Building historical gridded weather data sets for fire program analysis

Critical fire danger conditions place tremendous stress on forests and other vegetative communities. Such periods typically encompass a broad spatial scale with some degree of local variability. Assessing this local variability can play a key role in understanding the extent/severity of the fire conditions; however, we seldom have spatially explicit measurements of environmental conditions in forests. While remote sensing is one means of assessing the spatial extent and variability of environmental conditions, it is often difficult to put such measurements in a historical context as the period of record is generally rather short.

Many places around the world have long time series of routine weather observations that can be useful in evaluating historic fire danger conditions. This network of observing stations is very irregular with high concentrations of weather stations near heavily populated areas and relatively few in remote forested areas. The key to using this irregular network of observations to examine spatial patterns in fire danger is the method of spatial interpolation. This study examines the performance of different spatial interpolation methods in producing gridded fields of temperature, relative humidity, precipitation and wind speed. The interpolation methodologies used include inverse distance weighted average, multivariate linear regression, krigging and artificial neural networks based interpolation.