Mapping Mangroves in the Niger Delta using Machine Learning

Abstract:

Mangrove forests are a type of wetland ecosystem found in intertidal zones of tropical and sub-tropical climates. Despite being important ecological and economic assets, mangrove forests have historically been degraded and deforested due to anthropogenic or natural factors. Therefore, continuous monitoring through remote sensing is important to understand where mangrove forests are and if there is forest loss or gain over time. Mangrove mapping is the first step towards forest protection and regeneration. The performance of four machine learning algorithms were assessed in this study for mangrove forest mapping in the Niger Delta, Nigeria in 2018. Freely available Landsat-8 OLI, Sentinel-1 SAR GRD, and Sentinel-2 MSI data was used, and data processing was performed on two cloud computing platforms, Google Earth Engine and Google Colaboratory. Random Forest (RF), Classification and Regression Trees (CART), Support Vector Machine (SVM), and a Deep Neural Network (DNN) built with TensorFlow were all compared for mangrove forest mapping. Results show that the Random Forest, an ensemble decision tree algorithm, showed the highest overall performance at 91%. Time- and cost-efficient algorithms such as RF is recommended for mangrove forest mapping, and can therefore be used by decision makers to determine areas for mangrove protection and regeneration.