Exploring spatially varying relationships between forest fire and environmental factors at different quantile levels

The effect of driving factors on forest fire occurrence at various risk levels beyond average fire risk is of great interest to forest fire managers in practice. Using forest fire occurrence data collected in Fujian province, China, global quantile regression (QR) and geographically weighted quantile regression (GWQR) were applied to investigate the spatially varying relationships between forest fire and environmental factors at different quantiles (e.g. 0.50, 0.75, 0.90 and 0.99) of fire occurrence. These results indicated that: (1) at each quantile, the regression coefficients of both global QR and GWQR were negative for elevation, slope and Normalised Difference Vegetation Index, and positive for settlement density, national road density and grass cover; (2) low number of pixels with high fire occurrence in space might dramatically affect the analysis and modelling of the relationship between fire occurrence and a specific environmental factor; (3) according to GWQR, the relationships between forest fire and environmental factors significantly varied across the study area at different quantiles of fire occurrence; and (4) the GWQR models performed better in model fitting and prediction than the QR models at all quantiles. Therefore, the GWQR models could help decision makers to better plan for forest fire management and prevention strategies.