Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.504851
Title: Real-time extraction of the Madden-Julian oscillation using empirical mode decomposition and statistical forecasting with VARMA and neural network models
Author: Love, Barnaby Stuart
ISNI:       0000 0001 3834 0404
Awarding Body: University of East Anglia
Current Institution: University of East Anglia
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
Abstract:
A suite of real-time statistical forecast models of the Madden-Julian Oscillation (MJO), a large scale, quasi-periodic phenomenon and the dominant mode of variability in the tropics, is presented along with a novel approach to performing real-time intraseasonal data filtering. The study introduces the new technique of empirical mode decomposition (EMD) in a meteorological-climate forecasting context. It identifies empirical adjustments that can be made to the basic EMD method to produce a band-pass filter that is highly efficient at extracting a broad-band signal such as the Madden-Julian oscillation (MJO) with minimal end effects, such that it is suitable for use in realtime. This process is used to efficiently extract the MJO signal from gridded time series of outgoing longwave radiation (OLR) data. A range of statistical models were then tested for their suitability in forecasting the MJO signal in the OLR data, as isolated by the EMD. These were from the general classes of vector autoregressive moving average (VARMA) and neural network models. The models were developed using 17 years of OLR data from 1980 to 1996. Forecasts (hindcasts) were then made on the remaining independent data from 1998 to 2004. These were made in real-time, as only data up to the date the forecast was made were used. A VARMA (5,1) model was selected and its parameters determined by a maximum likelihood method. The median skill of forecasts was accurate (defined as an anomaly correlation above 0.6) at lead times up to 25 days. Four neural network models were created with a range of degrees of freedom in their forecasts. .These had their parameters determined using back propagation methods in a supervised learning manner. The neural network comparable to the VARMA (5,1) model in terms of forecast resolution also had accurate median forecast skill at lead times of 25 days, while the remaining neural network models had accurate median forecast skill at lead times ranging from 15-25 days but with greater spatial resolution. The statistical MJO forecasts developed here perform significantly better than any current· dynamical models and are at least as skillful as other statistical MJO forecasts. Operational model forecasts are available at http://envam1.env.uea.ac.uk/mjo_forecast.html.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.504851  DOI: Not available
Share: