Machine Learning Techniques of Land Use and Land Cover Classification Using Google Earth Engine: A Comparative Study

To manage assets sustainably and to comprehend how the landscape is changing due to climate change and the passage of time, it is essential to classify LULC accurately. Reliable classification techniques and excellent datasets are necessary for accurate and effective LULC classification. The performance of different satellite data and classification method combinations must be objectively evaluated in order to choose the most effective method for LULC classification given the increasing accessibility of satellite data, geospatial analysis tools, and classification algorithms. This study uses a case study for the Machilipatnam region utilizing two distinct years that 2017 and 2022 to evaluate the LULC classification performance of the widely used platform (i.e., Google Earth Engine) with multiple satellite datasets (i.e., Landsat, Sentinel). This will make it simpler to select the most efficient LULC categorization technique. In contrast, Google Earth Engine creates LULC maps for the various time periods using SVM, classification and regression tree (CART), and Naive Bayes as its three classifiers. The total accuracy of each classifier is computed during the investigation for all conceivable pairings of satellite data, classification platforms, and procedures to determine how well the classifiers are performing. Findings indicate that the SVM classifier is superior to the other two classifiers for the two years 2017 and 2022. Particularly, in the year 2017, the SVM exhibits an overall accuracy of 0.85, CART exhibits an accuracy of 0.73, and Naive Bayes exhibits an accuracy of 0.57. Similarly, in the year 2022, SVM classifier has the highest accuracy (0.78), followed by CART with 0.71 and Naive Bayes with 0.57.

I. Introduction

This chapter provides a summary of the information proposed objectives, background, and applications. Additionally, it describes the circumstances in which it is relevant to, including the requirements, scope and constraints.