Mapping wildfire occurrence at regional scale

When assessing fire danger, interpolation of the dependent variable—historic fire occurrence—is required in order to statistically compare and analyze it with human factors, environmental parameters and census statistics. To confirm the compatibility between the distinct data types, occasionally, for this kind of spatial analysis, historical observations of the primary wildland fire (given as x and y coordinates) must be transformed either to continuous surfaces or to area data. The simple overlay approach converts single point observations to area data. However, this procedure assumes lack of spatial uncertainties that would otherwise result in serious errors caused by the positional inaccuracies of the original point observations. we used kernel density interpolation to convert the original data on wildland fire ignition into an expression of areal units, defined by a raster grid and, subsequently, by the administrative borders of the municipalities in two study areas in Spain. By overlaying a normal bivariate probability density function (kernel) over each point observation, each ignition point was considered an uncertain point location rather than an exact one.