Along the south coast of Australia sea-level variations over longer time-scales are related to meteorological systems which move from west to east. The corresponding sea-level variations also migrate from west to east as demonstrated by Provis and Radok (1979) and illustrated in the monthly reports from the National Tidal Centre for the Australian Baseline Sea-level Monitoring Project. It would be useful in terms of emergency services if the water levels associated with such storm events could be predicted in advance.

Previous unpublished work by Cardno determined that the sea-levels from Portland could be used to predict the water level at Geelong 18 hours in advance. In order for such predictions to be determined for the study area, reliable tide-gauge data must be obtained. The closest long-term reliable tide gauges to the study area are located at Lorne and Queenscliff.

In order to determine if sea-level rise under storm condition could be predicted in advance, the measured sea-levels at Portland and Lorne were compared. Non-tidal variations in sea-level are defined by determining the difference between the measured and predicted sea-level, that is, by the residual. As storms travel from west to east, the residual at Portland should therefore be an indication of potential surge to be added to the predicted tide at Lorne.

The initial time lag between Portland and Lorne was determined using a lagged cross-correlation based upon the residual at Portland and the non-tidal water level at Lorne. This indicated that it took seven hours for sea-level variations measured at Portland to reach Lorne.

In order to provide the best prediction algorithm, a technique known as “Artificial Neural Networks” was applied. This is a method of using multi-variable correlations to find relationships between one set of variables and another. The main feature of the method is the use of a set of “hidden” variables in the correlation which allows for a better fit.

One of the disadvantages of predicting sea-level with a neural network program is that the prediction algorithm is relatively complicated. In order to simplify the process for operational purposes, the possibility of using a simple addition of the residual from Portland seven hours earlier to the predicted tide at Lorne was investigated. The results of this prediction in comparison to the predicted sea-level and actual measured sea-level at Lorne are shown in Figure 5-4 for a storm event.

Figure 5-4     Comparison of ANTT predicted, actual measured and predicted sea-level at Lorne using the Portland residual seven hours earlier added to the Lorne predicted tide.

The data in Figure 5-3 shows that the simple method provides a useful prediction of the timing and height of the storm-tide. Comparison with the more complex neural network predictions showed that the simpler system performed as well for practical purposes and hence this method is recommended. A cross-correlation of the measured sea-levels and those from the simplified prediction method are shown in Figure 5-4.

Figure 5-5     Cross-correlation of measured sea-level at Lorne with predicted sea-level using tide + residual

The statistics of the differences between the measurements and the Portland-residual prediction method using all the available data in 1994, 2001 and 2004 are shown in Table 5-4. The statistics of the difference indicate that by taking the Portland residual seven hours prior and adding it to the predicted tide level at Lorne, an accuracy of typically much less than 0.1 m can be achieved. In fact, 98% of the predicted values for each year examined differ by less than 0.1 m from the observed value. This indicates that the sea-level at Lorne under storm-surge conditions can be predicted to an accuracy of 0.1 m by adding the residual from Portland seven hours prior to the predicted Lorne tide level, e.g. sea-level at Lorne (10:00) = Predicted Lorne tide level (10:00) + Portland residual (03:00).

Table 5-4     Statistics of the difference between the measured sea-level at Lorne and the predicted sea-level using the Portland residual for 1994, 2001 and 2004.

The sea-level at Lorne is applicable to storm-tide levels at Barwon Heads and Ocean Grove. For flooding at Queenscliff and Point Lonsdale, the residual from Portland should be added to the predicted tide at Queenscliff. This prediction system should provide emergency managers with useful information in the planning of evacuations and emergency response, if inundations events are likely to cause significant flood depths for a significant amount of time.