Reservoir Computing approach to Great Lakes water level forecasting

Summary The use of echo state network (ESN) for dynamical system modeling is known as Reservoir Computing and has been shown to be effective for a number of applications, including signal processing, learning grammatical structure, time series prediction and motor/system control. However, the performance of Reservoir Computing approach on hydrological time series remains largely unexplored. This study investigates the potential of ESN or Reservoir Computing for long-term prediction of lake water levels. Great Lakes water levels from 1918 to 2005 are used to develop and evaluate the ESN models. The forecast performance of the ESN-based models is compared with the results obtained from two benchmark models, the conventional recurrent neural network (RNN) and the Bayesian neural network (BNN). The test results indicate a strong ability of ESN models to provide improved lake level forecasts up to 10-month ahead – suggesting that the inherent structure and innovative learning approach of the ESN is suitable for hydrological time series modeling. Another particular advantage of ESN learning approach is that it simplifies the network training complexity and avoids the limitations inherent to the gradient descent optimization method. Overall, it is shown that the ESN can be a good alternative method for improved lake level forecasting, performing better than both the RNN and the BNN on the four selected Great Lakes time series, namely, the Lakes Erie, Huron-Michigan, Ontario, and Superior.

[1]  Upmanu Lall,et al.  Nonlinear Dynamics of the Great Salt Lake: Nonparametric Short-Term Forecasting , 1996 .

[2]  Victor Privalsky,et al.  Statistical Analysis and Predictability of Lake Erie Water Level Variations , 1992 .

[3]  Paulin Coulibaly,et al.  Nonstationary hydrological time series forecasting using nonlinear dynamic methods , 2005 .

[4]  José Carlos Príncipe,et al.  Analysis and Design of Echo State Networks , 2007, Neural Computation.

[5]  Jeffrey L. Elman,et al.  Finding Structure in Time , 1990, Cogn. Sci..

[6]  Harald Haas,et al.  Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication , 2004, Science.

[7]  MohammadSajjad Khan,et al.  Application of Support Vector Machine in Lake Water Level Prediction , 2006 .

[8]  J. Garbrecht,et al.  Using climate predictions in Great Lakes hydrologic forecasts. , 2006 .

[9]  Simon Haykin,et al.  Decoupled echo state networks with lateral inhibition , 2007, Neural Networks.

[10]  Thomas E. Croley,et al.  Using NOAA's New Climate Outlooks in Operational Hydrology , 1996 .

[11]  Frank H. Quinn,et al.  Evaluation of Potential Impacts on Great Lakes Water Resources Based on Climate Scenarios of Two GCMs , 2002 .

[12]  B. Bobée,et al.  Artificial neural network modeling of water table depth fluctuations , 2001 .

[13]  Ian T. Nabney,et al.  Netlab: Algorithms for Pattern Recognition , 2002 .

[14]  José Carlos Príncipe,et al.  Special issue on echo state networks and liquid state machines , 2007, Neural Networks.

[15]  Barak A. Pearlmutter.  Gradient calculations for dynamic recurrent neural networks: a survey , 1995, IEEE Trans. Neural Networks.

[16]  P. Coulibaly,et al.  Downscaling Precipitation and Temperature with Temporal Neural Networks , 2005 .

[17]  Stanley A. Changnon,et al.  Temporal Behavior of Levels of the Great Lakes and Climate Variability , 2004 .

[18]  B. Bobée,et al.  Improving extreme hydrologic events forecasting using a new criterion for artificial neural network selection , 2001 .

[19]  J J Hopfield,et al.  Neurons with graded response have collective computational properties like those of two-state neurons. , 1984, Proceedings of the National Academy of Sciences of the United States of America.

[20]  Paulin Coulibaly,et al.  Comparison of neural network methods for infilling missing daily weather records , 2007 .

[21]  D. Prokhorov,et al.  Echo state networks: appeal and challenges , 2005, Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005..

[22]  Yves Chauvin,et al.  Backpropagation: theory, architectures, and applications , 1995 .

[23]  Martin F. Lambert,et al.  Bayesian training of artificial neural networks used for water resources modeling , 2005 .

[24]  G. V. Puskorius,et al.  A signal processing framework based on dynamic neural networks with application to problems in adaptation, filtering, and classification , 1998, Proc. IEEE.

[25]  J. Bruce,et al.  Great Lakes Levels and Flows: Past and Future , 1984 .

[26]  Paulin Coulibaly,et al.  Bayesian neural network for rainfall‐runoff modeling , 2006 .

[27]  Herbert Jaeger,et al.  Echo state network , 2007, Scholarpedia.

[28]  Christopher M. Bishop,et al.  Neural Network for Pattern Recognition , 1995 .

[29]  Garrison W. Cottrell,et al.  2007 Special Issue: Learning grammatical structure with Echo State Networks , 2007 .

[30]  Paulin Coulibaly,et al.  Seasonal reservoir inflow forecasting with low-frequency climatic indices: a comparison of data-driven methods , 2007 .

[31]  Robert J. Abrahart,et al.  Neural network modelling of non-linear hydrological relationships , 2007 .

[32]  Jing Peng,et al.  An Efficient Gradient-Based Algorithm for On-Line Training of Recurrent Network Trajectories , 1990, Neural Computation.