- Title
- Using palaeoclimate information to improve stochastic modelling for water management
- Creator
- Armstrong, Matthew
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2024
- Description
- Research Doctorate - Doctor of Philospohy (PhD)
- Description
- Various palaeoclimate reconstructions have identified the occurrence of ‘megadroughts’. These ‘megadroughts’ are much longer and more severe than those recorded via the instrumental measurements. Because instrumental measurements are used to infer climate risk when designing water supply infrastructure and management plans, a ‘megadrought’ invokes a concern for water security. However, such concerns should be viewed within the context of the (a) limitations of using palaeoclimate proxy records as a source of climate information and (b) existing methods used in water management to estimate climate risk (i.e. inferring climate risk using a stochastic model calibrated to instrumental measurements). The goal of this thesis is to (a) use palaeoclimate proxy records to evaluate stochastic model performance and parameter stationarity under long term, centennial scale variability and (b) present a stochastic modelling framework that incorporates proxy centennial scale variability. To achieve this goal, this thesis had five objectives: 1. Evaluate the persistence signal in Antarctic ice core records. It was found that the persistence signal in annual snowfall accumulation and mid-latitude rainfall records are statistically similar. In contrast, ice core Na+ records tended to have slightly higher persistence than mid latitude rainfall. This analysis informed the subsequent use of ice core information in palaeoclimate informed stochastic modelling and climate risk assessment. 2. Evaluate different stochastic models using millennium-length palaeoclimate proxy records. It was found that a stochastic model calibrated to instrumental measurements cannot simulate long term, centennial scale climate variability. This means that traditional stochastic modelling approaches (i.e. calibrating to a ~100 year instrumental record) are unable to simulate risk arising from aleatory uncertainty and centennial scale climate variability. However, several models capable of simulating long term, centennial scale climate variability when calibrated to extended, multi centennial timeseries were identified. Two such models, the ARMA(1,1) and ARFIMA(0,D,0) models, were used for subsequent objectives. 3. Evaluate the role of sampling bias, conditioning error, and likelihood approximation when inferring stochastic model parameters under centennial-scale variability. Using synthetic timeseries (generated from an ARMA(1,1) or ARFIMA(0,D,0) model) and Bayesian calibration methods, it was found that exact and conditional, approximate likelihoods return similar posteriors. 4. Evaluate stochastic model parameter stationarity using millennium-length palaeoclimate proxy records. It was found that stochastic model mean and standard deviation are likely (a) non stationary at multi centennial and millennial timescales and (b) stationary at centennial timescales. Furthermore, stochastic model persistence is likely stationary over centennial, multi centennial, and millennial timescales. 5. Calibrate stochastic model persistence using ice core information within a Bayesian framework. For the final objective, a Bayesian framework for calibrating a stochastic rainfall model using palaeoclimate proxy data is presented. The framework uses proxy data from an Antarctic ice core and instrumental measurements from southeast Australia to calibrate a catchment-scale stochastic rainfall model. The proxy data is used to define a Bayesian prior for instrumental persistence. This extracts the proxy persistence signal, which is representative of broader regional persistence, without using the proxy to predict catchment-scale rainfall. When validated, the proposed model reproduces the observed drought risk. However, compared with the ‘standard’ model calibrated using a non-informative persistence prior, the palaeoclimate informed model can simulate much longer and more severe droughts. When answering Objectives 4 and 5, it became apparent that centennial scale variability, aleatory uncertainty, and parameter uncertainty results in irreducibly ‘wide’ statistical uncertainty. This means that water supply systems must be robust under a future range of drought risk that is irreducibly ‘wide’. In the final discussion, ‘wide’ uncertainty is discussed within the context of ‘deep’ uncertainty associated with risk arising from anthropogenic climate change and the ‘murky’ uncertainty associated with imperfect knowledge, system complexity, and the subjective nature of socio political values. Approaches for managing water under ‘wide, deep and murky’ uncertainty are also discussed.
- Subject
- palaeoclimate; stochastic modelling; water management; megadroughts
- Identifier
- http://hdl.handle.net/1959.13/1497136
- Identifier
- uon:54302
- Rights
- Copyright 2024 Matthew Armstrong
- Language
- eng
- Full Text
- Hits: 2451
- Visitors: 2480
- Downloads: 84
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | ATTACHMENT01 | Thesis | 7 MB | Adobe Acrobat PDF | View Details Download | ||
| View Details Download | ATTACHMENT02 | Abstract | 343 KB | Adobe Acrobat PDF | View Details Download |