Evaluation of fire weather forecasts using PM2.5 sensitivity analysis

Fire weather forecasts are used by land and wildlife managers to determine when meteorological and fuel conditions are suitable to conduct prescribed burning. In this work, we investigate the sensitivity of ambient PM2.5 to various fire and meteorological variables in a spatial setting that is typical for the southeastern US, where prescribed fires are the single largest source of fine particulate matter. We use the method of principle components regression to estimate sensitivity of PM2.5, measured at a monitoring site in Jacksonville, NC (JVL), to fire data and observed and forecast meteorological variables. Fire data were gathered from prescribed fire activity used for ecological management at Marine Corps Base Camp Lejeune, extending 10-50 km south from the PM2.5 monitor. Principal components analysis (PCA) was run on 10 data sets that included acres of prescribed burning activity (PB) along with meteorological forecast data alone or in combination with observations. For each data set, observed PM2.5 (unitless) was regressed against PCA scores from the first seven principal components (explaining at least 80% of total variance). PM2.5 showed significant sensitivity to PB: 3.6 +/- 2.2 mu g m(-3) per 1000 acres burned at the investigated distance scale of 10-50 km. Applying this sensitivity to the available activity data revealed a prescribed burning source contribution to measured PM2.5 of up to 25% on a given day. PM2.5 showed a positive sensitivity to relative humidity and temperature, and was also sensitive to wind direction, indicating the capture of more regional aerosol processing and transport effects. As expected, PM2.5 had a negative sensitivity to dispersive variables but only showed a statistically significant negative sensitivity to ventilation rate, highlighting the importance of this parameter to fire managers. A positive sensitivity to forecast precipitation was found, consistent with the practice of conducting prescribed burning on days when rain can naturally extinguish fires. Perhaps most importantly for land managers, our analysis suggests that instead of relying on the forecasts from a day before, prescribed burning decisions should be based on the forecasts released the morning of the burn when possible, since these data were more stable and yielded more statistically robust results. (C) 2016 Elsevier Ltd. All rights reserved.