Sentinel-1 and -2 time-series data-fusion for olive tree identification in heterogeneous land surfaces using Google Earth Engine


Akçay H. M. , Kaya Ş. , Sertel E. , Algancı U. , Aksoy S.

2nd Intercontinental Geoinformation Days (IGD, 5 - 06 May 2021, pp.156-162

  • Publication Type: Conference Paper / Full Text
  • Page Numbers: pp.156-162

Abstract

Olive, a crucial crop for the economies of Mediterranean countries, is expanded to Aegean, Mediterranean, Marmara, SouthEast and Black Sea regions of Turkey. Identification of olive trees in heterogeneous land surfaces, particularly in mountainous regions is essential for exploitation of un-grafted olive trees. In this study, several samples of olive tree, agriculture, bare-land, urban, forest and sparse vegetation fields located between Bayındır and Tire districts of Izmir province in Turkey, are randomly selected. Independent two sample sets are generated to train the classifier (70%) and for the validation (30%). Several data fusion combinations of time series of Sentinel-1 and Sentinel-2 satellite data with various spectral indices are performed with random forest classifier on Google Earth Engine environment. A new spectral index, named as "DVI Red index (DVIR)" is generated and experimented in the study, as well. Results demonstrated that "Sentinel-1, Sentinel-2 and 10 indices" data fusion performed best overall accuracy (95.5%) as "Sentinel-1 and new ratio index (DVIR)" data fusion performed highest user's accuracy (97.2%) for olive class. Of 10 spectral indices standalone classifications, DVIR ranked the first for overall accuracy (94.8%) and the third for olive class user's accuracy (84.4%).