Flood aftermath assessment using radiometric normalization and change detection techniques

Floods are among the most dangerous natural hazards related to global warming effects, causing damages and loss of human life all over the world. During the last two decades floods became one of the worst threats involving human environment, both in developing and developed countries. Remote Sensing has proven an effective tool to produce flood maps and assess damages due to flood events, thanks to a wide number of sensors today operative, from radar to optical, from high to low spatial resolution. This paper investigates the influence of the radiometric normalization on flood mapping retrieval. Different radiometric normalization techniques were tested (manual and automatic selection of pseudo-invariant features) and different models were applied to data (linear regression and parametric model based on parabolic functions). When using radiometric normalization, results showed an increment between 2.48% and 3.75% for Overall Accuracy and between 0.051 and 0.075 for the Kappa coefficient. The maximum accuracy was reached with a manual selection of the pseudo-invariant features and a linear regression transformation (Overall Accuracy of 87.98% and a kappa coefficient of 0.760). These results enforce the acknowledgement of radiometric normalization influence in change detection analyses, thus favoring further investigation and study of the pseudo-invariant features selection methods and of the transformation models used.