Application of HYSPLIT-STILT to Identify Source Regions of CO2 Affecting Monitored Levels in San Jose, California

Comparisons of simulated versus observed diurnal trends of CO2 concentrations for typical sea-breeze conditions in central San Jose, California will be presented. The simulations are of hourly CO2, composited over several sea-breeze days during spring of 2016, from HYSPLIT run in backwards-concentration STILT-emulation mode (HYSPLIT-STILT). Simulations will be compared to observed hourly CO2 levels measured on the composited days by continuous monitoring at San Jose State University (SJSU). Meteorological fields to HYSPLIT-STILT are from WRF V3.8 run with 1-km inner domain. The simulated CO2 levels at a given hour are calculated by multiplying HYSPLIT-STILT twelve-hour back-concentration footprints corresponding to the hour by a CO2 surface emission flux field point by point across the footprint for each of these twelve hours, and adding the results.  This emission field has been constructed by us based on the most recent Bay Area Air Quality Management District CO2 emission inventory. The process is repeated for each hour of the simulation to arrive at simulated diurnal CO2 trends.

The STILT methodology is direct and efficient at identifying source regions of air most affecting measurements at a particular site. Footprints from application of the method directly identify these source regions. It is an efficient method because, unlike traditional grid-modeling approaches, the entire pollution field across the modeling domain does not need to be calculated. The potential for advancing understanding of urban air quality that is focused on a region around a particular site is therefore high, especially within multi-source, complex urban airshed like San Jose. Here, we apply the technique to sea-breeze conditions, the climatologically prevailing conditions from spring and summer in San Jose.

Our first application of the method, over a two-day period from 1200Z May 16 to 1800Z May 18, 2016 (journal article in preparation), reproduces the basic hourly CO2 variations at the site. Model calculated period-averaged concentration are 427 ppm, very close to the measured values at the SJSU site of 430 ppm. This match indicates the basic accuracy of the HYSPLIT-STILT application and of the emission field to which the footprints are coupled. Discrepancies between model and observations are seen in the details of the hour-to-hour variation, which are being researched. We anticipate by the time of the January meeting to have a better understanding of the modeling issues affecting these discrepencies, which will be reported on at the conference.

Interesting simulated patterns of source regions affecting SJSU site measurements are seen in the similation. An example is seen in the accompanying figure for an early-morning hour of May 17, 2016. Here, source regions for early-morning CO2 air at the site are simulated both north and south of SJSU associated with branches of the sea-breeze that track northwards to San Jose through Monterey Bay to the south, and east and southwards to San Jose through the Crystal Springs / San Bruno Gap area of the San Francisco Bay peninsula to the north. This happens due to the shift in wind direction during the evening from strong daytime/early evening northwesterly to weaker nighttime/early-morning southeasterly. The time of shift to southeasterly flow is simulated to begin to the south of San Jose around midnight, subsequently propogating northwestward through the city to produce a northwesterly to southeasterly wind shift at SJSU in the pre-dawn hours. This is broadly similar to what is seen in wind observations at SJSU and other nearby sites in the area for this evening. Because of this, simulated HYSPLIT particles during early-morning hours disperse into both branches of the sea-breeze, and the model footprints in the figure reflect this.  

Work is ongoing to further evaluate the utility of the modeling method for the region. Sensitivity to WRF simulation details, particularly with respect PBL scheme, is ongoing. WRF-CHEM modeling is also being planned to compare to HYSPLIT-STILT for particular cases. Better modeling of near-field sources, not as well resolved by the HYSPLIT-STILT 1-km grid, is also being explored by use of dispersion models.