Modeling Spatiotemporal Wild Fire Data with Support Vector Machines and Artificial Neural Networks

Abstract

Forest fires have not only devastating catastrophic environmental consequences, but they also have serious negative social impact. Exploitation of related spatiotemporal data could potentially lead to the development of reliable models, towards forecasting of the annual burned area. This paper takes advantage of the regression capabilities of modern Soft Computing approaches. More specifically numerous Artificial Neural Networks' and e-Regression Support Vector Machines' models were developed, each one assigned locally to a distinct Greek forest department (GFD). The whole research effort was related to Greek wild fires incidents for the period 1983-1997 and to all of the GFDs. The performance of both methods has proven to be quite reliable in the vast majority of the cases and a comparative analysis was also used to reveal potential advantages or weaknesses. © Springer-Verlag Berlin Heidelberg 2013.