Evaluation study - operational RWsOS Meren : analysis of D-HYDRO model performance during high waters of winter 2023-2024

This study evaluates the performance of the D-HYDRO (Delft3D FM) hydrodynamic models during extreme storm events and the prolonged high water period in winter 2023-2024 in the IJsselmeer region (Netherlands). The aim is to create a baseline for model accuracy during high-water periods and increase understanding of model performance, strengths, and weaknesses. The models analyzed are the updated D-HYDRO models from the operational RWsOS-Meren system, based on Delft-FEWS software. The study focuses on fine- and coarse-gridded D-HYDRO models with KNMI-HARMONIE meteorological forcing, including ijsselmeer4_harmonie, ijsselmeer4c_harmonie, markermeer4_harmonie, markermeer4c_harmonie, and veluwerandmeren4_harmonie. For some models, archived operational model output from the MATROOS archive is used, while for others, model results were re-created using archived meteorological forcing data. The reforecasting results closely aligned with operationally archived data, with a negligible RMSE difference (<0.8 cm), confirming the validity of the reforecasting approach for evaluating model performance. Fine-gridded D-HYDRO models showed good accuracy during the studied period, with RMSE ranging from 2–4 cm at lake stations and around 5 cm at most riverine stations. Weighted average lake levels were predicted with higher accuracy (1–2 cm RMSE) due to the lake level correction applied during modeling. During Storm Pia (December 20th – 22nd, 2023) and Storm Henk (January 2nd – 5th, 2024), model performance deteriorated slightly, with RMSE increasing by 1-3 cm, and water level peaks were further underestimated, particularly during Storm Pia. D-HYDRO models performed similarly to the older WAQUA models, with no significant improvement or decline in overall model performance. Differences in model performance can be considered negligible overall, but performance does differ at specific stations (e.g., Kampen). Coarse-gridded models showed reduced accuracy compared to their fine-gridded counterparts, particularly in riverine areas, where RMSE increases significantly (up to 10–30 cm). However, predictions of weighted average lake level and water levels at lake stations remain relatively unaffected, making coarse-gridded models useful for broader application in ensemble forecasting. Therefore, D-HYDRO models are concluded to be sufficiently accurate for operational use under normal conditions and during extreme events, despite a slight reduction in model performance during those periods.