Abstract by Lynsie Warr
Approximation of Gaussian Processes in Precipitation Modeling for High Mountain Asia
The High Mountain Asia region stores more water in glaciers than any other region (except the poles) and provides water to one-fifth of the world's population. Recently, temperatures in High Mountain Asia have increased more quickly than the global average, and precipitation patterns are changing. As a result, the risk of flooding, crop damage, water shortages, and mudslides is increasing downriver of the glaciers. Since precipitation patterns have significant effects on local populations, accurate precipitation models are essential. To better understand changing precipitation in High Mountain Asia, we use large climate data sets which vary in geographical coverage and in spatial and temporal resolutions. Gaussian processes are commonly used to model spatiotemporally correlated data, but fitting a full Gaussian process to large data sets is computationally expensive and unrealistic for our problem. This study explores approximation of Gaussian processes (e.g. local approximate Gaussian processes) in modeling a precipitation surface in High Mountain Asia and examines the merits of such models.