Open AccessMachine Learning Techniques for Land Use/Land Cover Classification of Medium Resolution Optical Satellite Imagery Focusing on Temporary Inundated Areas
Author(s) -
Boudewijn van Leeuwen,
Zalán Tobak,
Ferenc Kovács
Publication year - 2020
Publication title -
journal of environmental geography
Language(s) - English
Resource type - Journals
eISSN - 2060-467X
pISSN - 2060-3274
DOI - 10.2478/jengeo-2020-0005
Subject(s) - land cover , artificial neural network , remote sensing , satellite , satellite imagery , multispectral image , computer science , land use , support vector machine , thematic map , data set , image resolution , implementation , random forest , cover (algebra) , environmental science , artificial intelligence , cartography , geography , mechanical engineering , civil engineering , aerospace engineering , engineering , programming language
Classification of multispectral optical satellite data using machine learning techniques to derive land use/land cover thematic data is important for many applications. Comparing the latest algorithms, our research aims to determine the best option to classify land use/land cover with special focus on temporary inundated land in a flat area in the south of Hungary. These inundations disrupt agricultural practices and can cause large financial loss. Sentinel 2 data with a high temporal and medium spatial resolution is classified using open source implementations of a random forest, support vector machine and an artificial neural network. Each classification model is applied to the same data set and the results are compared qualitatively and quantitatively. The accuracy of the results is high for all methods and does not show large overall differences. A quantitative spatial comparison demonstrates that the neural network gives the best results, but that all models are strongly influenced by atmospheric disturbances in the image.