The pyrogeography of sub-Saharan Africa: a study of the spatial non-stationarity of fire–environment relationships using GWR

Aim Analyze the relationship between fire incidence and some environmental factors, exploring the spatial non-stationarity of the phenomenon. Location Sub-Saharan Africa. Methods Geographically weighted regression (GWR) was used to study the relationship between fire incidence and a set of environment covariates, which comprise land cover, anthropogenic and climatic variables, over an eight year period (1981-1983, 1985-1991). GWR was compared to Ordinary Least Squares (OLS), and the hypothesis that GWR represents no improvement over the global model was tested. Local regression coefficients were mapped, interpreted and related with fire incidence. Analysis was carried out using a 0.5x0.5 degree grid. Results GWR revealed local patterns in parameter distributions, and also reduced the spatial autocorrelation of model residuals. All the predictors were non-stationary and in terms of goodness of fit the model replicates the data very well (R2=87%). About 45% of the local models had an R2 above 0.75 and only 26% had an R2 below 0.57 (the global R2). Herbaceous cover has the most significant relationship with fire incidence, with climate variables being more important than anthropogenic ones in explaining the response variability. Both precipitation seasonality and soil water coefficient estimates exhibit locally different signs, which would have been ignored by a global approach. Main conclusions Results highlighted the need to accommodate for the effects of spatial non-stationarity in fire-environment relationship, which may be hidden in a global model. This study provides an improved understand of spatial fire-environment relationships, indirectly contributing towards the reduction of uncertainties in burned area estimates and to make better predictions of broad-scale fire patterns. GWR is shown to be a valuable complement to global spatial analysis methods, revealing details of fire incidence patterns and data properties that global methods average out and fail to detect.

SA Ana;  PEREIRA José M.C.;  CHARLTON Martin E.;  MOTA Bernardo;  BARBOSA FERREIRA Paulo;  FOTHERINGHAM Alexander S.; 

2011-09-15

SPRINGER HEIDELBERG

JRC49092

1435-5930,   

http://www.springer.com/economics/regional,    science/journal/10109,    https://publications.jrc.ec.europa.eu/repository/handle/JRC49092,   

10.1007/s10109-010-0123-7,