Open AccessLow-Frequency Climate Response and Fluctuation–Dissipation Theorems: Theory and Practice
Author(s) -
Andrew J. Majda,
Boris Gershgorin,
Yuan Yuan
Publication year - 2010
Publication title -
journal of the atmospheric sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.853
H-Index - 173
eISSN - 1520-0469
pISSN - 0022-4928
DOI - 10.1175/2009jas3264.1
Subject(s) - climate model , forcing (mathematics) , nonlinear system , linear subspace , mathematics , frequency response , climate change , linear regression , set (abstract data type) , computer science , econometrics , statistics , physics , mathematical analysis , ecology , geometry , quantum mechanics , electrical engineering , biology , programming language , engineering
The low-frequency response to changes in external forcing or other parameters for various components of the climate system is a central problem of contemporary climate change science. The fluctuation–dissipation theorem (FDT) is an attractive way to assess climate change by utilizing statistics of the present climate; with systematic approximations, it has been shown recently to have high skill for suitable regimes of an atmospheric general circulation model (GCM). Further applications of FDT to low-frequency climate response require improved approximations for FDT on a reduced subspace of resolved variables. Here, systematic mathematical principles are utilized to develop new FDT approximations on reduced subspaces and to assess the small yet significant departures from Gaussianity in low-frequency variables on the FDT response. Simplified test models mimicking crucial features in GCMs are utilized here to elucidate these issues and various FDT approximations in an unambiguous fashion. Also, the shortcomings of alternative ad hoc procedures for FDT in the recent literature are discussed here. In particular, it is shown that linear regression stochastic models for the FDT response always have no skill for a general nonlinear system for the variance response and can have poor or moderate skill for the mean response depending on the regime of the Lorenz 40-model and the details of the regression strategy. New nonlinear stochastic FDT approximations for a reduced set of variables are introduced here with significant skill in capturing the effect of subtle departures from Gaussianity in the low-frequency response for a reduced set of variables.