Fig 1.
Study area map showing air quality monitoring stations used in analysis and example of VIIRS SNPP hotspots for one day and night between 2019-04-28 midday and 2019-04-29 midday.
150 km buffer around Chullora is black circle. West, South and North region were used for calculating separate fire area variables. Basemap: © OpenStreetMap contributors.
Table 1.
Predictor variables used in modelling.
Fig 2.
Histograms of daily mean PM2.5 of all five AQS and maximum daily PM2.5 from all five AQS (i.e. value from AQS with highest daily mean PM2.5).
“Daily” here means midday to midday.
Fig 3.
Scatter plots of mean Sydney PM2.5 vs predictor variables, with single variable GAM smooths fit via the “mgcv” package in R.
Table 2.
GAM accuracy results for models with all variables and best models from the model selection process.
Fig 4.
Bayesian mean effects plots for each variable for mean daily Sydney PM2.5 model.
The grey bands represent, from darkest to lightest grey, 0.5, 0.8 and 0.95 credible intervals. Effect of each fire area variable is shown with other two fire area variables held at zero. Other variables held at mean values. U and V wind components have been converted to wind speed and direction effects (bands show 0.5 credible interval). Wind directions are standard wind direction in degrees, i.e. 180 = southerly wind.
Fig 5.
Predictive distributions from the mean and maximum daily Sydney PM2.5 models.
Predictions for three levels of fire area west, with other variables at set levels: Lag PM2.5 = 10 μgm-3, ventilation index = 500 m2 s−1, temperature = 10 C, fire areas south and north = 100 ha, coast wind speed and direction = 10 km h-1 and easterly (sea breeze), inland wind speed and direction = 10 km h-1 and westerly. Dotted and dashed vertical lines are thresholds defined at 15 μgm-3 and 25 μgm-3 respectively, with percent of distribution > thresholds indicated in text within the plots.
Fig 6.
Tile plot of predictions for mean Sydney PM2.5 model (top) and maximum Sydney PM2.5 model (bottom).
Each coloured grid square shows the percent chance, under different predictor conditions, of exceedance above 25 μgm-3 (darker = higher chance, also in red text), i.e. percent of predictive distribution > 25 μgm-3. Other variables held at: daily temperature 14 C, fire areas north and south = 0 ha, inland (Katoomba) wind speed and direction = 15 km h-1 and westerly. Wind speeds and directions were calculated from the relevant U and V wind variables.