Radiometric normalization for change detection in peatlands: a modified temporal invariant cluster approach

Abstract

Radiometric normalization is a vital stage in any change detection study due to the complex interactions of radiance and irradiance between the Earth's surface and atmosphere. Compensation for variables such as sun's angle, surface profile, atmospheric conditions, and sensor calibration coefficients are essential in achieving a radiometrically stable data base of multi-temporal, multi-spectral imagery for a change detection study. In this study, five Landsat Enhanced Thematic Mapper Plus (ETM+) images taken over the east coast of Ireland in 2001 were geometrically corrected and topographically normalized for further processing and analysis. Assessment of various vegetation indices showed that the enhanced vegetation index 2 (EVI2) gave the highest accuracy in identifying the various vegetation types and habitats in the Wicklow Mountains National Park. The initial analysis of radiometric normalization with temporal invariant clusters (TICs) gave poor results due to the spectral heterogeneity of urban pixels within each image. A revised TIC subset normalized method was developed using regional growth parameters in urban environments to limit the spatial and spectral extent of pixels used in the TIC scene normalization process. Correlation analysis between the TIC-subset-normalized ETM+ data and Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) absolute corrected data produced coefficient of determination (R²) values between 0.88 and 0.98. Such results demonstrated the robustness of the TIC subset normalization procedure when correcting for atmospheric variability between images while maintaining spectral integrity. Statistical analysis on master slave and TIC-subset-normalized slave data using cumulative distribution curves derived from image histograms showed an 86.93% reduction in the maximum difference between master and slave data due to the TIC subset normalization process. This procedure of radiometric normalization is suitable in landscapes with a low density of spectrally stable targets.

Acknowledgements

The authors wish to thank the Environmental Protection Agency of Ireland (EPA) for their financial support under the STRIVE fellowship. The authors also wish to thank the US Geological Survey for providing access to the ETM+ data, as well as the NPWS for supplying habitat maps of the Wicklow Mountains National Park.