Mangrove Extent Mapping Using Machine Learning and a Fusion of Optical and Radar Images
Keywords: Algorithm; Decision Trees; Cloud Cover; NDVI; Radar; Tropical Coastal Zone
Abstract: Mangrove complexity and cloud cover effects, among others, make it difficult to classify mangrove forests in tropical coastal zones using simply passive remote sensing techniques. The method described in this paper combines optical and radar data to overcome the difficulties of identifying mangrove stands in cloudy areas. Google Earth Engine geospatial processing platform was used to extract multiple scenes of Landsat surface reflectance Tier1 and synthetic aperture radar. The images were enhanced by creating a feature that removes clouds from the optical data and using speckle filters to remove noise from the radar data. The random forest algorithm was used for mangrove classification. Classification was evaluated using three scenarios: classification of optical data only, classification of radar data only, and combination of optical and radar data. The scenario that uses both optical and radar data fared better, according to our findings. For the classification of optical data only, radar data only, and a combination of optical and radar data, the overall accuracy for 2019 was 98.9 %, 84.6 %, and 99.1%, respectively. This research has shown that it is possible to map mangrove correctly, enabling on-site conservation practices in the climate change environment.
Submission Category: Machine learning algorithms
Submission Number: 34
Loading