Skip to main content
Have a personal or library account? Click to login
The linearisation of maps in data assimilation Cover

The linearisation of maps in data assimilation

Open Access
|Dec 2013

Figures & Tables

Fig. 1

(a) pdf of generalised normal distribution (11) with p=1 for σ=0.8 (blue), σ=0.4 (green) and σ=0.1 (red); (b) cloud fraction Φ (black) and Φ(0) for various σ as in (a); (c) Φ′ (black) and Φ(1) for various σ as in (a); (d) linearisation error assuming true pdf has σ=0.4, using the TLA (black) and the BLA with various σ as in (a).

Fig. 2

100 000 iterates of the Duffing map (33) with a=2.75, b=0.15 starting from (12,12 (plus sign).

Fig. 3

Mean square analysis error for the Duffing map example, using TLA filter (black) and BLA filter (blue) for various values of the linearisation error parameter α. The dotted curve is the lowest mean square analysis error achievable using the TLA and any value of β.

Table 1

Minimum variance smoother

Given first guesses x^00,P00:Fori=0,1, … Predict:   x^i+1i=E[fi(x^ii+εi]       Pi+1i=E[(fi(x^ii+εi-x^i+1i)×       (fi(x^ii+εi-x^i+1i)T]+Qi Smooth:   Ui=E[εi(fi(x^ii+εi)T]       Li=Ui(Pi+1i+Ri+1-1       x^ii+1=x^ii+Li(yi+1-x^i+1i       Pii+1=Pii-Ui(Pi+1i+Ri+1-1UiT Update:   Ki+1=Pi+1i(Pi+1i+Ri+1-1       x^i+1i+1=x^i+1i+Ki+1(yi+1-x^i+1i       Pi+1i+1=(I-Ki+1Pi+1iEnd For i
Table 2

Smoother using the BLA

Given first guesses x^00,P00:Fori=0,1,… Predict:   x^i+1i=fi(0(x^ii       Pi+1i=fi(1(x^iiPiifi(1(x^iiT+Qi+Q˜iBL Smooth:   Ui=Piifi(1(x^iiT       Li=Ui(Pi+1i+Ri+1-1       x^ii+1=x^ii+Li(yi+1-x^i+1i       Pii+1=Pii-Ui(Pi+1i+Ri+1-1UiT Update:   Ki+1=Pi+1i(Pi+1i+Ri+1-1       x^i+1i+1=x^i+1i+Ki+1(yi+1-x^i+1i       Pi+1i+1=(I-Ki+1Pi+1iEnd For i
Table 3

Variational form of smoother using the BLA

Given first guesses x^00,P00:Fori=0,1, …       Find δi,δi+1 which minimise (44)       x^ii+1=x^ii+δi       Pii+1=[(J(δi,δi+1)-1]11       x^i+1i+1=fi(0(x^ii+δi+1       Pi+1i+1=[(J(δi,δi+1)-1]22End For i
Fig. 4

Mean square analysis error for the Duffing map example for various values of the linearisation error parameter α. Filters are in black and smoothers in blue. Dotted curves use the TLA and solid curves use the BLA.

Fig. 5

Mean square analysis error in roughened Duffing map (with p=100) for various values of the linearisation error parameter α. Filters are in black and smoothers in blue. Dotted curves use the TLA and solid curves use the BLA.

Language: English
Page range: 18840 - 18840
Submitted on: May 24, 2012
Accepted on: Mar 11, 2013
Published on: Dec 1, 2013
Published by: Stockholm University Press
In partnership with: Paradigm Publishing Services

© 2013 Timothy J. Payne, published by Stockholm University Press
This work is licensed under the Creative Commons Attribution 4.0 License.