Background/aim Both mechanistic and Land-Use-Regresison (LUR) models have been extensively used for air quality assessment. Mechanistic models are expensive to collect the data for and run at a high spatial resolution. LUR models usually lack the required physical sensibility for a high temporal resolution. The Optimised Dispersion Model (ODM) is a fresh approach, which attempts to combine the two.

Methods Available data included: half-hourly ambient monitoring of nitrogen oxides; GPS-based traffic activity data; emission estimates from every registered industrial stack; and hourly wind fields at 1 km spatial separation. These data were combined into a puff-like model, which uniquely included 10 optimisation parameters. The values of the parameters were automatically calculated to provide a least-squares fit with measurements at each time-point. This new formulation included the concentrations in one time-point as sources for the next time-point, making the model more realistic and increasing computational efficiency, such that is was possible to only take into account a limited number of cells affecting each other cell.

Results The new model provided a modest improvement (half-hourly cross-validated MSPC of 0.25 vs 0.22) over a previous version at the coastal plain, and a much greater improvement (0.20 vs 0.05) at the Haifa district. The improvement was especially great for areas influenced by industrial stacks and at times when the wind direction was highly heterogeneous in space. The models were most similar when the wind was easterly. Re-circulation of pollutants was observed in the new model results, which may have played a part in the higher residuals obtained by previous versions.

Conclusion The new ODM can be used in different areas in an automated fashion and without changing the model formulation. For this, input of only reasonable quality is needed (continuous ambient measurements, wind direction and locations and relative magnitude of emission sources). Given these data, it performs better than other popular models and provides results in high spatio-temporal resolution.