Prediction of forest fire occurrences depends on spatial scale and inter-annual climatic variations in the Western Ghats of India.

Speaker : Quentin RENARD, International Volunteer, Ecology, IFP

Forest fires represent a recurrent management problem in the Western Ghats (WG) of India. Although most of fires happen during the dry season, information on the spatial distribution of fires is needed to improve fires prevention. Based on geographical coordinates of fire occurrences detected by MODIS satellite during the period 2003-2007 and on different sets of environmental variables (i.e. climate, topography and vegetation), we used the Maxent algorithm to build predictive distribution models of fire occurrences. We also aimed at providing a quantitative understanding of the environmental controls regulating the spatial distribution of forest fires at different spatial scales within the WG. We used Hierarchical Partitioning to assess the independent variables contributions to the goodness-of-fit of the models. Based on these independent contributions, we selected the most significant predictors at each spatial scale in order to build the most parsimonious and accurate predictive models of fire occurrences.

Although areas predicted as highly suitable for forest fires were mainly localized on the eastern slopes of the Ghats (that support dry to moist deciduous forest habitats), spatial predictions and models accuracy differed significantly depending on the sets of environmental variables and the spatial scale considered. Importance of independent variables contribution was also function of spatial scale. Relative contributions were more evenly distributed among variables at small spatial scales than at the scale of the entire WG, where a few variables related to climate were predominant. We also noticed preponderant relative contributions of climatic seasonal variables at all spatial scales as well as an increasing independent contribution of the vegetation layer with decreasing spatial scale.

In order to rationalized the prevention resources to be engaged for forest fires management in the WG, our results suggest that spatial predictions of fire occurrences should be made in two steps. Firstly, large fire-prone areas can be identified at the regional scale in paying a special attention to climatic conditions of the monsoon season prior to the fire season. Such model will provide a good indication of the fuels moisture content during the fire season. Secondly, local models mainly based on the type of vegetation may help identifying the most endangered sites within the fire-prone areas.

Organiser : Department of Ecology, French Institute of Pondicherry.

Venue : Jawaharlal Nehru Conference Hall, French Institute of Pondicherry, 11, Saint Louis Street, Pondicherry - 605 001