Development of image classification toolkit for remote sensing in Google Earth Engine

Abstract

Google Earth Engines provides accessibility to plentiful databases for image acquisition and analysis, while also providing tools and a user-friendly API to accomplish the tasks at hand. Utilizing Google Earth Engine’s analysis and tool creation capability the aim of the study was to create a tool to monitor landcover change within the Chobe District in Botswana that can be easily executable from users with minimal expertise in the remote sensing field. The three machine learning techniques used are Naïve Bayes, Support Vector Machine, and Random Forest which are paired alongside pooled sampling to create the tool. Z-Test and McNemar’s test are two quantitative methods used to compare each classifier’s performance from the resulting overall accuracy and Kappa value which is also calculated from within Google Earth Engine. Qualitative analysis methods are then used to compare the results of the best performing classifier from Google Earth Engine and results from a similar study which creates a landcover classifier for the same region on ArcGIS.