Published in Scientific Papers. Series E. Land Reclamation, Earth Observation & Surveying, Environmental Engineering, Vol. XIII
Written by Tamara MYSLYVA, Marciano DASAI, Christiaan Max HUISDEN, Petro NADTOCHIY, Yurii BILYAVSKYI
Machine learning (ML) algorithm-based models represent cutting-edge techniques used for mapping, quantifying, and modelling changes in land use and land cover (LULC) over time. In this study, a comparative analysis was conducted on the multilayer perceptron neural network (MLP) and support vector machine classification (SVM) applied to LULC change detection and forecasting within the coastal plain territory of Suriname. Sentinel-2A satellite data covering the period from 2017 to 2022 was utilised, along with additional variables such as the distance from rivers, roads, and administrative cities in each district and slope and digital elevation models in the prediction models. The SVM algorithm based predictive model, incorporating an urbanization transition sub-model, exhibited an impressive accuracy of 83.85%, surpassing the MLP algorithm-based model, which did not exceed 64.63%. Consequently, this model is recommended for generating LULC change prediction maps. These maps can serve as a crucial baseline for the Surinamese government, providing valuable insights for policy development and sustainable land use management.
[Read full article] [Citation]
