A Machine-Learning Approach to Studying Relationships Between Extremes in Geopotential Height and Surface Temperature

Using 500 hPa geopotential height (Z500) and near-surface air temperature (Tsfc) data from several reanalysis products, coupled with observed daily Tsfc from the Global Historical Climatology Network database, we're conducting a spatiotemporal analysis of relationships in daily extremes using connected component analysis and Gaussian Processes (GP) approaches. We explore trends and relationships in the occurrence of low/cold percentile anomalies (1st, 5th, and 10th percentiles) and high/hot percentile anomalies (90th, 95th, and 99th percentiles) in the Z500/Tsfc fields over 1979-2013, focusing on anomaly measures including frequency, duration, magnitude, and spatial extent. Our goal is to characterize relationships between upper-air drivers (500 hPa flow), which recent research has indicated may be increasing in meridional extent ("waviness"), and the trend toward fewer/weaker cold extremes and more/stronger hot extremes in Tsfc observed over much of the global land area in the last few decades.