Influence sampling of trailing variables of dynamical systems

Paul Krause. "Influence sampling of trailing variables of dynamical systems." Mathematics of Climate and Weather Forecasting 3.1 (2017): 51-63. <http://eudml.org/doc/288426>.

@article{PaulKrause2017,
abstract = {For dealing with dynamical instability in predictions, numerical models should be provided with accurate initial values on the attractor of the dynamical system they generate. A discrete control scheme is presented to this end for trailing variables of an evolutive system of ordinary differential equations. The Influence Sampling (IS) scheme adapts sample values of the trailing variables to input values of the determining variables in the attractor. The optimal IS scheme has affordable cost for large systems. In discrete data assimilation runs conducted with the Lorenz 1963 equations and a nonautonomous perturbation of the Lorenz equations whose dynamics shows on-off intermittency the optimal IS was compared to the straightforward insertion method and the Ensemble Kalman Filter (EnKF). With these unstable systems the optimal IS increases by one order of magnitude the maximum spacing between insertion times that the insertion method can handle and performs comparably to the EnKF when the EnKF converges. While the EnKF converges for sample sizes greater than or equal to 10, the optimal IS scheme does so fromsample size 1. This occurs because the optimal IS scheme stabilizes the individual paths of the Lorenz 1963 equations within data assimilation processes.},
author = {Paul Krause},
journal = {Mathematics of Climate and Weather Forecasting},
keywords = {data assimilation; determining variables; Dyson’s formula},
language = {eng},
number = {1},
pages = {51-63},
title = {Influence sampling of trailing variables of dynamical systems},
url = {http://eudml.org/doc/288426},
volume = {3},
year = {2017},
}

TY - JOUR
AU - Paul Krause
TI - Influence sampling of trailing variables of dynamical systems
JO - Mathematics of Climate and Weather Forecasting
PY - 2017
VL - 3
IS - 1
SP - 51
EP - 63
AB - For dealing with dynamical instability in predictions, numerical models should be provided with accurate initial values on the attractor of the dynamical system they generate. A discrete control scheme is presented to this end for trailing variables of an evolutive system of ordinary differential equations. The Influence Sampling (IS) scheme adapts sample values of the trailing variables to input values of the determining variables in the attractor. The optimal IS scheme has affordable cost for large systems. In discrete data assimilation runs conducted with the Lorenz 1963 equations and a nonautonomous perturbation of the Lorenz equations whose dynamics shows on-off intermittency the optimal IS was compared to the straightforward insertion method and the Ensemble Kalman Filter (EnKF). With these unstable systems the optimal IS increases by one order of magnitude the maximum spacing between insertion times that the insertion method can handle and performs comparably to the EnKF when the EnKF converges. While the EnKF converges for sample sizes greater than or equal to 10, the optimal IS scheme does so fromsample size 1. This occurs because the optimal IS scheme stabilizes the individual paths of the Lorenz 1963 equations within data assimilation processes.
LA - eng
KW - data assimilation; determining variables; Dyson’s formula
UR - http://eudml.org/doc/288426
ER -