LAUSR

Shapero et al.1 claim the conclusions of our study2 are undermined by “erroneous assumptions and misapplied technical approaches.” However, their letter ignores most of the evidence that we provided in our article, incorrectly describes the methods we used, and fails to identify any errors in our work. We quantified the contribution of preharvest sugarcane burning to air concentrations of fine particulate matter [PM ≤2:5 lm in aerodynamic diameter (PM2:5)] using multiple independent data sets and lines of evidence. We showed that ground-based PM2:5 monitors and satellite-derived surface PM2:5 observations both recorded higher average PM2:5 concentrations in Florida’s sugarcane-growing region during harvest burning season than the rest of the year, a pattern not seen elsewhere on the Florida peninsula. This pattern matched the magnitude and spatial extent of PM2:5 expected from sugarcane fires, which we simulated in a state-of-the-art atmospheric dispersion model using emissions derived from sugarcane burn authorization records. The mean diurnal cycle of PM2:5 in Belle Glade, Florida, a city surrounded by sugarcane fields, also featured a peak shortly after sugarcane fires began in the morning; this peak did not appear in non-harvest months. The consistency and corroboration between these independent sources provided a confident estimate of PM2:5 concentrations caused by sugarcane fires. Our results also cohere with numerous past studies showing that PM2:5 from many sources causes premature mortality3,4 and that PM2:5 from agricultural fires specifically is associated with around 600 premature deaths per year in the United States.5 Much of the letter by Shapero et al.1 concerns a statistical significance test (p value) at one surface monitor site, but their critique does not apply to the methods we used. As we wrote in our paper, PM2:5 surface observations were averaged over harvest and non-harvest seasons (6 months each) before performing the significance test. The statements by Shapero et al.1 about clustered standard errors on subseasonal timescales are, therefore, irrelevant and misrepresent our paper. Shapero et al. are incorrect in saying that our analysis neglected “meteorological conditions” or the “temporal specifics of actual harvest activities.”1 In reality, we estimated the contribution of sugarcane fires to PM2:5 using an atmospheric dispersion model driven by high-resolution meteorological data from the National Oceanic and Atmospheric Administration, and we accounted for the date and location of every sugarcane fire, as well as the times of day when harvest fires occurred. Shapero et al. also say that our analysis contained large uncertainties and high biases,1 but all the sources of uncertainty they listed (e.g., fuel loading, emission factors, plume rise, secondary aerosol) were explicitly accounted for in the confidence intervals reported in our article. Our study focused on the years 2009–2018 because the satellite-derived PM2:5 data,6 which informed the health impacts assessment, were not available for later years at the time of our analysis. If we look at the PM2:5 monitor in Belle Glade (Figure 1), as Shapero et al.1 suggest, we see that the mean PM2:5 concentrations were consistently higher during harvest seasons than non-harvest seasons for 2009–2017, except for a couple years when large wildfires burned nearby during summer. In more recent years, harvest season PM2:5 concentrations have not been as elevated, coinciding with new restrictions on sugarcane burning implemented in 2019.7 The change in harvest season mean PM2:5 after 2019 is therefore consistent with sugarcane fires contributing to PM2:5 during the years of our study. Future work should examine whether the recent PM2:5 changes in Belle Glade are regionally representative or limited to the vicinity of the monitor, and whether the changes persist in future harvest seasons. In summary, we maintain that the letter by Shapero et al.1 contains errors, incorrectly describes our analysis, and does not identify any source of uncertainty that was not already accounted for in our article. We stand behind our analysis and the central findings of our article.