The FLEXPART Lagrangain particle dispersion model is widely used in the research community to study the transport of anthropogenic emissions, biomass burning, and ozone. FLEXPART originally was designed to ingest meteorological data produced by the EMCWF global model. It since has been adapted to GFS input data as well. FLEXPART recently was modified to ingest high spatial resolution meteorological data from the Weather Research and Forecasting (WRF) mesoscale model. This new FLEXPART-WRF (F-W) model can utilize gridded WRF data on multiple nested grids at hourly intervals. Furthermore, parameters such as friction velocity, planetary boundary layer height, and surface heat flux now are taken directly from the WRF output instead of being internally calculated as in the original FLEXPART model. This makes F-W desirable for applications that require accurate representations of mesoscale phenomena, such as in regions of complex terrain. There have been no sensitivity studies that evaluate the performance of F-W using real-world emissions data. Our paper will evaluate F-W results in a dispersion simulation against smoke plumes from biomass burning in a region of complex terrain in southern Asia.

Previous studies have shown that resolving the mesoscale eddies and vertical mass transport associated with complex terrain is vitally important, especially in short-range dispersion simulations. Additionally, it has become apparent that higher resolution data are needed when emissions sources are sub-grid scale due to the interpolation error associated with transforming the gridded wind data to the source location. Therefore, dispersion models that rely on external meteorological data are limited by the structure of, and method in which those data were created. Our paper will highlight the advantages of using the WRF mesoscale model to drive dispersion forecasts.

The evaluation was done in two parts: First, WRF was integrated out to 72 h using horizontal and vertical grid resolutions ranging from 48 km to < 2 km and 30 to 60 levels, respectively. The WRF simulated winds and temperatures then were compared to observations, while the WRF precipitation rates were compared to satellite-derived estimates. Root mean square errors (RMSE) were computed for winds, and object-based (OB) analyses were performed on the precipitation fields to compute skill scores. The ability of WRF to produce accurate precipitation rates was evaluated instead of vertical velocity under the assumption that more accurate vertical mass transport produces more accurate precipitation patterns. Then, once the WRF grid configuration that produced a desirable minimum RMSE was determined, F-W was integrated for the same 72 h duration using the WRF data as input. Next, the F-W simulated CO plume structure was compared to the observed MODIS mass concentration and CALIPSO cloud-aerosol lidar attenuated backscatter products. CO was used in our simulations since biomass burning releases large amounts of carbon, which approximately follows the path of the air-borne aerosols ejected by the fires.

In contrast to tracer experiments, this evaluation method allows us to verify the performance of the WRF and F-W system on observed phenomena since fires occurred near the region and there was remote sensing at some point during the simulations. Additionally, this procedure allows us to evaluate the performance of WRF when traditional statistical measures such as RMSE break down at high resolutions, as suggested by numerous studies.