P-Surge Model - MDL
The Probabilistic storm Surge (P-Surge) model (Taylor and Glahn) is a probabilistic approach to storm surge modeling that accounts for the uncertainty in the wind forecast. P-Surge uses the National Hurricane Center's (NHC) official forecast along with historic 5-year average errors in its track, size and intensity to derive an ensemble of inputs for the parametric wind model used within SLOSH. Each input represents a specific portions of the error space defined by the 5-year average error, allowing each storm surge output from SLOSH to have a well defined statistical weight. P-Surge combines the resulting storm surge output and weights into probabilistic products. P-Surge is consistent with NHC's message and utilizes the best available parametric information as NHC's official forecast has historically outperformed other guidance.
MDL developed P-Surge version 1.0 between 2003 and 2008 in response to Hurricane Isabel (2003). P-Surge version 1.0 computed probabilistic inundation information based on storm surge alone. MDL developed P-Surge version 2.0 in 2014 by introducing gridded tide predictions and updating the computational grids to provide more accurate overland inundation guidance. Additional products were created with more time steps as well as Above Ground Level (AGL) products to provide users with more useful guidance. P-Surge version 2.0 provided probabilistic inundation information based on storm surge and tide. MDL's latest development was P-Surge version 2.7 in 2017, which extended the forecast hours from 78 to 102 hours, upgraded the south Florida basin, and providing better support for NHC’s operational needs
The National Weather Service's (NWS) Meteorological Development Lab (MDL), in furtherance of its mission to help protect life and property, developed the Probabilistic storm Surge (P-Surge) model between 2003 and 2008 to provide storm surge inundation guidance for tropical cyclones which takes into account the uncertainty in the forecast.
P-Surge utilizes a parametric wind model which uses the National Hurricane Center’s (NHC) official forecast along with historic errors in its track, size, and intensity. This allows P-Surge to be consistent with NHC’s message and utilize the best available parametric information as NHC’s official forecast has historically outperformed other guidance.
Methodology and Products:
The P-Surge model is an ensemble of SLOSH model runs, that provides real-time probabilistic predictions based on the current National Hurricane Center (NHC) hurricane forecast. Each ensemble member’s input is derived from the current NHC hurricane forecast along with the associated 5 year average along track, cross track and intensity errors. Additionally P-Surge utilizes error distributions based on the derived size of the storm. P-Surge then computes the possible storm surge from each of these and combines them together to get a better idea of the probable storm surge within the next 102 hours.
P-surge Error Statistics:
NHC determines the 5 year error statistics by considering the following:
- Official forecasts initiated and verified within 10-45N and 60-100W.
- Official forecasts which started with at least hurricane strength.
Intensity Error Sampling:
- Three samples for Strong (30%), Medium (40%), Weak (30%) Intensity.
Along Track Error Sampling:
- Seven equally weighted samples (14%) for different forward speeds.
Cross Track Error Sampling:
- Cross track samples cover 90% of the area (± 1.64 sigma) under the normal distribution assigning the 5 year mean absolute error to 0.7979 sigma.
- To be dense enough, the storms were chosen to be one radius of maximum winds apart from each other at the 48 hour forecast projection.
Size Error Distribution and Sampling:
- The size error is bounded, invalidating the normal distribution assumption. Instead, five error distributions were derived by grouping storms based on initial size and determining the historic size guidance skill in the Atlantic. Once an appropriate error distribution is chosen, it's sampled three times similarly to the Intensity errors.
The result is the number of ensemble members is 3*7*X*3 where X depends on the density of the cross track sampling. The weight of each member is the product of the weight of each error sample: e.g. 0.3 * 0.14 * 0.1 * 0.4 = 0.00168