Insights into WRF-Chem Sensitivity in a Coastal Region of Morocco: Chemical Mechanisms, Nesting Options, and Physical Parameterization

The performance of the Weather Research and Forecasting model coupled with chemistry (WRF-Chem) depends on the accuracy of input data and parameterisation schemes. This study examined the sensitivity of WRF-Chem version 4.4.2 to various domain configurations, chemical mechanisms, and physics parameterizations. The model was applied to predict air quality pollutants (PM₁₀, O₃, CO, NO₂, SO₂) and meteorological variables (wind speed and temperature) for the first time over the Moroccan city of Agadir, which features complex topography and land use. Multiple simulations tested the sensitivity of these variables to different chemical mechanisms (MOZART, RACM, GOCART), nesting configurations (three nested domains at a 1:4 ratio and four nested domains at a 1:3 ratio), and planetary boundary layer (PBL) parameterizations (YSU, MYJ, MYNN2, QNSE). The modelled pollutant and meteorological data were then validated against surface observations using various statistical metrics. Results indicated that ozone (O₃) and NO₂ show limited sensitivity to domain configuration and chemical mechanisms, implying they are primarily influenced by local factors, especially emission patterns. Conversely, CO exhibits greater sensitivity to nesting choices due to its dependence on accurately capturing local emissions and atmospheric mixing. Similarly, concentrations of PM₁₀, SO₂, and CO strongly depend on the representation of physical and chemical processes within domain configurations and chemical schemes. Ozone estimates, however, are highly affected by physical parameterizations, particularly those influencing temperature, highlighting the importance of temperature and sunlight in ozone formation. Sensitivity analysis revealed that optimal configurations differ by variable, based on both spatial variation and statistical assessment. For physics parameterizations, the YSU PBL scheme was preferred for predicting wind speed and temperature. Among chemical schemes, GOCART proved most effective for PM₁₀, SO₂, and CO, while MOZART was best suited for O₃ and NO₂. The most efficient nesting setup was identified as three nested domains at a 1:4 ratio, balancing accuracy with computational efficiency. These findings suggest significant potential for improving air quality predictions in the region, though further refinement-particularly for PM₁₀, which exhibited the lowest model accuracy-remains necessary for future research.