Data Assimilation of forecasted errors in hydrodynamic models using inter-model correlations

Data-assimilation techniques of the Kalman filter type are considered to be the state-of-the-art approach for combining data information and deterministic numerical models with the objective of operational forecasting. This paper introduces, as an alternative, a faster and simpler data-assimilation technique that exploits inter-model correlations to distribute predicted errors. This scheme is performed in two steps: (i) prediction of the deterministic model errors at observation points using so-called local linear models and (ii) distribution of the forecasted errors over the computational domain employing a scheme based on deterministic inter-model correlations which describe the spatial nature of error structure. The method’s advantage is that systematic error can be predicted by the error correction scheme, while the dynamics remain described by the deterministic model, which also establishes a basis for the spatial error distribution scheme. This relatively simple approach is inspired by original Kalman filter techniques but distinguishes error prediction and distribution in two different stages, hence allowing for data-driven error forecasting and off-line correction. In order to test the scheme’s performance, a deterministic model of an artificial bay was constructed and run. The system was driven by specific forcing conditions and characterized by physical parameters that, in subsequent simulations, were deliberately manipulated to introduce errors into the model and test the scheme’s capability.