| FLORIDA STATE UNIVERSITY | ||
 |
||
 FSU Home |
Beaches and Shores Resource Center |
BSRC Home
|
SHORT TERM STORM SURGE FORECASTING This note discusses very short term (order of 1-3 hours) forecasting of storm surge water level of tide gage data for a gage site on the East Coast of the U.S. Water level observations and tide predictions were utilized to construct a surge signal (the difference between the observed water level and the predicted tide). An autoregressive (AR) modeling approach was adopted to predict short term (1-3 hour) surge water levels. The AR model defined here consists of a lag space model where surge signal with mean removed is a function of previous values of the surge and of a noise input. Various issues to resolve in utilizing a forecasting approach include determining the length of the series utilized in the fit (stationarity issue) and determining the model order (number of parameters to utilized in the model). A forecast approach was utilized on a number of large storm surge events where the best judge of forecast ability was considered to be the direct comparison of forecast values versus observed values. An example of the "out of model" forecast fit to the data for 1 forecast step (1 hour) utilizing a block size of 48 (hours) and a model order of lag 3 (hours) is provided in Figure 1 for the December 1992 Northeaster storm. Figure 2 provides a 3 forecast step (3 hour) comparison of observations and forecasted surge levels for the same storm. It should be noted that during the rising event, as the forecast step increases, so does the phasing error between the actual observation and the forecast. For surge prediction between gaging sites, regression may be utilized. A two site regression plot for use in potential missing data fill in of the historical records is given in Figure 3. As ordinary least squares regression identifies observation error being in only one gage station, an additional least squares regression line with error balanced in both stations is also shown in Fig.3. Regression is also of use in multi gage prediction schemes should various gages within a gaging network fail. Should you have any questions concerning forecasting or regression for your specific project needs call: Todd Walton, Ph.D., P.E. (850-644-2847). REFERENCES: Walton, T. 1999. "Discussion: Predicting Caspian Sea Surface Water Level By ANN and ARIMA Models," Journal of Waterway, Port, Coastal, and Ocean Engineering, ASCE, Vol.125, No.1. Walton, T. and Garcia, A. 2001. "Discussion: Back-Propagation Neural Network in Tidal-Level Forecasting," Journal of Waterway, Port, Coastal, and Ocean Engineering, Vol.127, No.1, ASCE, 57-58. Walton, T.L., Jr. (1989). Simulating Great Lakes Water Levels for Erosion Prediction, Journal of Coastal Research, Vol. 5, No. 3, pp. 377-389. Walton, T.L., Jr. and Borgman, L.E., (1990). Simulation of Non-Stationary, Non-Gaussian Water Levels on the Great Lakes, Journal of Waterways, Ports, Coastal and Ocean Division, ASCE, Vol. 116, No. 6. © Copyright 2001 Todd L. Walton Jr.
|
||||||||
