PROBLEMS OF VERIFICATION OF NUMERICAL MODELING OF MARINE REGIONS

Abstract

This paper presents an approach to assessing the discrepancies between model data and in-situ observations (i. e., one of the stages of data verification), based on the assumption of continuous variability in a natural system (a marine region) and methods of functional analysis . The essence of this approach is that the in-situ and model datasets of seawater temperature are considered discrete analogues of a real continuous process of change in one of the parameters of a natural system. Thus, the problem of assessing the “proximity” of discrete time series can be replaced by a comparison of their continuous prototypes, a well-established methods in classical functional analysis. It is shown that the use of the point-wise proximity criteria (MAE) or the root mean square deviation (RMSE) is sufficient to detect discrepancies between measured (reference) and model (verified) datasets for any thermodynamic parameter (in this paper, seawater temperature) in a specific marine region. However, the dimensionless criteria of data discrepancy proposed in the paper may be useful for determining the quality of the model’s accounting for subgrid processes. The effectiveness of these criteria is illustrated by comparing Baltic Sea Physical Reanalysis data with instrumentally measured seawater temperature during the autumn cooling of coastal waters in the Baltic Sea.