Abstract:
With the frequent occurrence of climate-induced natural disasters in recent years, research on climate change is urgently needed. To get a better understanding of how climate change is developing, statisticians over the past several decades have proposed various statistical methods to quantify the climate change problem. For example, models for estimating climate trends were introduced and other trend-feature detection approaches were developed. Among which, Change-Point analysis is widely used in detecting major turning points in climate time series. In addition, the rate of climate change over time is also an important trend feature to analyze.
In this project, we present and compare two statistical estimation methods to model climate trends. The narrowest-over-threshold Change-Point detection method captures the significant phase transitions in the time series trends. The piecewise linear regression model can then be constructed between the detected Change-Points. We advance the Narrowest-over-threshold Change-Point detection method by providing uncertainty measures of the Change-Point locations via time series moving block bootstrapping. In addition, local polynomial regression modelling is also discussed and implemented in this project. Based on these regression models, we further analyze the rate of change for trends using the modified Mann-Kendall test to test for significant changes in trend over time. The methodologies introduced are eventually applied on four different climate time series.
This report is available upon request. If you are interested, feel free to contact me!