Landsat Land Use Classification for Assessing Health Risk from Industrial Air Pollution

Resumo

This study investigated to what extent risk estimates can be modified by involving in the process readily available space data and well-tested processing methods. We classify Landsat data for these types using the support vector algorithm and small characteristic training sites for each type. The dispersion modeling problem, unlike most classification tasks, is tolerant of unclassified areas. We show that the classification obtained is better than available global maps based on MODIS or Landsat. For a large chemical plant, we perform dispersion modeling and calculate the maximal hourly concentrations and acute risk from the main pollutant. We compare several versions of calculated risk based on the surface parameters assessed from global maps and variants of Landsat classification to show that the latter are twice as accurate as the former (with a ~20 and ~40% error, respectively). Risk estimates are shown to vary considerably (by ~25%) depending on the yearly set of Landsat data used, so that using multi-year data is a must, unless land use changes considerably over the period. Thus, in assessing the hazard from air pollution from any specific plant, which is an obligatory procedure for establishing the plant’s sanitary protection zone and obtaining the pollutant emission permit, it is desirable to use Landsat data.