Please login first
A Comparison of Supervised Classification Algorithms in Guayaquil Land Use and Land Cover Data: An Evaluation with Landsat and MapBiomas
* 1, 2 , * 3 , * 2, 3, 4
1  Faculty of Mechanical Engineering and Production Sciences, ESPOL Polytechnic University, Campus Gustavo Galindo, Guayaquil, 090902, Ecuador
2  Laboratory of Geoinformation and Remote Sensing, ESPOL Polytechnic University, Campus Gustavo Galindo, Guayaquil, 090902, Ecuador
3  Faculty of Engineering in Earth Sciences, ESPOL Polytechnic University, Campus Gustavo Galindo, Guayaquil, 090902, Ecuador
4  Centro de Investigación y Proyectos Aplicados a las Ciencias de la Tierra, ESPOL Polytechnic University, Campus Gustavo Galindo, Guayaquil, 090902, Ecuador
Academic Editor: Hossein Azadi

Abstract:

Land use and land cover classification (LULC) is a method used for sustainable land management and studying land use change over time, particularly in expanding areas such as the city of Guayaquil, Ecuador. This study compares three supervised classification algorithms—Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN)—applied to the mosaic created from multispectral Landsat-9 images of 2023, obtained through Google Earth Engine (GEE). Validation was carried out by comparing the results with reference data from the MapBiomas Ecuador project, which has shown high reliability in South American contexts. The models were trained using samples obtained through visual interpretation of the Sentinel image mosaic, categorizing four classes (forest, crops, non-vegetated areas, and water). Standard metrics such as the kappa coefficient, overall accuracy, and confusion matrices were used to assess the performance of the algorithms. The results showed that the SVM algorithm performed the best (kappa = 0.91, accuracy = 93%), surpassing RF (kappa = 0.88) and ANN (kappa = 0.86). SVM demonstrated a stronger ability to manage nonlinearly separable classes, and its robustness against band dimensionality explains its superior performance, which aligns with previous findings in remote sensing. SVM showed greater spatial similarity with the patterns identified by MapBiomas, particularly in defining urban areas and water bodies. This research supports using SVM as an effective tool that requires minimal computational resources in equatorial regions, where obtaining images with low cloud covers presents an additional challenge. Furthermore, the importance of validating machine learning models using open and reliable sources is emphasized. Its application is recommended for urban planning studies, coverage change monitoring, and environmental impact assessments.

Keywords: Supervised classification; Landsat; SVM; MapBiomas; LULC; Guayaquil
Top