Scholarlink

<p>Like many other countries the world over, Zimbabwe’s towns and cities are experiencing a substantial changes in land use/land cover. These changes are associated with significant negative environmental, social and economic effects, and are attributed to many variables, which then demand the use of accurate land use/land cover classification techniques. This study compared the effectiveness of four supervised machine learning (ML) algorithms, namely; Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB), and Classification and Regression Trees, for land use/land cover classification in Google Earth Engine (GEE). Landsat satellite images for 2000, 2005, 2010, 2015 and 2020 were used to generate land use/land cover maps for Bulawayo, the second largest city in Zimbabwe, using the GEE platform. The effectiveness of the machine learning algorithms was evaluated using F1-score, kappa coefficient, and overall accuracy. The findings of the study revealed that there was a significant increase in built-up areas between 2000 and 2020, likely due to population growth. Conversely, the areas for water bodies, bare land and vegetation decreased during the study period, likely converted for development. The results obtained from F1-scores showed that Random Forest is the most effective algorithm for classifying built-up, water bodies, bare land, and vegetation classes. Support Vector Machine generally achieved good performance, while Classification and Regression Trees and Naive Bayes exhibited lower performance. The overall accuracy and kappa coefficient were used to validate the land use/land cover maps generated by each algorithm. All four algorithms achieved satisfactory accuracy, with Random Forest demonstrating the highest overall accuracy (91.59%) and kappa coefficient (0.8845). This study demonstrated the great potential of machine learning algorithms, particularly Random Forest, for accurate land use/land cover mapping in Bulawayo, Zimbabwe. The findings of this study provide valuable insights into land use changes and are capable of informing urban planning and environmental management strategies in the city.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Comparative Analysis of Supervised Machine Learning Algorithms for Land Use and Land Cover Classification in Google Earth Engine

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Norman Karimazondo,
  • Narissara Nuthammachot,
  • Warangkana Jutidamrongphan,
  • Luke Makarichi,
  • Kingsley Okpara

摘要

Like many other countries the world over, Zimbabwe’s towns and cities are experiencing a substantial changes in land use/land cover. These changes are associated with significant negative environmental, social and economic effects, and are attributed to many variables, which then demand the use of accurate land use/land cover classification techniques. This study compared the effectiveness of four supervised machine learning (ML) algorithms, namely; Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB), and Classification and Regression Trees, for land use/land cover classification in Google Earth Engine (GEE). Landsat satellite images for 2000, 2005, 2010, 2015 and 2020 were used to generate land use/land cover maps for Bulawayo, the second largest city in Zimbabwe, using the GEE platform. The effectiveness of the machine learning algorithms was evaluated using F1-score, kappa coefficient, and overall accuracy. The findings of the study revealed that there was a significant increase in built-up areas between 2000 and 2020, likely due to population growth. Conversely, the areas for water bodies, bare land and vegetation decreased during the study period, likely converted for development. The results obtained from F1-scores showed that Random Forest is the most effective algorithm for classifying built-up, water bodies, bare land, and vegetation classes. Support Vector Machine generally achieved good performance, while Classification and Regression Trees and Naive Bayes exhibited lower performance. The overall accuracy and kappa coefficient were used to validate the land use/land cover maps generated by each algorithm. All four algorithms achieved satisfactory accuracy, with Random Forest demonstrating the highest overall accuracy (91.59%) and kappa coefficient (0.8845). This study demonstrated the great potential of machine learning algorithms, particularly Random Forest, for accurate land use/land cover mapping in Bulawayo, Zimbabwe. The findings of this study provide valuable insights into land use changes and are capable of informing urban planning and environmental management strategies in the city.