Quantifying sources of uncertainty in regional climate model scenarios for Ireland.
PhD thesis, National University of Ireland Maynooth.
This thesis develops a novel framework for model skill assessment and the
generation of probabilistic future climate scenarios. Traditional approaches to model
validation assume that skill in simulating the mean climate is a valid indicator of skill
in modelling the climate system. However, without information about how errors
arise, conclusions cannot be drawn about whether models are genuinely skilful.
Initially, verification statistics are used to assess model skill in simulating
seasonal means and variability of Irish climate for 1961-1990. Significant biases
were identified, however without further analysis, these biases cannot be attributed to
a cause. Therefore, a spatial analysis, including EOF analysis, was undertaken which
indicated that biases may be either spatially consistent (systematic) or inconsistent
(random), an important distinction. Next, representation of a key large-scale driver of
Irish climate, the North Atlantic Oscillation, was examined for a representative subsample
of models. Skill in simulating the NAO was found to vary considerably
between models. Therefore, assessing statistics of mean climate may not be the
optimum way to characterize model skill, as deficiencies in the representation of
large-scale drivers may not be detected.
Both quantitative and qualitative information from the skill assessments was
used to inform probabilistic ensemble projections of future climate using Bayesian
Model Averaging. In some cases, weighting scheme variation affects the ensemble
PDF shape. In other cases, PDFs are similar when different weights are used, but the
relative contributions of ensemble members vary. This is a crucial finding, as this
underlying variation may not be immediately apparent, but may affect the confidence
attached to the PDF. Therefore, robustness of ensemble generation methods must be
considered when determining the level of confidence attached to a projection.
Finally, the implications of these results for climate decision-making are
discussed and recommendations for the use of climate models in decision-making are
||regional climate models;
||Faculty of Social Sciences > Geography
||14 Feb 2011 12:03
Repository Staff Only(login required)
||Item control page