File:Comparisons between four fire emission inventories.webp

English: "a, Correlation matrix showing the Pearson correlations between station observed all-source daily PM2.5, and GEOS-Chem simulated and machine learning (ML) estimated all-source PM2.5 based on different fire emission inventories in 2012. b, Correlation matrix showing the Pearson correlations between GEOS-Chem simulated and ML estimated fire-sourced PM2.5 based on different fire emission inventories in 2012. c, Correlation matrix showing the Pearson correlations between station observed all-source daily O3, and GEOS-Chem simulated and ML estimated all-source O3 based on different fire emission inventories in 2012. d, Correlation matrix showing the Pearson correlations between GEOS-Chem simulated and ML estimated fire-sourced O3 based on different fire emission inventories in 2012. e, Table showing the performance of GEOS-Chem outputs and machine learning estimates based on different fire emission inventories in 2012 in validation against station observed daily all-source PM2.5, all-source O3 and smoke PM2.5. RMSE, root mean squared error, in µg/m3. For all the panels in this figure, the ML estimates were based on the ML model trained in out-of-sample stations (according to the 10-fold spatial cross-validation). For each specific fire emission inventory (e.g., FINN), the ML predictors include corresponding GEOS-Chem simulated all-source PM2.5 or O3 (e.g., the FINN GEOS-Chem all-source PM2.5) and other predictors mentioned in equation 1 (or 2) in the Methods section."