Browsing by Subject "Model validation"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Open Access
    Evaluation of the operational Mercator Global Ocean analysis and forecast system modelled current vector product with in-situ ADCP data in the Southern Benguela upwelling system
    (2025) Gammon, Emily; Clark; Barry; Schmidt, Kevin; Vichi, Marcello
    Global ocean current vector datasets are frequently used to force, or are assimilated into, local and regional particle tracking and pollutant dispersal models, metocean forecasting platforms, climate analyses, and natural resource management applications. The Operational Mercator global ocean analysis and forecast system (Mercator data product), made publicly available in 2022, provides near-real-time vector components of global ocean currents at 50 vertical levels (depths) in the water column with a horizontal resolution of 1/12 degree. Validation of ocean forecasting and simulation models is required to assess a modelled product's ability to represent the local environment. Validation is done using in-situ data such as those gathered by Acoustic Doppler Current Profilers (ADCP). In this study, an upward-facing Nortek Signature 250 ADCP was deployed at a depth of 123 m off the Southern Namibian coast for ten months. The resulting current profile data were used to validate the Mercator data product for the same area. Data collected by the ADCP and simulated data from the NEMO model were matched spatially (nearest-neighbour) and temporally (rolling average), and the similarity of current vector components was statistically investigated. Spearman Rank-Order Correlation, root-mean-square-error and bias were calculated for the full time series. The seasonal percent occurrence of current vector magnitude and direction were subsequently derived to assess the Mercator data product's ability to represent seasonal variability. The Mercator data product captured the seasonal variation in the magnitude of the nearshore southern Benguela Upwelling System for both the zonal and meridional components well. The Mercator data product represented much lower complexity in the direction of bulk flow GAMMON iii within the system but did contain the daily oscillation pattern measured by the ADCP. The Benguela Current is a physically complex eastern boundary current spanning three countries, used extensively for fishing, mining, maritime transit, and oil extraction. The ocean current outputs from the Mercator data product could allow for improved pollution modelling (oil dispersion or plume modelling), improved fuel optimisation models for maritime transit, and productivity prediction to support fisheries management without the cost of deploying devices to measure ocean currents in situ.
  • Thumbnail Image
    Item
    Open Access
    How deep is your model? Network topology selection from a model validation perspective
    (2022-01-03) Nowaczyk, Nikolai; Kienitz, Jörg; Acar, Sarp K; Liang, Qian
    Deep learning is a powerful tool, which is becoming increasingly popular in financial modeling. However, model validation requirements such as SR 11-7 pose a significant obstacle to the deployment of neural networks in a bank’s production system. Their typically high number of (hyper-)parameters poses a particular challenge to model selection, benchmarking and documentation. We present a simple grid based method together with an open source implementation and show how this pragmatically satisfies model validation requirements. We illustrate the method by learning the option pricing formula in the Black–Scholes and the Heston model.