Research paperSensitivity analysis of a data assimilation technique for hindcasting and forecasting hydrodynamics of a complex coastal water body
Introduction
Rapid development of electro communication technologies, instrumentation such as satellites and radars since the 1990s has resulted in great improvements in oceanic observing systems. These powerful tools can monitor sea surface properties such as surface currents, sea surface temperature (SST) and waves at a fine spatial resolution over a relatively large domain and at relatively high temporal frequencies. Although abundant measurements can provide near real-time information of the variables of interest, how to take the advantage of observations to produce accurate forecast is another challenge for researchers. Data assimilation (DA), which has been widely used in meteorology for some time, is beginning to be applied more often in oceanic hindcasting/forecasting modelling systems. Two principle types of DA algorithms have been applied in oceanic modelling systems: sequential and variational DA algorithm (Robinson and Lermusiaux, 2000). The former methodology linearly combines measurements into models; examples of this approach are Optimal Interpolation (OI), Ensemble Optimal Interpolation (EnOI), nudging and Ensemble Kalman Filter (Chen, 2013, Evensen, 2003, Gopalakrishnan and Blumberg, 2012, Oke et al., 2010, Ren et al., 2014). Variational approaches use a cost function which measures the distance of analysis states between measurements and model background states over assimilation windows using either three-dimensional variational or four-dimensional variational DA algorithms (Moore et al., 2011, Yu et al., 2012).
However, very few of these DA models have been applied in the complex area of shallow inshore waters which are heavily influenced by wind dynamics, such as Galway Bay located on the west coast of Ireland. Dabrowski (2005) studied the inner Galway Bay for the purpose of flushing properties assessment. He analyzed the transport of dye concentration from models with and without wind forcing, and found that Galway Bay circulation could be classified as being highly wind-dependent. However, accurate offshore winds are not easily measured. As the CODAR radar system has monitored surface currents in Galway Bay since 2011, this research develops a DA forecasting system of surface currents for Galway Bay via assimilating radar data.
Estimation of the system's background error covariance is a crucial aspect for most DA system (Oke et al., 2010). OI and EnOI algorithms are quite similar, the difference between them lies in defining model background error covariance. Ensemble of model background states is used to compute the model background error covariance in EnOI, details are presented in Evensen (2003) and Oke et al. (2010). The size of the ensemble is an empirical parameter to be determined in a practical EnOI assimilation system. Computation of the model background error covariance from an ensemble in an EnOI assimilation system needs additional computational cost compared with the OI assimilation system. Particular assimilation parameters such as correlation lengths need to be defined in OI assimilation systems (Kaplan et al., 2000, Rienecker, 1991, Rienecker and Adamec, 1995). Very little information has been published regarding ranges of these parameters or optimum values for use in particular applications. Additionally Ragnoli et al. (2012) suggest that the influences of previous DA on forecast was short-lived based on analyzing the surface velocity components time series at ten spatial points. In order to further investigate the potential to improve modelling performance for both hindcasts and forecasts, sensitivity analysis of horizontal correlation length and DA cycle length (CL) have been carried out to assess the impact of variations of these parameters on model results in this work.
The structure of this paper is outlined as follows: Section 2 describes the CODAR high frequency radar system, the three-dimensional hydrodynamic model EFDC, Optimal Interpolation DA algorithm and assessment criteria. Section 3 presents the results of hindcasting and forecasting. Discussion and conclusions are presented in Section 4.
Section snippets
CODAR radar observations
A CODAR system measures near-surface ocean currents in a coastal area (Kaplan et al., 2005). One such system has been deployed in Galway Bay on Ireland's west coast since 2011 summer (see Fig. 1). When the radar signal scatters off a wave whose wave length is exactly equal to half of the transmitted signal wavelength, the radar signal can return information of surface currents (Haus et al., 2000, Wang et al., 2004). A single HF radar station determines radial components of surface currents
Results
Improvement of model forecasting accuracy through the application of DA is a multi-step process. The first step of the evaluation of the OI algorithm is to analyse effects of the spatial correlation length on model accuracy and determine the best fit model length scale, see details in Section 3.1. The second step is to examine influences of the DA CL on forecasting and to extend the improvements for forecasting as studied in Section 3.2. The main results from various assimilation models are
Discussion and conclusions
Investigations into sensitivity of both spatial correlation length and DA CL on hindcasting and forecasting of surface currents were performed in detail in this study. The main conclusions in the work are:
- (a)
Detailed sensitivity experiments of horizontal correlation length indicate that model was not sensitive to this parameter. Although hourly assimilation of the radar data improved hindcast, hourly updating of surface currents in the models did not have significant influences on forecasting.
Acknowledgements
This research is supported by funding from China Scholarship Council (CSC) and National University of Ireland, Galway. We would like to thank Informatics Research Unit for Sustainable Engineering (IRUSE) for providing the weather data and Ireland's High-Performance Computing Centre (ICHEC) for providing computation services. The authors also would like to express their gratitude to the anonymous reviewers for their very useful and constructive suggestions and comments that helped in improving
References (41)
Ensemble Kalman filter for data assimilation
Comput. Geosci.
(2013)- et al.
Remote Radar Measurement of Shelf Currents off Key Largo, Florida, U.S.A
Estuar., Coast. Shelf Sci.
(2000) - et al.
Sensitivity of Self-Organizing Map surface current patterns to the use of radial vs. Cartesian input vectors measured by high-frequency radars
Radars Comput. Geosci.
(2015) - et al.
The Regional Ocean Modeling System (ROMS) 4-dimensional variational data assimilation systems: part I – System overview and formulation
Prog. Oceanogr.
(2011) - et al.
Physical and numerical investigation of the hydrodynamic implications of aquaculture farms
Aquac. Eng.
(2013) - et al.
Observation and modeling of tide- and wind-induced surface currents at Galway Bay
Water Sci. Eng.
(2015) - et al.
Assimilation of altimeter data into a quasigeostrophic ocean model using optimal interpolation and eofs
J. Mar. Syst.
(1995) - et al.
Impact of assimilating altimeter data on wave predictions in the western Iberian coast
Ocean Model.
(2015) - et al.
Variational assimilation of HF radar surface currents in a coastal ocean model off Oregon
Ocean Model.
(2012) - Agostinho, P., Gil, A., Galeano, J.C., 2012. N.U.I Galway CODAR SeaSonde HF Radar System. Qualitas remos, Galway,...
Atmospheric Data Analysis
The Ensemble Kalman Filter: theoretical formulation and practical implementation
Ocean Dyn.
Assimilation of HF radar-derived surface currents on tidal-timescales
J. Oper. Oceanogr.
Background errors in a global wave model determined from altimeter data
J. Geophys. Res.
Case study: modeling of sediment transport and wind-wave impact in Lake Okeechobee
J. Hydraul. Eng.
A new approach to linear filtering and prediction problems
J. Basic Eng.
Cited by (8)
Computational assessment of smooth and rough parameter dependence of statistics in chaotic dynamical systems
2021, Communications in Nonlinear Science and Numerical SimulationCitation Excerpt :The sensitivities we are interested in are derivatives of long-time averages of observables with respect to parameters. These derivatives are used in gradient-based approaches for design, optimization and control [1–6], variational data assimilation [7,8] and uncertainty quantification applications [9,10] in various fields of science and engineering. In chaotic dynamical systems describing complex phenomena, such as the Earth’s climate [11,12], molecular transition [13], and turbulent flow [14], the high sensitivity to perturbations poses a challenge to computing the parametric derivatives of statistics.
A fusion-based data assimilation framework for runoff prediction considering multiple sources of precipitation
2023, Hydrological Sciences JournalA multiscale sequential data assimilation system and its application to short-term traffic flow prediction
2020, Sensors and MaterialsCharacteristics of coastal currents based on High Frequency radar and ADCP observations in the Strait of Georgia
2018, IOP Conference Series: Earth and Environmental Science