Faculty of Mathematics and Natural Sciences - Earth Observation Lab

Faculty of Mathematics and Natural Sciences | Geography Department | Earth Observation Lab | News | Archive | New paper in Remote Sensing by doctoral researcher Akpona Okujeni

New paper in Remote Sensing by doctoral researcher Akpona Okujeni

We evaluate the use of five different advanced regression techniques for quantitative mapping of urban land cover from HyMap data at 3.6 and 9 m...

A comparison of advanced regression algorithms for quantifying urban land cover

We evaluate the use of five different advanced regression techniques for quantitative mapping of urban land cover from HyMap data at 3.6 and 9 m. We adopt the concept of synthetically mixed training data (Remote Sensing of Environment, Volume 137, Oct. 2013), which helps to overcome the lack of quantitative training information required for empirical modeling. This paper presents a substantial follow-up study that allows readers to derive valuable knowledge on how to use different machine learning regression techniques. Our paper includes a comprehensive analysis and a thorough discussion of the findings. We highlight the excellent suitability of kernel-based SVR and KRR in combination with synthetically mixed training for quantifying thematically meaningful and spectrally challenging urban categories at multiple spatial scales.

We carried out the regression analyses with the following Toolboxes:

EnMAP-Box - developed at the Geomatics Lab, HU zu Berlin

ARTMO-Toolbox - developed at the Image Processing Laboratory, Universitat de Valènci