Using Multitemporal and Multisensoral Images for Land Cover Interpretation with Random Forest Algorithm in the Prykarpattya Region of Ukraine

Autor: Mykola Korol, Olena Vovk, Olha Tokar, Serhii Havryliuk, Lubov Kolyasa
Rok vydání: 2018
Předmět:
Zdroj: Advances in Intelligent Systems and Computing ISBN: 9783030010683
DOI: 10.1007/978-3-030-01069-0_5
Popis: Random Forest (RF) classification algorithm was used for investigation of land cover dynamics in the Prykarpattya region of Ukraine. This approach was applied for two types of images – Landsat and Sentinel-2 with different spatial resolution and obtained in different time (multitemporal). For correct comparison of classifications resulting from the Landsat and Sentinel-2 images the same ground truth data were used for forming the signatures for image interpretations. All classifications were done using a script developed for R software involving a special library realizing the Random Forest algorithm for image interpretation. For land cover investigation main land cover types (coniferous forest, deciduous forest, water, urban territory, grasslands, and additionally areas under clouds) were interpreted using different multitemporal images with different (similar) spatial resolutions. For comparison of the results, all images were formed with similar spatial resolution. Accuracy of classification was estimated using a few indexes, including OOB (Out-of-Bag) and Kappa. Using obtained land cover classifications, thematic maps of land cover were formed, and land cover change was analyzed.
Databáze: OpenAIRE