Fine resolution modeling with CMAQ-adjoint

Abstrakt

Inverse modeling of emissions based on 4DVar approach appears to be a very promising method for better specification of emission sources. Existing experiments utilizing different CTM models and different kinds of observations have been performed in global scale and in coarse resolution in mesoscale. Fine horizontal resolution brings specific benefits for adjoint modeling; more variability is captured in species like NO2 which are highly dependent on time and location and observations from stations with smaller representativity are easier comparable with model values. The main limitation of adjoint modeling in fine resolution is the computational complexity of the method; efficient parallelization is a necessary condition for practical feasibility of such experiments. Some limits of the parallel scalability are given in CMAQ model code itself and additional limitations are introduced in CMAQ-adjoint code. We identify the parts which are critical for efficient parallelization of the code by profiling the CMAQ-adjoint on application and system level under different conditions; we propose and test some enhancements of the code leading to improvement of scalability. The performance of the method is tested in the framework of an experiment performed on a fine resolution domain covering Central Europe.