Comparison of machine and deep learning algorithms using Google Earth Engine and Python for land classifications

Abstract

Classifying land use and land cover (LULC) is essential for various environmental monitoring and geospatial analysis applications. This research focuses on land classification in District Sukkur, Pakistan, employing the comparison between machine and deep learning models. Three satellite indices, namely, NDVI, MNDWI, and NDBI, were derived from Landsat-8 data and utilized to classify four primary categories: Built-up Area, Water Bodies, Barren Land, and Vegetation. The main objective of this study is to evaluate and compare the effectiveness of comparison of machine and deep learning models. The machine learning models including Random Forest achieved an overall accuracy of 91.3% and a Kappa coefficient of 0.90. It accurately classified 2.7% of the area as Built-up Area, 1.9% as Water Bodies, 54.8% as Barren Land, and 40.4% as Vegetation. While slightly less accurate, Decision Tree model provided reliable classifications. Deep learning models showed significant accuracy, of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). The CNN model achieved an impressive overall accuracy of 97.3%, excelling in classifying Water Bodies with User and Producer Accuracy exceeding 99%. The RNN model, with an overall accuracy of 96.2%, demonstrated strong performance in categorizing Vegetation. These findings offer valuable insights into the potential applications of machine learning and deep learning models for perfect land classifications, with implications for environmental monitoring management and geospatial analysis. The rigorous validation and comparative analysis of these models contribute to advancing remote sensing techniques and their utilization in land classification tasks. This research presents a significant contribution to the field and underscores the importance of precise land classification in the context of sustainable land management and environmental conservation.