Premium
Evaluation and optimization of sampling errors for the Monte Carlo Independent Column Approximation
Author(s) -
Räisänen Petri,
Barker Howard W.
Publication year - 2004
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1256/qj.03.215
Subject(s) - monte carlo method , variance reduction , sampling (signal processing) , statistics , variance (accounting) , importance sampling , radiative transfer , mathematics , markov chain monte carlo , standard deviation , environmental science , meteorology , computer science , physics , accounting , filter (signal processing) , quantum mechanics , business , computer vision
The Monte Carlo Independent Column Approximation (McICA) method for computing domain‐average broadband radiative fluxes is unbiased with respect to the full ICA, but its flux estimates contain conditional random noise. McICA's sampling errors are evaluated here using a global climate model (GCM) dataset and a correlated‐ k distribution (CKD) radiation scheme. Two approaches to reduce McICA's sampling variance are discussed. The first is to simply restrict all of McICA's samples to cloudy regions. This avoids wasting precious few samples on essentially homogeneous clear skies. Clear‐sky fluxes need to be computed separately for this approach, but this is usually done in GCMs for diagnostic purposes anyway. Second, accuracy can be improved by repeated sampling, and averaging those CKD terms with large cloud radiative effects. Although this naturally increases computational costs over the standard CKD model, random errors for fluxes and heating rates are reduced by typically 50% to 60%, for the present radiation code, when the total number of samples is increased by 50%. When both variance reduction techniques are applied simultaneously, globally averaged flux and heating rate random errors are reduced by a factor of ∼3. Copyright © 2004 Royal Meteorological Society