Time series of land surface temperature from daily MODIS measurements for the prediction of fire hazard

Abstract

Prolonged heat and absence of rainfall drive vegetation into water stress conditions that lead to an increase of its temperature. Since stressed vegetation is more prone to fire, it is expected that remote sensing mapping of temperature anomalies might be a viable tool to predict fire hazard. The identification of these anomalies requires the prior definition of a reference temperature against which compare actual recorded temperatures. This can be achieved by using long time series of satellite data and the HANTS (Harmonic ANalysis of Time Series) algorithm. The objective of this research was the characterisation of fire hazard using temporal series of land surface temperature (LST) derived from Terra-MODIS measurements. The investigation was based on a sequence of MODIS LST data from 2000 to 2008 in Campania (13595 km2), Italy, and on a data set of 7700 fires recorded in the area in the same period. Missing and/or cloudy LST data were reconstructed by means of the HANTS algorithm applied to annual sequences of daily observations. The coefficients of the Fourier analysis were assessed against spatial patterns of fire occurrence. HANTS algorithm was also used to construct daily reference temperature maps against which to evaluate temperature anomalies and cumulated temperature anomalies. Results show that fires tend to occur in areas characterised by specific values of several Fourier coefficients with high significance, and to avoid the other areas. The amplitude of the second harmonic is the only Fourier coefficient dictating mean fire size. The mean fire size and the proportion of large fires correlate with both daily and cumulated thermal anomalies.