Neighbourhood selection for local modelling and prediction of hydrological time series

The prediction of a time series using the dynamical systems approach requires the knowledge of three parameters; the time delay, the embedding dimension and the number of nearest neighbours. In this paper, a new criterion, based on the generalized degrees of freedom, for the selection of the number of nearest neighbours needed for a better local model for time series prediction is presented. The validity of the proposed method is examined using time series, which are known to be chaotic under certain initial conditions (Lorenz map, Henon map and Logistic map), and real hydro meteorological time series (discharge data from Chao Phraya river in Thailand, Mekong river in Thailand and Laos, and sea surface temperature anomaly data). The predicted results are compared with observations, and with similar predictions obtained by using arbitrarily fixed numbers of neighbours. The results indicate superior predictive capability as measured by the mean square errors and coefficients of variation by the proposed approach when compared with the traditional approach of using a fixed number of neighbours.

[1]  Jianming Ye.  On Measuring and Correcting the Effects of Data Mining and Model Selection , 1998 .

[2]  Shie-Yui Liong,et al.  Singapore Rainfall Behavior: Chaotic? , 1999 .

[3]  P. B. Wright,et al.  Homogenized long‐period Southern Oscillation indices , 1989 .

[4]  J. Elsner,et al.  The weather attractor over very short timescales , 1988, Nature.

[5]  Leonard A. Smith.  Identification and prediction of low dimensional dynamics , 1992 .

[6]  Gwilym M. Jenkins,et al.  Time series analysis, forecasting and control , 1971 .

[7]  Feizhou Lai,et al.  Time Series Analysis of Water Quality Data in Pearl River, China , 1989 .

[8]  H. Akaike.  A new look at the statistical model identification , 1974 .

[9]  Bellie Sivakumar,et al.  Chaos theory in hydrology: important issues and interpretations , 2000 .

[10]  G. Schwarz.  Estimating the Dimension of a Model , 1978 .

[11]  P. Grassberger,et al.  Measuring the Strangeness of Strange Attractors , 1983 .

[12]  George Sugihara,et al.  Nonlinear forecasting as a way of distinguishing chaos from measurement error in time series , 1990, Nature.

[13]  Farmer,et al.  Predicting chaotic time series. , 1987, Physical review letters.

[14]  Yingcun Xia,et al.  ON THE ESTIMATION AND TESTING OF FUNCTIONAL-COEFFICIENT LINEAR MODELS , 1999 .

[15]  Demetris Koutsoyiannis,et al.  Deterministic chaos versus stochasticity in analysis and modeling of point rainfall series , 1996 .

[16]  H. Abarbanel,et al.  Prediction in chaotic nonlinear systems: Methods for time series with broadband Fourier spectra. , 1990, Physical review. A, Atomic, molecular, and optical physics.

[17]  H. Kantz,et al.  Nonlinear time series analysis , 1997 .

[18]  Renzo Rosso,et al.  Comment on “Chaos in rainfall” by I. Rodriguez‐Iturbe et al. , 1990 .

[19]  B. Hao,et al.  Directions in chaos , 1987 .

[20]  Wai Keung Li,et al.  A note on the corrected Akaike information criterion for threshold autoregressive models , 1998 .

[21]  Henry D. I. Abarbanel,et al.  Analysis of Observed Chaotic Data , 1995 .

[22]  Takens,et al.  Estimation of the dimension of a noisy attractor. , 1994, Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics.

[23]  P. B. Wright,et al.  Relationships between Indices of the Southern Oscillation , 1984 .

[24]  A. Jayawardena,et al.  Noise reduction and prediction of hydrometeorological time series: dynamical systems approach vs. stochastic approach , 2000 .

[25]  George J. Mpitsos,et al.  Variability and Chaos: Neurointegrative Principles in Self-Organization of Motor Patterns , 1989 .

[26]  A. Jayawardena,et al.  Analysis and prediction of chaos in rainfall and stream flow time series , 1994 .