A wavelet‐based method for the assessment of hydrodynamic model behavior against observed sea level variability

Abstract

Sea level outputs from hydrodynamic models are often assessed against tide gauge data by comparing the standard deviations of modeled and observed time series, and by calculating the Pearson correlation coefficient and centered root mean square error (cRMSE) between the two. Together these metrics provide a compact evaluation of model skill; however, they are based on the implicit assumption that this skill is independent of both the timing and the duration of sea level deviations from the mean. In this study, we describe a complementary approach, utilizing the spectral information contained in numerical model outputs as well as in tide gauge records. We use a Ricker wavelet transform to express observed and modeled sea level in time‐frequency space, demonstrating our method for the Baltic Sea, an inland sea characterized by absence of strong tides. More precisely, we run a well‐established setup of the Nucleus for European Modelling of the Ocean (NEMO), and compare outputs against hourly data from tide gauges along the coast of Finland. We use the wavelet power spectrum and related quantities to assess how well our simulations capture sea level perturbations in the mesoscale and synoptic timescale ranges, and to illustrate seasonality in model errors at these different scales. In particular, our results show that spectral analysis can reveal qualitative differences in model performance between tide gauge stations; differences which the standard non‐spectral metrics fail to capture.