Error Analysis for an Algorithm That Reduces Radiative Transfer Calculations in High‐Resolution Atmospheric Models
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
AbstractThis study assesses characteristics of spatial and temporal errors for net (shortwave + longwave) surface irradiances and atmospheric heating rates predicted by the Partitioned‐Gauss‐Legendre Quadrature (PGLQ) algorithm, whose aim is to economize computation of radiative transfer (RT) in high‐resolution cloud system‐resolving models (CSRMs). Most results reported here are for PGLQ having made 200 times fewer calls to the RT models than the Independent Column Approximation, which applies the RT models to each column in CSRM domains. Due to the nature of GLQ, PGLQ yields almost unbiased domain averages. This is shown for data pertaining to two convective cloud systems produced by a CSRM. Relative to the Independent Column Approximation, PGLQ works with much reduced amounts of information, and so it has the potential to produce rare, but sizable, localized errors. Two methods are employed to assess the “randomness” of spatial and temporal series of PGLQ flux and heating rate errors; one of which is developed here. Regarding spatial transects of net surface irradiance errors ΔFNET, ~30–60% of them are considered by both assessment techniques, simultaneously, to be indistinguishable, at the 95% confidence level, from fully uncorrelated sequences. Correspondingly, for time series of ΔFNET (at 8‐s time step), ~15–20% of 128‐step series, and ~10–70% of 30‐step series are considered by both methods, simultaneously, to be random. For time series of errors for net radiative heating rates, ~40% of cloudless series get classed, by both tests simultaneously, as random, compared to ~20% of those that contain some cloud.
