The best thing about being a statistician is that you get to
play in everyone’s backyard
Hurricane Harvey was one of the most destructive and costliest natural disasters in modern American history. Not only did the extreme rainfall and flooding of Harvey cause nearly $125 billion in damage, but the extreme weight of the floodwater (nearly 275 trillion pounds of water) actually pushed the Earth’s crust down by nearly 2 cm!
Disasters like Hurricane Harvey underscore just how important it is to effectively monitor and measure extreme precipitation events. Scientists often quantify the rarity of an extreme precipitation event using the Average Return Interval, or ARI for short, which describes the “typical” times between event occurrences. For example, an extreme precipitation event with an ARI of 10 years corresponds to an event which occurs on average once every 10 years. However, because “extreme” storms are relatively uncommon and the characteristics of each storm vary over time and location, it is quite difficult to actually assess just how “extreme” an event really is.
Fortunately, the branch of statistics known as extreme value theory (EVT) is perfectly suited to address these problems! Under the framework of EVT, precipitation intensities are assumed to be random draws from some underlying probability distribution, and characterizing extreme value behavior is equivalent to characterizing the upper tail of this distribution. Satellite-based retrieval algorithms based on the measurements made by the Tropical Rainfall Measuring Mission (TRMM) and the more recent Global Precipitation Measurement (GPM) satellites have provided a rich source of precipitation data at the global scale for quantifying extreme precipitation events.
In a joint collaboration with Drs. Yaping Zhou and George J Huffman from NASA Goddard Space Flight Center, I helped develop a new methodology for computing ARIs corresponding to extreme precipitation events based on data collected from the TRMM satellites. Our methodology involves grouping similar regions into clusters, then independently fitting a statistical model to the extreme precipitation data in each of the clusters. Importantly, the ARI maps produced by our model could be used by policymakers for disaster monitoring and prevention.
Publications and Presentations:
You can access my posters and slides below: