Please use this identifier to cite or link to this item:
http://hdl.handle.net/11375/25121
Title: | Development of Multi-Model Ensembles for Climate Downscaling in Ontario, Canada |
Authors: | Li, Xinyi |
Advisor: | Li, Zhong |
Department: | Civil Engineering |
Publication Date: | 2020 |
Abstract: | Climate change has widespread impacts on the environment, economy, and municipal planning. Thus, generating accurate, high resolution climate predictions could aid in the assessment of the impacts of climate change on a local scale. Multi-model ensembles have been proven to improve the accuracy of climate prediction, and machine learning techniques are a promising tool for temperature downscaling. The thesis investigates machine learning and statistical methods in the development of multi-model ensembles for climate downscaling. Firstly, three neural network algorithms are used to develop multi-model ensembles for daily mean temperature downscaling, including Multi-layer Perceptron (MLP), Time-lagged Feed-forward Neural Network (TLFN) and Nonlinear Auto-Regressive Network with exogenous inputs (NARX). The inputs and outputs are the simulated daily mean temperatures obtained from six Regional Climate Models (RCMs) collected from the North American Coordinated Regional Downscaling Experiment (NA-CORDEX) archive and observed daily mean temperatures collected from the Digital Archive of Canadian Climatological Data, respectively. A case study of Big Trout Lake in Ontario, Canada is carried out as a preliminary study to evaluate the performance of the proposed neural network models. The results show that the neural network based ensembles outperformed each of the individual regional climate models and generated predictions with smaller fluctuations. Secondly, the thesis investigates and compares the applicability and performance of machine learning and statistical methods in developing multi-model ensembles for downscaling long-term daily temperature. The machine learning methods include Long Short-Term Memory (LSTM) networks and Support Vector Machine (SVM) and the statistical methods include arithmetic ensemble mean (EM) and Multiple Linear Regression (MLR). These ensembles share the same input and output variables with the preliminary study. The performance of the proposed machine learning and statistical ensembles are evaluated at twelve meteorological stations across Ontario, Canada. The results show that multi-model ensembles with machine learning or statistical techniques all performed well at downscaling daily temperature, and had similar performance with relatively high accuracy. This is the first attempt to apply advanced machine learning techniques and compare them with statistical methods in developing multi-model ensembles for downscaling in Canadian communities. The results provide a technical basis for applying statistical and machine learning methods to generate long-term high-resolution daily temperature projections. The generated climate projections will also provide useful information to support climate adaptation in Ontario. |
URI: | http://hdl.handle.net/11375/25121 |
Appears in Collections: | Open Access Dissertations and Theses |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Li_Xinyi_201912_MASc.pdf | 3.05 MB | Adobe PDF | View/Open |
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.