Accession Number : ADA300497
Title : Modeling Long Term Dependence, Nonlinearity and Periodic Phenomena in Sea Surface Temperatures Using MARS.
Descriptive Note : Technical rept.,
Corporate Author : NAVAL POSTGRADUATE SCHOOL MONTEREY CA DEPT OF OPERATIONS RESEARCH
Personal Author(s) : Lewis, Peter A. ; Ray, Bonnie K.
PDF Url : ADA300497
Report Date : 29 SEP 1995
Pagination or Media Count : 38
Abstract : We study a time series of 20 years of daily sea surface temperatures (SSTs) measured off the California coast. The SSTs exhibit quite complicated features, such as effects on many different time scales, nonlinear effects, and long range dependence. We use a modified MARS algorithm to obtain univariate adaptive spline threshold autoregressive (ASTAR) models for the SSTs and discuss practical modeling issues, such as handling of cycles and long range dependence. We approximate a nonlinear long memory model by allowing very long autoregressive terms in the ASTAR model. This large order ASTAR model is better predictively and descriptively than any other univariate model explored. Use of other concurrent predictor time series, in particular, categorical predictor variables such as wind direction, to extend the threshold autoregressive model for the SSTs to semi-multivariate adaptive spline threshold autoregressive (SMASTAR) models for the SSTs is also discussed. It is shown that SMASTAR modeling, with an added categorical time of year predictor, can also be used to model nonlinear structure in the data which is changing with time of year. Models for the SSTs are evaluated using out of sample forecast RMSEs, residual diagnostics, model skeletons, and sample functions of simulated series. Computational issues, such as choice of parameters in the MARS algorithm, in particular the span parameter, are discussed in an appendix.
Descriptors : *OCEAN SURFACE, *TIME SERIES ANALYSIS, *SURFACE TEMPERATURE, MATHEMATICAL MODELS, SIMULATION, COASTAL REGIONS, PREDICTIONS, THRESHOLD EFFECTS, LONG RANGE(TIME), VARIABLES, RESIDUALS, SCALE, WIND DIRECTION, NONLINEAR ANALYSIS, SEA WATER, DIURNAL VARIATIONS, SPLINES.
Subject Categories : Physical and Dynamic Oceanography
Distribution Statement : APPROVED FOR PUBLIC RELEASE