Extreme events like earthquake and climate extremes have huge economic and environmental impact. Taking climate extremes as examples, which can be broadly defined as highly anomalous climatic conditions on sub-daily to annual or longer time scales, there are various extremes. Examples include heat waves, heavy rain, and severe winds; specific phenomena such as cyclones, tornadoes, or floods, and also extremes resulting from a combination of biophysical impacts such as aridity/droughts (affected by rainfall and evaporation), heat stress (temperature and humidity) and bush fires (rainfall, temperature and wind).
Regardless of the type of events, past extremes have imposed considerable industry costs stemming from stresses and damage to infrastructure and direct impacts on commodities, such as loss or damage to crops and livestock. Exponentially increasing economic losses, coupled with an increase in deaths due to these events, have served to focus attention on the possibility that these events are increasing in frequency and intensity. Australia has the unfortunate distinction of being a world leader in insured losses from extreme weather events. Although it makes up only 2 per cent of the global reinsurance market, Australia accounted for 6 per cent of global losses in the 5 years to 2013. These adverse effects of climate extremes highlight both the need for effective monitoring of these events as well as understanding the effects of historical extremes as well as future changes in the extremes.
Linking atmospheric and hydrological or crop models is challenging because of a mismatch of spatial and temporal resolutions in which the models operate: dynamic hydrological or crop models need input at relatively fine temporal (daily or hourly) scale, but the outputs from general circulation models (GCM) are usually not realistic at the same scale, especially for extremes, even though fine scale outputs are available. Weather generators, often including spatial downscaling and/or temporal disaggregation components, are designed to produce finer spatio-temporal-scale data from reliable larger spatio-temporal-scale data. They are often used to help us to handle possible future climate changes as well as extremes.
To generate representative future weather data, CSIRO has developed weather generators. Some of them have specific consideration of extremes. Examples include using hyperbolic distribution, quantile regression, Markov Chain plus Burr XII distribution.
The project can be carried out in several different directions, with different output in mind.
1. Refine and implement daily and subdaily weather generators that CSIRO has developed in computationally efficient language like C/C++, and package them into an R package with well-documentation; Possible challenges include how to adjust the techniques for hourly generator with parameter tuning; adaptive model setting; system testing, and comparison with other widely accessible systems.
2.. Refine and implement threshold selection for genearlised Pareto distribution (GPD). The techniques could be based on surprise, incremental structure, and some diagonal plots.
3. Compare various extreme modelling techniques with special consideration of return period calculation. Modelling techniques could be Burr XII, genearlised Pareto distribution + automatic threshold selection, Hyperbolic, skew t-distribution, and Bayesian mixture. Comparison angles can be accuracy, sensitivity and interpretability. Automatic threshold selection for GPD needs to be packaged into an R package.
4. Spatial or spatio-temoral statistical modelling of extreme events
- Familiarity with programming languages, like C/C++ or R
- Basics of related knowledge, like statistical computation, software engineering, statistical modelling like various distributions,
- Interest in solving real-world problems
- statistical modelling techniques (optional)
- R free software for Statistical Computing
- Scarrott, Carl, and Anna MacDonald. "A review of extreme value threshold estimation and uncertainty quantification." REVSTAT–Statistical Journal10.1 (2012): 33-60.
- Kokic, Philip, Huidong Jin, and Steven Crimp. "Improved point scale climate projections using a block bootstrap simulation and quantile matching method." Climate dynamics 41.3-4 (2013): 853-866.
- Shao, Quanxi. "Notes on maximum likelihood estimation for the three-parameter Burr XII distribution." Computational statistics & data analysis 45.3 (2004): 675-687.
- Shao, Quanxi, Louie Zhang, and Q. J. Wang. "A hybrid stochastic-weather-generation method for temporal disaggregation of precipitation with consideration of seasonality and within-month variations." Stochastic Environmental Research and Risk Assessment: 1-20. (published online in Nov 2015)
- Lee, J., Yanan Fan, and Scott A. Sisson. "Bayesian threshold selection for extremal models using measures of surprise." Computational Statistics & Data Analysis 85 (2015): 84-99.
- Software package development experience
- Stronger R and C++ programming skills, that will be valuable for future statistical data analysis or data mining
- State-of-art of extreme analysis and modelling techniques
- Real world problem solving;
extreme, threshold, weather generator