REAL-TIME FORECAST OF AIR-POLLUTION EPISODES IN THE VENETIAN REGION .2. KALMAN PREDICTOR

Real-time prediction of air pollution means forecast of future ground-level concentrations on the basis of current information about meteorology, emission and concentrations themselves. The paper illustrates how the episode forecast performance of the advection-diffusion model (described in Part 1) can be improved by: (i) embedding the model into a stochastic version, by introducing into the numerical solution scheme of the advection-diffusion equation proper random terms (noises), specified in a statistical sense and representative of all the inaccuracies of the model itself (wrong inputs, numerical errors, simplifications of the actual physical mechanism, ...); (ii) applying the well-known Kalman prediction technique for real-time forecast to such stochastic version of the model. The results of such a procedure have been satisfactory in the case of the Venetian lagoon sulphur dioxide pollution. In particular, the four-hour ahead forecasts of episodes have been drastically improved with respect to those obtained by means of the 'deterministic predictor' illustrated in Part 1. © 1979 IPC Business Press Ltd.