Accurate, stochastic representations of rainfall structures and weather patterns in the space-time dimension are a challenging task. Recently, efforts have been focused on the simulation of large spatial fields, representation of higher-order statistics, simulation of spatial extremes and overcoming the problem of overdispersion - an underrepresentation of inter- and intraannual variance in weather generator simulations. In this dissertation, these issues are adressed by presenting three different multisite methodologies - a 'conventional' rainfall generator using orthogonal Markov chains with Richardson-type separation in event-amount generation (multisite, PXEOF-enhanced orthogonal Markov chain model methodology), a more novel approach using multivariate EOFs to express precipitation in the region as a two-component combination of deterministic evolution patterns and corresponding stochastic amplitude coupled with an autoregressive moving average model (multisite, ARMA-enhanced PXEOF model methodology) and, finally, a multivariable extension for the simulation of four meteorological variables with improved interannual variability on the station level (multivariable, multisite PXEOF-EEOF model methodology). Based on above methodologies, 1,000 to 10,000 years of daily simulated weather for 196 stations (20 stations in the case of the multisite, multivariable framework) in Peninsular Malaysia were generated. Statistical characteristics of the synthetic dataset are examined in comparison with the observational record and comparisons between models are made. Regarding the ARMA-enhanced PXEOF model methodology, the need for an autoregressive model component to improve short-term rainfall dependence is demonstrated and model evaluation is focused on a slightly 'neglected' topic, often missing from model evaluations in the literature - spatial rainfall footprints and areal statistics. For the multivariable, multisite PXEOF-EEOF model methodology, the spatial and cross-variable correlation structure as well as the effect of introducing interannual correlations is investigated in further detail. The thesis concludes by summarizing benefits and challenges of using multivariate EOFs in weather generators and a recommendation for the shift towards a more parsimonious model framework with modular structure is made.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:693997 |
| Date | January 2015 |
| Creators | Schmid, Matthias |
| Contributors | Toumi, Ralf |
| Publisher | Imperial College London |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/10044/1/39974 |
Page generated in 0.002 seconds