|InterJournal Complex Systems, 383
|Manuscript Number: |
Submission Date: 505
|Hidden States for Modeling Interactions between disparate Spatiotemporal Scales|
Subject(s): CX.01, CX.08, CX.06
Category: Brief Article
Understanding and modeling complex systems is difficult because different time and length scales are coupled. For example global seasonal climate conditions drive local weather dynamics. We describe a variety of hidden Markov model (HMM) which can model such coupling, specifically, a seasonal model of precipitation. Our model assumes the existence of unobserved (or hidden) "weather states" which follow a Markov chain. The states of the Markov chain have output models associated with them which model the local weather behavior, in our case, precipitation. Seasonally dependent transition probabilities between the states provide for the coupling of the local weather behavior to the global seasonal climate behavior. Two-parameter gamma distribution output models are fit to the precipitation. We use a variant of the Baum Welch forward backward algorithm, developed for research in natural language processing, to do maximum a posteriori estimation of all model parameters.
|Submit referee report/comment|