A Computational Framework for Modelling and Analyzing Ice Storms

Ranjini Swaminathan, Mohan Sridharan, Katharine Hayhoe

arXiv preprint · 13 May 2018

In one sentence. Framework learns ice storm occurrence from atmospheric variables with weak labels, evaluated on reanalysis and GCM data under future climate scenarios.

Abstract

Ice storms are extreme weather events that can have devastating implications for the sustainability of natural ecosystems as well as man made infrastructure. Ice storms are caused by a complex mix of atmospheric conditions and are among the least understood of severe weather events. Our ability to model ice storms and characterize storm features will go a long way towards both enabling support systems that offset storm impacts and increasing our understanding of ice storms. In this paper, we present a holistic computational framework to answer key questions of interest about ice storms. We model ice storms as a function of relevant surface and atmospheric variables. We learn these models by adapting and applying supervised and unsupervised machine learning algorithms on data with missing or incorrect labels. We also include a knowledge representation module that reasons with domain knowledge to revise the output of the learned models. Our models are trained using reanalysis data and historical records of storm events. We evaluate these models on reanalyis data as well as Global Climate Model (GCM) data for historical and future climate change scenarios. Furthermore, we discuss the use of appropriate bias correction approaches to run such modeling frameworks with GCM data.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.