Research On The Identification Of Mangrove Forest Based On Multi-source Remote Sensing Data

The mangrove ecosystem dominates the ecological environment of coastal wetlands in the tropical and subtropical regions of the world,providing a variety of ecological and economic systems services.However,mangroves belong to the world's most threatened fragile ecosystems and have fallen dramatically over past half century.Remote sensing is an important tool for monitoring regional mangrove ecosystem distribution,species differentiation,and health status.Mangrove resource survey and dynamic monitoring based on remote sensing technology have become an indispensable means of mangrove scientific research and management.The difficulty in extracting mangrove remote sensing information lies in the low,inundated mangroves and semi-mangroves growing near the land.This paper takes the mangrove ecosystem in the Gaoqiao area of the Zhanjiang Mangrove National Protected Area in Guangdong Province as an example.Optical and radar remote sensing data,analyze the key features of mangrove remote sensing information identification,developing effective methods for mangrove remote sensing information identification,and provide scientific support for mangrove remote sensing dynamic monitoring and scientific management.The main research contents and research results are as follows:(1)Applicability and accuracy analysis of different data sources and different information identification methods in mangrove remote sensing monitoring.The 2017 high-resolution 1(GF-1),Sentinel-2A,and Landsat OLI remote sensing images were selected,and two classification methods of object-oriented classification and support vector machine were used to identify the mangrove distribution information and compare and analyze.The results show that the Sentinel-2A data of the three data sources has the highest recognition accuracy for large plaque mangroves,among which,the normalized difference humidity index,the short-wave infrared texture information-heterogeneity,the red edge band texture information-angle the second-order moment has a higher contribution rate to the recognition of mangrove remote sensing information;the high-resolution GF-1 data corresponds to the small mangrove plaques with better characterization ability.In the classification method,for higher resolution Sentinel-2A and GF-1 data,object-based classification method is more suitable,and medium resolution Landsat data is suitable for pixel-based classification,such as support vector machine classification.The three data sources are consistent with the main body of identification information of the mangrove concentrated distribution area.The inconsistencies are reflected in the mangroves growing on the land and near the coastline.This is also the difficulty in remote sensing identification of mangroves.(2)Based on the two short-wave infrared and red-edge bands of Sentinel-2A data,the characteristic vegetation index of mangrove remote sensing information-9)LSWI was constructed.Using the index to identify the mangrove information in the study area,the vegetation index is superior to the traditionally used normalized difference humidity index(NDMI),especially to improve the recognition accuracy of mangroves growing near the coastline.The overall accuracy of mangrove remote sensing information identified in the subtropical Gaoqiao area was 87.8%,and the Kappa coefficient was 0.755,which was 3.6% higher than the NDMI vegetation index.(3)Integrating Sentinel-1A microwave remote sensing information based on Sentinel-2A optical remote sensing,further improving the discrimination between semi-mangrove and terrestrial vegetation.Studies have shown that because mangroves/semi-mangroves have different leaf structures than terrestrial plants,Sentinel-1A radar data differs in the backscattering coefficients of mangroves and terrestrial plants in the polarization modes of VH and VH/VV.Larger,therefore,based on Sentinel-2A data,the backscatter information of Sentinel-1A VH mode,VH/VV mode and VH+VH/VV mode are integrated to identify mangroves.It is found that under the superimposed polarization combination of Sentinel-2A +VH+VH/VV,mangrove and nearshore vegetation have the highest separability,with user accuracy of 94.9%,mapping accuracy of 92.5%,and total classification accuracy of 93.8.%.Therefore,by combining the radar wave scattering characteristics in the optical data,the discrimination between the near-shore semi-mangroves and terrestrial vegetation can be improved.