Examining Contiguous Extreme Events over the United States

The current study examines the areal extent and duration of extreme events across the United States, in a high-resolution gridded dataset of observed temperature and precipitation.

The space-time volume spanned by a gridded dataset [lon, lat, time] is transformed to an extreme-event volume space E, containing mostly non-event cells (E_x,y,t = 0), but also a population of sub-volumes (regions in space and time) that are marked with having an event value (E_x,y,t = 1), given they exceed some threshold value. The event designation can be achieved by applying an absolute threshold criterion, such as marking all gridboxes that have a value of precipitation greater than 1 inch/day (for example) to identify heavy rainfall events. Alternatively, the event criteria could involve percentile-based thresholds computed over a reference period; for example, marking all gridboxes having a maximum temperature value in excess of the 90^th percentile (for each lon-lat box) to identify heat wave events.

These exceedance extreme event “blobs” display contiguity in space and time, and can be subsequently analyzed using simple binary image processing techniques, providing a method for identifying and classifying extreme events (within a gridded dataset) with respect to certain bulk properties. This enables the ranking of past observed events, such as heavy prolonged rainfall events by the total volume, or excessive heat events by areal extent, without predetermining the spatial- or duration-based limits on the event boundary. This technique can be further employed to analyze past trends and future projections in climate model simulations.