Assessment of uncertainty to estimate burned area from different spatial resolution satellite imagery using a neuro-fuzzy classifier
Abstract
Remote sensing data from different instruments (AVHRR, MODIS, LANDSAT) and spatial resolutions (30m, 500m, 1Km, 4Km) were used to assess the impact of the spatial resolution in burned area mapping. Uncertainty was estimated with a neuro-fuzzy classifier. High resolution remote sensing images (Landsat5/TM) and ground data were used initially to select diverse scenes affected by the fire. The study region was located in the north-west region of the Iberian Peninsula, where several fires occurred in August 2006. A pixel approach neuro-fuzzy classifier was designed to identify burned areas on those high resolution scenes but only using those bands in similar spectral region, comparable between sensors. The classifier was applied to all of the images in order to compute the burned area uncertainty driven by the image resolution. Results show the inverse relationship between the spatial resolution of the images and the burned areas in terms of uncertainty. Burned pixel neighbourhood conditions could be used by the classifier in order to improve uncertainty burned area estimations.
- Publication:
-
37th COSPAR Scientific Assembly
- Pub Date:
- 2008
- Bibcode:
- 2008cosp...37..976R