Assessing Seasonal Coastal Sea Level Prediction Skill of Statistical and Dynamical Forecast Techniques

As the frequency of coastal flooding is increasing under climate change, an accurate and reliable coastal flooding forecast on seasonal time scale is needed. To provide guidance on future development of forecasting methods, we assess the forecast skill of nine models for sea surface height at two pilot stations: San Diego and Charleston. The new generation dynamical forecast system has higher skill than last generation, yet further improvement is expected if a better sea surface height field is assimilated into its initial condition. Most of the models can issue relatively reliable and sharp forecast for San Diego but to a much less extent for Charleston. The Receiver Operating Characteristic curve reveals that many of these models cannot issue enough ‘positive’ forecasts for both upper and lower tercile even with low threshold. The statistical downscaling and ECCO hybrid tend to improve the deterministic skill but are less effective for probabilistic skill. The existence of trend complicates the skill assessment. The trend mostly impacts Charleston as the east coast saw a larger trend over the last few decades. An alternative way to use the model output of sea level is proposed by adding the observed sea surface height to the model predicted sea level tendency. Such correction tends improve the skill of the models with poorer initialization but is not effective to models with realistic initial condition. For a accurate forecast, adding missing error variance to the statistically downscaled hindcast may improve its reliability.