Interpretable machine learning prediction of fire emissions and 1 comparison with FireMIP process-based models 2

Annual burned areas in the United States have increased twofold during the past decades. With more 8 large fires resulting in more emissions of fine particulate matter, an accurate prediction of fire emissions is critical 9 for quantifying the impacts of fires on air quality, human health, and climate. This study aims to construct a machine 10 learning (ML) model with game-theory interpretation to predict monthly fire emissions over the contiguous US 11 and to understand the controlling factors of fire emissions. By comparing the predicted fire PM2.5 emissions from 12 the interpretable ML model with the Global Fire Emissions Database (GFED) observations and predictions from 13 process-based models in the Fire Modeling Intercomparison Project (FireMIP), the ML model is also used to 14 diagnose the process-based models to inform future development. Results show promising performance for the ML 15 model, Community Land Model (CLM), and Joint UK Land Environment Simulator-Interactive Fire And Emission 16 Algorithm For Natural Environments (JULES-INFERNO) in reproducing the spatial distributions, seasonality, and 17 interannual variability of fire emissions over CONUS. Regional analysis shows that only the ML model and CLM 18 simulate the realistic interannual variability of fire emissions for most of the subregions (r>0.95 for ML and 19 r=0.14~0.70 for CLM), except for Mediterranean California, where all the models perform poorly (r=0.74 for ML 20 and r<0.30 for the FireMIP models). Regarding seasonality, most models capture the peak emission in July over 21 western US. However, all models except for the ML model fail to reproduce the bimodal peaks in July and October 22 over Mediterranean California, which may be explained by the coarse spatial resolutions of the processed-based 23 models or atmospheric forcing data or limitations in model parameterizations for capturing the effects of Santa Ana 24 winds on fire activity. Furthermore, most models struggle to capture the spring peak in emissions in southeastern 25 US, probably due to underrepresentation of human effects and the influences of winter dryness on fires in the 26 models. As for extreme events, both the ML model and CLM successfully reproduce the frequency map of extreme 27 emission occurrence but overestimate the number of months with extremely large fire emissions. Comparing the 28 fire PM2.5 emissions from the interpretable ML model with process-based fire models highlights their strengths and 29 uncertainties for regional analysis and prediction and provides useful insights on future directions for model 30 improvements. 31