Remote sensing identification of marine floating raft aquaculture area based on sentinel-2A and DEM data

Abstract

Marine floating raft aquaculture forms an integral component of the monitoring of coastal marine environments. It is essential to accurately obtain the spatial distribution of marine floating raft aquaculture to gain the fullest understanding of the development of marine fishery production, optimization of the spatial layout of aquaculture, and protection of the marine environment. The Sentinel-2 Multispectral Instrument (MSI) is used to acquire optical imagery at a high spatial and temporal resolution, sampling 13 spectral bands in the visible, near-infrared, and short-wave infrared parts of the spectrum. This research reports how a decision-tree-based procedure was developed to map marine floating raft aquaculture using Sentinel-2A MSI imagery and DEM (Digital Elevation Model) data. Three indices and spectral features were used in this algorithm to differentiate marine floating rafts from other land-cover and land-use types in Fangchenggang City, China. These included the Differential Ratio Floating Raft Index (DRRI), newly proposed in the paper, the Normalized Difference Vegetation Index (NDVI), and visible reflectance. Additionally, a comparison was made between the decision tree classification method (DT) and the random forest (RF) and support vector machine (SVM) methods. The results demonstrate that these three methods can obtain raft information with high accuracy. Finally, the classification results were merged into aquaculture rafts and non-aquaculture rafts. The overall accuracy for DT was 98.20% and 1.28 and 4.99 percentage points higher than RF and SVM, respectively. The user accuracy for marine floating rafts for DT (98.25%) was also markedly higher than that of RF and SVM methods (93.97% and 86.50%, respectively). The producer accuracy for marine floating rafts through the DT method was 98.17%, 0.81 percent lower than that of RF, and 1.03 percent lower than that of SVM. The decision-tree method does not assume strict data distribution parameters, optimization of the application of multispectral imagery and elevation data becomes possible, and combing with the DRRI index, then results in higher classification accuracies of marine floating rafts. When using multi-source data of different types and distributions to map marine floating rafts, a decision-tree method, therefore, appears to be superior to RF and SVM classifiers.