Investigation of the April 2000 forest fire in Kangwon Province, Korea, using RADARSAT data

A series of forest fires in Kangwon Province in April 2000 was one of the most serious ones in Korea in recent years. A set of multi-temporal RADARSAT data was used to identify the burned area from the undamaged background forest. First, the backscattering coefficient (σ⁰) of SAR image was computed and examined. Depending on the slope of the scattering surface, σ⁰ changes slightly (1~3 dB) in burnt areas, and there are consistent trends in the front- and back-slopes. Secondly, multitemporal texture analysis of the RADARSAT data was carried out using Markov random field models. The lognormal multiplicative autoregressive random field (MAR) model was employed to determine the textural attributes of the SAR image. The contextual fusion of multi-temporal changes in textures was evaluated using the fuzzy concept. The performance of texture analysis alone could not give a clear result, but the contextual fusion clarified the severely burned areas