veenstrajelmer
commented
4 months ago The analysis was done again with data downloaded on 8-4-2024, so after the corrections in DDL data on 22-02-2024. Many stations were not changed. For DELFZL there were significant improvements.
With the original data (combination of waterbase and pre-feb-2024 DDL data), there was a continuous upwards trend visible in the bias, with periodic corrections. Note that the period was already updated with respect to the slides/figure in the original issue description. Also, the figure shows a combination of waterbase (to 2018) and DDL data (from 2018):
When looking at updated DDL data (retrieved on 8-4-2024), it is clear that there was a linear correction and the bias is now more constant. However, in 2003 something seems off:
This is even more clearly visible when looking at the surge signal only:
The new figures contain only DDL data. This timeseries is incomplete in the start of the period of interest. This missing data was also documented in https://github.com/Rijkswaterstaat/wm-ws-dl/issues/39.
TODO: add other stations and updated slides
TvLoon-RWS
FZ shared the attached presentation called "Overleg_Verloop_Nederlandse_Waterstandmetingen_20220421.pdf". These slides show the monthly means of measured minus modelled waterlevels (bias corrected). It starts on slide 2 with 8 stations along the coast that show a strong spatial relation. The black line is the average of the stations:
The following slides show several stations that sometimes show unexpected deviations from this average black line, for instance a temporary vertical offset in 2004 and 2005 for VLISSGN:
Or difference that indicate a temporal offset trend like from 2009 to 2022 for BROUWHVSGT08 and DELFZL. DELFZL also shows a instantaneous correction in 2003, after which the trend continues:
It concludes with a spatial overview of the bias in 2013-2025, which shows many white (zero) dots along the coast but also several outliers:
Stations in the Waddensea and the RMM estuary could not be well represented in the model, so these can be ignored for now. Also, HOEKVHLD is affected by the Rhine ROFI and this is not well represented in the model. However, we expect that most of the other coastal stations to be sort of coherent. At least for Huibertgat, Brouwershavensegat and Cadzand we would expect bias values near zero.
The original presentation: Overleg_Verloop_Nederlandse_Waterstandmetingen_20220421.pdf