Application of geostationary satellite and high-resolution meteorology data in estimating hourly PM2.5 levels during the Camp Fire episode in California

Abstract

Wildland fire smoke contains large amounts of PM_2.5 that can traverse tens to hundreds of kilometers, resulting in significant deterioration of air quality and excess mortality and morbidity in downwind regions. Estimating PM_2.5 levels while considering the impact of wildfire smoke has been challenging due to the lack of ground monitoring coverage near the smoke plumes. We aim to estimate total PM_2.5 concentration during the Camp Fire episode, the deadliest wildland fire in California history. Our random forest (RF) model combines calibrated low-cost sensor data (PurpleAir) with regulatory monitor measurements (Air Quality System, AQS) to bolster ground observations, Geostationary Operational Environmental Satellite-16 (GOES-16)'s high temporal resolution to achieve hourly predictions, and oversampling techniques (Synthetic Minority Oversampling Technique, SMOTE) to reduce model underestimation at high PM_2.5 levels. In addition, meteorological fields at 3 km resolution from the High-Resolution Rapid Refresh model and land use variables were also included in the model. Our AQS-only model achieved an out of bag (OOB) R² (RMSE) of 0.84 (12.00 μg/m³) and spatial and temporal cross-validation (CV) R² (RMSE) of 0.74 (16.28 μg/m³) and 0.73 (16.58 μg/m³), respectively. Our AQS + Weighted PurpleAir Model achieved OOB R² (RMSE) of 0.86 (9.52 μg/m³) and spatial and temporal CV R² (RMSE) of 0.75 (14.93 μg/m³) and 0.79 (11.89 μg/m³), respectively. Our AQS + Weighted PurpleAir + SMOTE Model achieved OOB R² (RMSE) of 0.92 (10.44 μg/m³) and spatial and temporal CV R² (RMSE) of 0.84 (12.36 μg/m³) and 0.85 (14.88 μg/m³), respectively. Hourly predictions from our model may aid in epidemiological investigations of intense and acute exposure to PM_2.5 during the Camp Fire episode.

Keywords: AOD; GOES16; PM2.5; Remote sensing; SMOTE; Weighted Random Forest; Wildland fire.

Fig. 1.

Study domain of California. EPA AQS monitors are pictured in red, PurpleAir sensors are in blue.

Fig. 2.

Panel of density scatter plots of 10-fold spatial CV measured vs. predicted PM 2.5 concentrations from (A) AQS-only Model, (B) AQS + Weighted PurpleAir Model, and (C) AQS + Weighted PurpleAir + SMOTE Model, and the 10-fold temporal cross-validation measured vs. predicted PM_2.5 concentrations from the three models (D) AQS-only Model, (E) AQS + Weighted PurpleAir Model, and (F) AQS + Weighted PurpleAir + SMOTE Model. The dotted red line designates the slope and intercept while the solid blue line designates a 0 intercept with a slope of 1.

Fig. 3.

Predictions from all three models (AQS-only, AQS + Weighted PurpleAir, AQS + Weighted PurpleAir + SMOTE) at 12:00 pm on November 16th, 2018. Area of the smoke plume remains the same in all models; however, PM 2.5 levels increase as PurpleAir and SMOTE is added.

Fig. 4.

Hourly prediction maps of PM 2.5 in μg/m³ from the weighted RF and SMOTE model in California on November 16, 2018. Recorded ground measurements were highest on this day.

Fig. 5.

Comparison of hourly prediction maps of PM 2.5 in μg/m³ with the true color composite images from MODIS at noontime on November 8 and November 16, the day the Camp Fire started and the day with the highest recorded ground measurements, respectively.