Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields: Comparison of Three Techniques

Abstract

This paper presents three simple techniques for fusing observations and numerical model predictions. The techniques rely on model/observation bias being considered either as error free, or containing some uncertainty, the latter mitigated with a Kalman filter approach or a spatial smoothing method. The fusion techniques are applied to the daily maximum 8-hour average ozone concentrations observed in the New York state area during summer 2001. Classical evaluation metrics (mean absolute bias, mean squared error, correlation, etc.) show that fused predictions are not better than a simple interpolation of observations. However, fused maps better reproduce the spatial texture of the model predictions.