Close Window

Simulating the occurrence of hot pixels along the Amazon forest fringe

Rafaella Silvestrini, Universidade Federal de Minas Gerais, rafaufmg@yahoo.com.br (Presenting)
Britaldo Silveira Soares-Filho, Universidade Federal de Minas Gerais, britaldo@csr.ufmg.br
Hermann Rodrigues, Universidade Federal de Minas Gerais, hermann@csr.ufmg.br
Daniel Curtis Nepstad, Woods Hole Research Center, dnepstad@whrc.org

Forest fire models have become an important tool for assessing the resilience of forest environments in anthropogenic landscapes as well as the feedbacks between deforestation and climate, which eventually may lead the Amazon ecosystem into an irreversible cycle of self-destruction. A basic component of such models involves simulating the occurrence of fire ignition sources due to weather conditions and land-use practices. In this work, we have developed a model that simulates the occurrence of hot pixels, representing sources of fire along the Amazon forest fringe, based on the integration of climate and land-use data. The model was calibrated using NOAA-12 night satellite hot pixel data for 2003 and then validated for the years 2002, 2004 and 2005. Firstly, we estimated the Weights of Evidence of a series of spatial variables - e.g. proximity to roads, towns, and deforested areas, land-use, and other biophysical factors - on the location of hot pixels. The resulting probability map was then combined with a climate risk probability map, derived from monthly VPD (vapor pressure deficit) data by means of logistic regression. Assessment of the integrated fire-prone probability map employing the ROC (Relative Operating Characteristic) method yielded fitness values above 0.8 for all months of 2003, greater than the 0.5 value that would be expected due to chance. The model simulates stochastically the quantity and location of hot pixels, alternating for the dry and wet seasons of the southern hemisphere the coefficients used to average the probability maps and two density distribution functions employed to draw random numbers - the Beta (0.985, 0.1) and Weibull (13,0.6) distributions. Simulated hot pixels matched fairly well the NOAA-12 hot pixel data both in terms of spatial and temporal distributions, showing a maximum yearly frequency deviation of only 15%. As a next step, this model will be coupled to a fire spreading mechanistic model in order to incorporate fire regimes into SimAmazonia-2, a basin-wide simulation model of Amazon landscape dynamics.

Science Theme:  LC (Land Use and Land Cover Change)

Session:  3C: Land Use and Fire

Presentation Type:  Oral (view presentation (4899 KB))

Abstract ID: 46

Close Window