Exploratory spatial data analysis of global MODIS active fire data

Abstract

We performed an exploratory spatial data analysis (ESDA) of autocorrelation patterns in the NASA MODIS MCD14ML Collection 5 active fire dataset, for the period 2001–2009, at the global scale. The dataset was screened, resulting in an annual rate of false alarms and non-vegetation fires ranging from a minimum of 3.1% in 2003 to a maximum of 4.4% in 2001. Hot bare soils and gas flares were the major sources of false alarms and non-vegetation fires. The data were aggregated at 0.5° resolution for the global and local spatial autocorrelation Fire counts were found to be positively correlated up to distances of around 200km, and negatively for larger distances. A value of 0.80 (p=0.001, α=0.05) for Moran's I indicates strong spatial autocorrelation between fires at global scale, with 60% of all cells displaying significant positive or negative spatial correlation. Different types of spatial autocorrelation were mapped and regression diagnostics allowed for the identification of spatial outlier cells, with fire counts much higher or lower than expected, considering their spatial context.