Sea level forecasts

Validation of Hirlam 7.2

KNMI intends to replace its operational numerical weather prediction model Hirlam by a new version (7.2) at the end of 2009.
For the users of the model data and data from models that depend on Hirlam, a validation of the new model results is relevant. This web site describes the validation of numerical sea level forecasts, produced by the storm surge model WAQUA/DCSM98 with input from Hirlam.

Validation experiments

Two kinds of validation experiments have been performed: a parallel run of the model with the new input and a rerun of 3 storms from the past.

Parallel runs

The WAQUA/DCSM98 model has been run in parallel to the operational version from the end of October 2008. However, due to late availability of the Hirlam 7.2 input data not all runs are complete. Therefore the model has been rerun for this test from 2008102306 – 2009082800, both with and without assimilation of observed water levels.
The present runs with data assimilation have not used all of the available water level observations. Therefore the comparison with the operational model might turn out unjustly (un?)favourable for the new Hirlam version.

Storms

In close cooperation with the Storm Surge Warning Service, the following storm periods were selected for a comparison:

Event	Simulation
	Start	End
16/17 December 2005	2005121300	2005121812
9 November 2007	2007110400	2007111000
1 March 2008	2008022500	2008030300

Results

Parallel runs

Scatter plots of the model skew surge versus the observed skew surge show very little difference between runs with Hirlam 7.2 input and operational runs with Hirlam 7.0 input.
Also, statistical parameters as a function of either t_f or as a running 3-month quantity, show only small differences. Most notable is the difference in the Kalman filter results, but that should be attributed to the difference of the assimilated data.

Storms

Storms have been run with 1 month spinup of the astronomical tide and 2 – 3 days spinup with meteorological fields from the operational archive.
Comparisons have been made with forecasts from the operational archive. The Hirlam versions used for these forecasts was 7.0 for the 2007 and 2008 storms and 6.3 for the 2005 storm.
Differences between both models are generally small. In a few cases one of the models overestimates the surge more than the other, but this is not always the same model.