ZDR
}Particle shape& orientation
+ analogous quantities for lidar
Radar wavelengths3 GHz=10 cm
35 GHz=8.6 mm
94 GHz=3.2 mm
~ uniform E field across particle at 3, 35 GHz-> Rayleigh Scattering
Applied E-field varies over particle at 94 Ghz
ie. Non-Rayleigh scattering
1 mm ice particle
Applied wave(radar pulse)
kR from 0 to 5 for realistic sizes
Lidar Wavelengths
• Small ice particles from 5m (contrails) to 10mm ish (thick ice cloud)
• Lidar wavelengths 905nm and 1.5m
• Wavenumbers k=20 to k=70,000
• Big span of kR need a range of methods
Current methodology:
Radar: approximate to idealised shapes
Prolate spheroid (cigar)
Oblate spheroid (pancake)
Sphere Mie theory for both Rayleigh and Non-Rayleigh regimes
Exact Rayleigh solutionT-Matrix for Non-Rayleigh
Rayleigh scattering
2 0
ICE AIR
n n
E
Applied field is uniform across the particle so have an electrostatics problem:
ICE AIR
ice permittivity
BCs: On the particle surface:
BCs: Far away from particle
E = applied field
E (applied)
n = normal vector to ice surface
Analytic solution for spheres, ellipsoids. In general?
Non-Rayleigh scattering
• Exact Mie expansion for spheres• So approximate ice particle by a sphere• Prescribe an ‘effective’ permittivity
– Mixture theories: Maxwell-Garnett etc.
• Pick the appropriate ‘equivalent diameter’
• How do you pick equiv. D? Maximum dimension? Equal volume? Equal area?
Non-spherical shapesRayleigh-Gans (Born) approximation:
Assume monomers much smaller than wavelength(even if aggregate is comparable to )
For low-densities, Rayleigh formula is reduced bya factor 0 < f < 1 because the contributions from the different crystals are out of phase
Crystal at point r sees the applied field at origin shifted by k . r radiansSo for backscatter, each crystal contributes ~ K dv exp(i2k.r)
so,
this is great because it's just a volume integral :-)
(ie. essentially the Fourier transform of thedensity-density correlation function)
Guinier regime Scaling regime
(kR)-21- (kR)23
4
Rayleigh – Gans results
Nice, but we've neglected coupling between crystals (each crystal sees only the applied field).
Westbrook, Ball, Field Q. J. Roy. Met. Soc. 132 897
fit a curve with the correctasymptotics in both limits:
Current approach for lidar:
Ray tracing of model particle shapes
•Hexagonal prisms•Bullet-rosettes•Aggregates•etc.
measured phase functions usually find no halos.surface ‘rougness’ ?
Is this real? And if so, at what k does it become important?
Geometric optics:
Q. is how good is G.O. at lidar wavelengths,where size parameter is finite?
Better methods: FDTD
• Solves Maxwell curl equations• Discretise to central-difference equations• Solve using leap-frog method
(ie solve E then H then E then H…)• Nice intuitive approach• Very general
• But…– Need to grid whole domain and solve for E and H
everywhere – Some numerical dispersion– Fixed cubic grid, so complex shapes need lots of
points– Stability issues– Very computationally expensive, kR~20 maximum
t
t
DH
HE
BEM
• Boundary element methods• Has been done for hexagonal prism
crystal• E and H satisfy the Helmholtz equation• Problem with sharp edges/corners of
prism (discontinuities on boundary)• Have to round off these edges & corners
to get continuous 2nd derivs in E and H• This doesn’t seem to affect the phase
function much so probably ok.
only one study so far!Mano (2000) Appl. Opt.
T-matrix
• Expand incident, transmitted and scattered fields into a series of spherical vector wave functions, then find the relation between incident (a,b) and scattered (p,q) coefficients
• Once know transition matrix T then can compute the complete scattered field
• Elements of T essentially 2D integrals over the particle surface
• Easy for rotationally symmetric particles (spheroids, cylinders, etc)
• But…– Less straightforward for arbitrary shapes– Numerically unstable as kR gets big
• OK up to kR~50 if the shape isn’t too extreme
2
2 22
11 1
1
T Tp a
T Tq b
Discrete dipole approximation
• Recognise that a ‘point scatterer’ acts like a dipole
• Replace with an array of dipoles on cubic lattice
• Solve for E field at every point dipole know scattered field
DDA continued…
( ) ( ) ( )j applied j jk k kk j
E r E r A E r
polarisability of dipole k
Electric field at j
Applied field at j
Tensor characterising fall off of the E field from dipole k, as measured at j
( ) ( ) ( )j applied j jk k kk j
E r E r A E r
• Model complex particle with many point dipoles
• Each has a dipole moment of (Ej is field at jth dipole)
• Every dipole sees every other dipole, ie total field at the lth dipole is:
applied
etc..
3d ( )j jp vK E
So need a self-consistent solution for Ej at every dipole
- Amounts to inverting a 3N x 3N matrix A
DDA for ice crystal aggregates
Discrete dipole calculations allow us to estimate the ‘true’ non-Rayleigh factor:
Rayleigh-Gans
discrete dipoleestimates
Want to parameterise a multiple scattering correction so we can map R-G curve to the real data based on:
volume fraction of ice (v/R )size relative to wavelength (kR)
3
Mean field approach to multiple scattering following the approach of Berry & Percival Optica
Acta 33 577
Mean-field approximation – every crystal sees same scalar multiple of applied field:
ie. multiple scattering increases with: - Polarisability of monomers via K()- Volume fraction F- Electrical size via G(kR)
so what's G(kR) ?
(essentially d.d.a. with 1 dipole per crystal)
v
Leading order form for G(kR)
Rayleigh-Gans corrected by d
2
Rayleigh-Gans
Fractal scaling leads to strong clustering and a probability density of finding to crystals a distance r apart:
this means that, to first order:
ie.
This crude approximation seems to work pretty well
-strong clustering and fact that kR is -fairly moderate have worked in our favour
(x=r/R)
Fit breaks down as D
DDA pros & cons
• Physical approach, conceptually simple• Avoids discretising outer domain• Can do any shape in principle
• Needs enough dipoles to – 1. represent the target shape properly– 2. make sure dipole separation <<
• Takes a lot of processor time, hard to //ise• Takes a lot of memory ~ N3 (the real killer)
• Up to kR~40 for simple shapes
Rayleigh Random WalksWell known that can use random walks to sample electrostatic potentialat a point.
For conducting particles () Mansfield et al [Phys. Rev. E 2001] havecalculated the polarisability tensor using random walker sampling.
Advantages are that require ~ no memory and easy to parallelise (each walker trajectory is an independent sample, so can just task farm it)
Problems: how to extend to weak dielectrics (like ice)? Jack Douglas (NIST)
Efficiency may be poor for small
- +
Transition probability at boundary
1
Conclusions
• Lots of different methods – which are best?• Computer time & memory a big problem• Uncertain errors• Better methods? FEM, BEM…?• Ultimately want parameterisations for
scattering in terms of aircraft observables eg. size, density etc.
• Would like physically-motivated scheme to do this (eg. mean-field m.s. approx etc)