Land Cover Classification and Machine Learning Techniques with AVIRIS and WorldView Data in the Arctic Coastal Plain

Abstract:

The Arctic Tundra is characterized by low growing vegetation and small water bodies. It is vulnerable to thawing and erosion which can impact ecosystem services, water resources, and habitat. Machine learning algorithms are advantageous for collecting land cover data using image classification capabilities. The Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) has 224 spectral bands and research applications that include environmental change, land cover, and hydrology. While AVIRIS imagery has a 5 meter spatial resolution, pansharpened WorldView imagery has a 0.5 meter spatial resolution. The localized hyperspectral AVIRIS data will be translated as a training dataset for land cover using pansharpened WorldView data. The training data is applied to machine learning algorithms such as random forest analysis to evaluate the translation from localized hyperspectral data to a 0.5 meter spatial resolution.