Influence of Atmospheric Correction Models on the Discriminatrion of Crops using Airborne Hyperspectral Imagery

Crop discrimination is a general application of hyperspectral remote sensing with supervised learning-based classification being the approach of choice. The quality of classification is substantially influenced by the atmospheric impacts on the radiance recorded. The primary goal of atmospheric correction is to extract from remotely sensed data of the target's surface reflectance or emittance values. We have assessed the influence of different atmospheric correction schemes on the quality of crop discrimination using airborne hyperspectral imagery acquired on a site in the Anand district of Gujarat. The variation of the classification performance as obtained from different machine learning algorithms has been assessed for quantifying the influence of different perspectives of atmospheric corrections on the hyperspectral imagery for crop classification. Results indicate a strong dependence on the classification of crops, especially with the imagery corrected using statistically dominant atmospheric correction methods. Despite marginal differences across the various radiative transfer modelling (RTM)-based models, a general trend of higher-level classification performance in crop classification indicates a uniform distribution of load factors and image quality metrics. Accuracy estimates from the imagery corrected using a semi-empirical model (e.g., QUAC) are only marginally lesser than the radiative transfer models while offering great flexibility in terms of computational requirements and scope for automated image correction. So, at times, an empirical model can also be a better method for atmospheric correction.