Slepian Functions - the best thing since sliced bread
Document Sample


Reconstruction of bandlimited data
on a sphere
using Slepian functions
Ciarán Beggan1, Frederik Simons2, Kathy Whaler3
1 British Geological Survey, Edinburgh, UK
2 Princeton University, Princeton, USA
3 University of Edinburgh, UK
Background
• Spherical Harmonics:
– Orthogonal basis functions on a sphere
– Represent data on a sphere in a globally
continuous manner
– Well understood; e.g. useful for spectral analysis
Why Slepian functions?
• Spherical Harmonic functions can be limited by:
– Data gaps
– Very high degrees have small valued coefficients
– Differing survey data sampling density
• Slepian functions:
– Concentrate energy into a region of interest (minimise effects
of data gaps)
– Allow a trade‐off between spatial and/or spectral fidelity
– Locally and globally orthogonal
Relation to Spherical Harmonics (SH)
• For positions (θ,φ) on the unit
ˆ
r
sphere, we have data d( ) with
structure to maximum degree L
• SH basis functions (Ylm):
– globally orthogonal basis to
represent real functions on a
sphere
– Requires (L+1)2 coefficients (flm) L
d r f lmYlm r
^
ˆ
lm
• Slepian basis functions (Gα): ( L 1) 2
– locally and globally orthogonal
– Optimal # of coefficients (sα) given
s G rˆ
1
by the Shannon Number:
Slepian basis functions
• Example: Circular region
– centred on θ = 20°; φ= 15°;
– radius = 15°;
– rotation (from north) = 10°;
• Shannon Number:
– Ns = (L+1)2 ∙ f where f is fractional area of sphere
• Basis functions optimally contained in region until Shannon
number (Ns = 23 for L=36)
• Trade‐off between energy concentration in region and
leakage outwith
Basis functions concentrated within region
Concentrated
on boundary
Basis functions concentrated outside region
Computing Slepian coefficients
• Similar to deducing SH coefficients except:
– Do not need to estimate all (L+1)2 coefficients
– Only solve for α coefficients using Singular Value
Decomposition (SVD) – a least squares inversion
with truncation
g lm
• Calculate the matrix and then solve for sα
– i.e.
L
G r g lmYlm r
d YT f G T s
lm
s g lm f lm
s G T d
lm
Approximation of data using Slepian coefficients
Reconstructing a sparsely sampled function
• Example:
– Randomly sample a function in space
– Simple case
• a random combination of Slepian functions of degree L= 1‐36
• Reconstruct using just 91 coefficients (not 1368)
• Examine RMS error between input and reconstruction
Sample function at discrete Generate Slepian basis functions Generate Slepian coefficients for Reconstruct the input function in a
points within a region within circular region the region using discrete data least squares manner using the
Slepian basis functions and
coefficients
Reconstruction of a sparsely sampled Slepian function
Reconstruction of a sparsely sampled Slepian function
Reconstruction using Slepian functions with real data
• POMME Model – Z component of the field
– L = 17‐72
– ≈800 randomly sampled points in each region
– 0.5° grid resolution
1
Bangui Anomaly
± 300nT
Region 1
• Leakage of energy out of region
Region 1
• Leakage of energy out of region
Increasing truncation level
Shannon Number
• Improves approximation within the region
– but induces severe leakage outside
Observations
• Reconstruction using 91 coefficients (SN =
91) not perfect – plenty of remaining
structure in the error
• Leakage of signal outside region is minimal
• Increasing grid resolution does not improve
RMS error (e.g. 0.1° lat/lon)
• However, increasing the truncation level
does improve the reconstruction – but with
leakage outside the region
Conclusions
• Allows trade‐off of spatial and
spectral energy in a orthogonal
spherical basis
• Compact coefficient representation
• Reconstruction using real data
appears to require N > Shannon
number (depends on Signal‐to‐Noise
ratio)
Future Work
• Estimation of specta in separate regions
– Ocean/continent power spectra
• Exclusion/down‐weighting of noisy data
– Use region selection to remove electro‐jet and/or auroral
regions
Acknowledgements
• Thanks to NERC (GEOSPACE)
• References:
– Dahlen, F.A. & F.J. Simons, Spectral estimation on a sphere in geophysics and
cosmology, Geophys. J. Int., 2008, 174 (3), 774–807
– Simons, F.J. & F.A. Dahlen, A spatiospectral localization approach to
estimating potential fields on the surface of a sphere from noisy, incomplete
data taken at satellite altitudes, Proc. of SPIE, 2007, 6701 (670117)
– Simons, F.J. & F.A. Dahlen, Spherical Slepian functions and the polar gap in
geodesy, Geophys. J. Int., 2006, 166
– Simons, F.J., F.A. Dahlen & M.A. Wieczorek, Spatiospectral concentration on a
sphere, SIAM Review, 2006, 48 (3), 504–53
Get documents about "