Approaches to quantifying and reducing uncertainty in GCMs over Southern Africa

Includes abstract.Includes bibliographical references (p.127-131).Global Circulation Models (GCMs) are the primary tool for simulating future climate changes. These models by necessity make use of various assumptions and simplifications due to computational constraints, and in so doing introduce biases and systematic error. Along with other sources of uncertainty regarding our understanding of the climate system and given the quasi-chaotic nature of the climate, climate projections differ between models whose climate simulation skill is poorly quantified. A new methodology is presented to assess the regional biases in GCMs and to, in part, compensate for some aspects of these biases. The study will focus on the Southern African region but could be replicated for other regions. Using Self-Organising Maps (SOMs), synoptic archetypal patterns are identified and the distribution and frequency of these patterns assessed. The use of synoptic archetypes to quantify model metrics presents a novel approach with many benefits over standard metrics, such as errors and means per variable. SOMs add a spatial and multi-variable dimension to the analysis as each metric is calculated based on its synoptic circulation pattern and associated to a set of atmospheric variables. Some persistent biases in the models are notable based on comparisons between the NCEP and GCM SOM node mapping, such as an overall cool bias in the models and a shift of the dominant high pressure cells and thus the westerly wave to the south. The weighting techniques provide insight into how much of the model bias is contributed by differences in synoptic frequency and what part is attributable to systematic biases in the models which result in a different mean state for a given synoptic process. The frequency correction enabled a correction of up to 25% of the difference between model and reanalysis data, but in most cases the change was far smaller than this. The differences in mean conditions remained the largest component of the bias. To correct for this the weighting was applied to the climate change anomaly (difference between future and control projections) per synoptic process to create a multi-model climate change component that is added to the NCEP baseline. This provides the most accurate depiction of future climate from the data provided. The models generally have different strengths, therefore the weighted multi-model solution allows models to give a greater contribution where they are skilful and less where they do not match the observed dynamics. Comparison of the magnitude of the climate change signal showed that winter states in the weighted multi-model composite had a smaller temperature increase and reduced rainfall compared to the unweighted results. In summer states there is greater warming and increased rainfall, especially over the oceans. This suggests the models are over estimating changes in temperature in winter and underestimating the increases in summer. Synoptic events are the primary driver of climate change impacts. Therefore errors in synoptic state will have a notable influence on the climate change projections and need to be fully considered in any climate change impact study. The use of the weighting technique helped to identify and reduce uncertainties in the climate change projections over Southern Africa