MULTI-SENSOR TECHNIQUES TO FLOOD MAPPING

Abstract

The aim of this article is to map the flooding area in the Tarauacá city of Acre State (Amazon basin, Brazil) using multi-sensor techniques. Floods are the most common disaster in the Amazon, even though flood and risk mapping has only recently been delimited by governmental issues. The methods to flood mapping included Synthetic Aperture Radar (SAR) (Sentinel-1/S1) and optical sensor (Sentinel-2/S2) data, separately, and in a fusion approach. The flood extend was measured on Sentinel-1 VV, on NDWI Sentinel-2, and on unsupervised classification from S1 and S2 fusion and classes: soil exposed, urban, vegetation, shadow, water and clouds. The resulting total areas vary by 8.23 km² (S1), 7.86 km² (NDWI-S2), and 11.87 km² (multi-sensor fusion). The fusion S1S2 results were validated from the calculation of global accuracy (70%), errors of omission (soil exposed 0; urban 31,25; vegetation 87,5; shadow, water and clouds 0), commission errors (soil exposed 50; urban 0; vegetation 75; shadow 6,25, water 6,25 and clouds 87,5), and the kappa index (0.59). Using multi-sensors is an alternative to calculate flood extension and can aid in mapping hydrological hazards in Amazon cities.