Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Australie → Statistical testing of downscaled rainfall data for water security assessment

University of Newcastle (2017)

Statistical testing of downscaled rainfall data for water security assessment

Parana Manage, Nadeeka

Titre : Statistical testing of downscaled rainfall data for water security assessment

Auteur : Parana Manage, Nadeeka

Université de soutenance : University of Newcastle

Grade : Doctor of Philosophy (PhD) 2017

Résumé
Rainfall is the key input for the hydrology predictions used in water supply reservoir simulations. In principal, rainfall data can be obtained from General Circulation Models (GCMs ; grid resolution of 250 km) at the global scale, and at finer resolutions from Regional Climate Models (RCMs). This thesis assesses the ability of RCM simulations to accurately reproduce observed rainfall statistics (temporal and spatial) important for water availability analysis in reservoir simulations. The dynamically downscaled 10 km resolution RCM simulations used in this thesis have been produced by the New South Wales (NSW)/Australian Capital Territory (ACT) Regional Climate Modelling (NARCliM) project. The assessment is performed for four time periods, two current day periods (1950-2009 and 1990-2009) and two future periods (2020-2040 and 2060-2080) ; four GCMs ; three configurations of the RCMs (called R1, R2, R3) for each GCM ; and five study regions along the NSW eastern seaboard. This thesis has three main foci. The first focus was on the assessment of the ability of NARCliM RCM data to generate the correct statistical properties of current day rainfall. Understanding the properties of the NARCliM generated rainfall and how they differ from observed rainfall is important when using the NARCliM data for hydrology modelling, particularly for models calibrated or designed using observed rainfall. Time series and spatial statistics of RCM simulations are compared against the ground based measurements for selected Bureau of Meteorology rainfall stations and 5 km gridded data from the Australian Water Availability Project (AWAP). The second focus was an assessment of the impacts of orography on regional rainfall, both observed and NARCliM rainfall. The third focus was on how the NARCliM rainfall properties will change in the future. Statistical testing of the NARCliM rainfall performed along the east coast of Australia suggest that bias corrected rainfall from the NARCliM RCMs are better at reproducing the daily rainfall distribution of AWAP rainfall than uncorrected, yet fail to reproduce the entire probability distribution of the rainfall, particularly the probabilities of the light rainfall events (rainfall less than 1 mm). The bias corrected NARCliM rainfall is able to reproduce both the magnitudes and spatial pattern of distribution statistics such as mean, standard deviation and coefficient of variation. The spatial correlation function of AWAP rainfall is well captured by both uncorrected and bias corrected NARCliM rainfall, however, both AWAP and NARCliM overestimate the correlation between rain gauges by 20-30%. Both the uncorrected and bias corrected NARCliM rainfall fail to reproduce the time series statistics such as autocorrelations of the rainfall. NARCliM overestimates the magnitudes of autocorrelations of AWAP, even though all NARCliM rainfall series are able to capture the seasonal cycle in the autocorrelation. Similar results for both the uncorrected and bias corrected NARCliM rainfall suggest that the bias correction does not influence the time series statistics, particularly the autocorrelations. The spatial pattern of both the time series and spatial statistics was generally well reproduced by the bias corrected data, though at high elevations there were some significant differences with AWAP. The ability to reproduce the spatial pattern of the rainfall statistics is an important but not sufficient requirement for hydrological studies, particularly reservoir modelling, since many reservoirs are located in areas of localised high rainfall. The relationship between the mean rainfall and elevation, and impacts from East Coast Lows (ECLs) on this relationship was investigated. The results suggest that there is a statistically significant positive linear relationship between observed rainfall and elevation, in a region close to the coast where the ECL impacts are known to be dominant. When the distance from the coast increases, this relationship becomes weaker. Compared with AWAP, the RCM R2 is best able to replicate the linear relationship between the gauged rainfall and elevation. A relationship for the rainfall combining distance from the coast and elevation difference between two points was also found. The RCM future projections suggest that mean rainfall will change by approximately -19% to 54% and the monthly rainfall coefficient of variation will change by -19% to 40%. All GCMs, except Commonwealth Scientific and Industrial Research Organisation (CSIRO-Mk3.0) project rainfall increases along the coast. There is a general trend that regions with reduced coefficient of variation for rainfall have a higher mean rainfall and vice versa, for all sites except for Sydney.

Mots clés : climate ; water security ; regional climate modelling

Présentation

Version intégrale (24 Mb)

Page publiée le 10 février 2018