HR: 1340h
AN: NG23D-0110 [Abstracts]
TI: Impact of Noise and Seasonality on the Detection and Nonlinear Prediction of Chaos From Finite
River-Flow Time Series
AU: * Ganguly, A R
EM: gangulyar@ornl.gov
AF: Oak Ridge National Laboratory, Computational Sciences and Engineering
MS 6085; 1 Bethel Valley Road, Oak Ridge, TN 37831
United States
AU: Khan, S
EM: skhan4@eng.usf.edu
AF: University of South Florida, Civil and Environmental Engineering
311 Kopp Engineering Building, Tampa, FL 33620
United States
AU: Saigal, S
EM: saigal@eng.usf.edu
AF: University of South Florida, Civil and Environmental Engineering
4202 East Fowler Avenue, Tampa, FL 33620
United States
AB:
Detection of possible chaos in hydrological time series can be useful for scientific understanding of the component processes
as well as for short-term predictability and predictive modeling. However, the presence of noise and seasonality makes the
detection of any nonlinear component, especially chaos, difficult in finite time series. This study utilizes approaches such
as correlation dimension (CD), phase space reconstruction (PSR) and artificial neural networks (ANN) for the detection of
possible chaos. The results on simulated data generated from the Lorenz system of equations (contaminated with various levels
of noise and periodicity) indicate the presence of thresholds in terms of "noise to chaotic-signal" and "seasonality to
chaotic-signal", beyond which the currently available set of tools are unable to detect the chaotic component. The simulation
results also demonstrate that the underlying chaotic or nonlinear component, if present, may be extractable from a time
series contaminated with noise and seasonality. We also show the impacts on predictive modeling, for example we illustrate
the possibility that a decomposition of the time series observations into periodic, nonlinear dynamical and noise components
can be utilized to improve predictive modeling through a best fit strategy that applies the most suitable methodology to each
component. Analysis of monthly streamflow data from the Arkansas River at Little Rock and daily streamflow data from the
Colorado River below Parker dam shows that the chaotic component can be detected in the Arkansas data but not in the Colorado
data. The extracted chaotic component from the Arkansas data is processed further to generate multi-step ahead predictions.
These results suggest that while chaos may be detectable in certain hydrological time series leading in many situations to
improved short-term predictability, not all hydrological time series exhibits detectable chaos.
Acknowledgment: Auroop R Ganguly gratefully acknowledges the Laboratory Directed Research and Development Program (SEED money
funds) of the Oak Ridge National Laboratory (ORNL), managed by UT-Battelle, LLC for the U.S. DOE under Contract No.
DE-AC05-00OR22725.
DE: 1800 HYDROLOGY
DE: 4400 NONLINEAR GEOPHYSICS (3200, 6944, 7839)
SC: Nonlinear Geophysics [NG]
MN: Fall Meeting 2005