| Forests in seasonally dry regions of the Brazilian Amazon have been burning at an increasing rate in recent years, and much of this increase has been due to the disturbance of forest canopy through selective logging. In this thesis, the results of field studies on fire and edge effects are extrapolated to landscape and regional scales through the use of remote sensing data and simulation modeling. A scale relation formalism is used to clarify the spatial and temporal components of this process. Remote sensing data are used to analyze land cover, fire counts and precipitation over a decade, and spatiotemporal characteristics of active fires in 1998. A simulation model of fire growth on gridded artificial landscapes at a 30 m resolution incorporates a model of flammability edge effects implemented with integrative spatial filters. The model reveals that there may be specific thresholds of canopy gap probabilities above which the contiguity of flammable fuel allows for the indefinite spatial spread of any single surface fire in the fuel bed. This condition can be described in terms of a critical percolation. The model is applied to landscapes derived from Landsat data, but the resolution of the 30 m data and model are shown to be problematic. The model resolution is improved to 6 m, and the possible use of IKONOS data is investigated. The fire spread model at the 6 m resolution is used to examine artificial landscapes with gaps of various sizes, from 36 to 4356 m2, and critical percolation thresholds are established. Scale relations and scale dependencies, including the shadowing of smaller gaps, are considered analytically. Finally, canopy gap size distributions are considered in the context of self-organized criticality, and critical percolation thresholds are established for a range of such distributions.; This study indicates that selective logging may dramatically increase the flammability of a forest if it removes more than ten to sixteen percent of the canopy on a random basis, and that certain distributions of gap sizes may attenuate this problem on an areal basis. |
