Using Spatial Statistics Tools on Remote-Sensing Data to Identify Fire Regime Linked with Savanna Vegetation Degradation

Introduction

Fire is a common management tool in African tropical and sub-tropical savanna. It is acknowledged to be a factor for explaining the disturbance of vegetation dynamics. Depending on the fire regime more specifically the occurrence and the seasonality of the fires, the vegetation cover of the savanna changes in terms of biomass quantity and structure (Eva & Lambin, 2000). But these fire effects vary widely according to local conditions because of many interactions with other environmental factors such as the rain regime, the stockbreeding pressure, the soil properties or the human activities (Borak et al., 2000). As a consequence, there is no general agreement on the conditions under which fires should be used to ensure the sustainability of the savanna (Jeltsh et al., 2000). “The problem of savanna tree-shrub-grass equilibrium appears to be extremely varied: there is not one savanna problem, but several savanna problems” (Schnell, 1971). Based on this assumption, clarifying and quantifying the importance and the role of fire on savanna vegetation cover taking into account the specificities of the local conditions is a major issue to provide accurate information to local managers while they build their management strategy.

Given the elements presented above, to analyse space-time variations of the plant cover dynamics of savanna together with space-time variations of the ways of using fires is the most appropriate approach (Eva & Lambin, 2000; Ehrlich et al., 1997; Bucini & Lambin, 2002; Devineau et al., 2010; Jacquin, Chéret, Sheeren, & Goulard, 2010). In the present work, issues on the importance and the role of the fire factor in the changes for a given savanna ecosystem are revisited at landscape scale using time-series of medium spatial resolution images to measure vegetation activity trend and fire regime over almost ten years. This is addressed through the implementation of spatial statistical tools. Our aim is not to produce strong thematic conclusion about fire regime and landscape management, but to suggest methodological tools and to test them on a real dataset.

From a methodological point of view, this study presents two interests compared to previous works (Eva & Lambin, 2000; Bucini & Lambin, 2002).

First, we used a time-series of images as input data. This kind of data enables to use temporal decomposition methods which are recognised to be more adapted than traditional change analysis technics to detect abrupt as well as subtle changes on vegetation cover (Verbesselt et al., 2009). These time series allows producing information on fire that combines seasonality and occurrence; occurrence can be including into the analysis as it is recognized to play a significant role in the explanation of vegetation changes. Using these time series, fire regime and vegetation changes are calculated at the same spatial resolution and in the same timing as recommended by Bucini and Lambin (2002).

Second, the relationship between fire regime and vegetation changes is usually studied on homogeneous environmental conditions. These are defined through a stratification technique. Quality of the data used for the stratification highly influences conclusions (Buicini & Lambin, 2002; Jacquin, Chéret, Sheeren, & Goulard, 2010). In this paper, we investigated how to examine the relationship between fire regime and vegetation changes with and without data on environmental factors. To this end, we compared two statistical approaches in which we directly or indirectly consider the existence of a spatial effect due to environmental factors.