An analysis of the output of the Hadley Centre Unified Model forecast for Southern Africa using Nonlinear Primary Component Analysis (NLPCA) for feature recognition.
Abstract
Analysis of general circulation model (GCM) performance often focuses on global scale features, using temporal means and anomalies. The evaluation of GCMs for regional climate forecast requires the recognition of important meteorological features at relatively short timescales. The simplest analysis of means and anomalies are not as relevant to validation of global model data at regional and seasonal scales. Visual analysis at relevant temporal levels may reveal synoptic features representative of observations that are more relevant to regional forecasts. A statistical method, such as the use of empirical orthogonal functions (EOFs) for primary component analysis is generally used to identify the temporal and spatial modes. A limitation of EOFs is feature recognition is restricted to linear functions of variable fields. The development and evolution of synoptic features are expected to be nonlinear. We use Nonlinear Primary Component Analysis (NLPCA) derived from a neural network (Hsieh, 2001) to analyse Hadley Centre Unified Model output fields in comparison to NCEP reanalysis data. This procedure has the capability to identify features that are nonlinear functions of model diagnostic fields, such as chaotic bifurcations or cycles, for example El Niño. NLPCA will allow the recognition of the primary spatial-temporal components of a physical-dynamical model. Seasonal GCM forecasts of the meteorology of southern Africa are evaluated to identify primary synoptic features and their relationship to ocean climate state. The objective is to determine the extent to which the global GCM can forecast relevant features at the regional scale of seasonal climate variation.
- Publication:
-
AGU Spring Meeting Abstracts
- Pub Date:
- May 2001
- Bibcode:
- 2001AGUSM...A41A05C
- Keywords:
-
- 3309 Climatology (1620);
- 3329 Mesoscale meteorology;
- 3339 Ocean/atmosphere interactions (0312;
- 4504);
- 4203 Analytical modeling;
- 4263 Ocean prediction