Integration Of UAV-LiDAR Data And Satellite Imagery For Multi-scale Mangrove Observation

Mangroves are evergreen woody plants composed of trees and shrubs and grow in the intertidal region of the tropical and subtropical coasts with low wave energy.Mangrove wetlands are among the most productive and carbon-dense ecosystems in the world.Mangrove wetlands play an important role in windbreaks,shoreline stabilization,purification of coastal water quality,and providing breeding and nursing grounds for marine and pelagic species as well as aesthetic value.Despite mangroves provide a wide range of ecosystem services and goods,global mangrove forests declined rapidly in last century,and the mangroves in China were no exception.Currently,global mangroves are still reduced rapidly.Therefore,it is crucial to monitor mangrove forests and evaluate their biophysical status in response to mangrove forest ecosystem degradation.Now most mangrove observation studies focused on a specific scale or a specific extent.There is no systematic or normal multi-scale mangrove observation method.Accurately mapping mangrove species is ongoing challenge in remote sensing.In addition,limited by the difficulty in acquiring enough field samples?mangroves grow in harsh environments that are hard to access?,most mangrove estimation studies are conducted over small areas.Against this background,based on the synergistic use of UAV-LiDAR data and satellite remote sensing imagery,this study proposed a new multi-scale observation method for mangroves.To examine this method,this study used Hainan Island that is the most structurally complex mangrove forest region in China as study area.On Hainan Island,an observation system was established consisting of regional scale,reserve scale and local scale,in which specific observation methods were developed.The specific observation objects are mangrove extent,spatial distribution of tree species,tree height and aboveground biomass?AGB?.The regional scale denotes a large region,such as a province or an island;the reserve scale denotes a national or provincial nature reserve;the local scale denotes a sub area of a reserve or a small area of mangroves.The main results indicated that:?1?Based on globally-and freely-available Sentinel-2 imagery,the proposed approach that integrates object-based and pixel-based imagery analysis methods successfully extracted regional mangroves.The overall classification accuracy of mangroves and non-mangroves in Hainan Island was 98.00%with Kappa coefficient of0.96.Hainan Island had 3697.02 ha of mangrove forests in 2018,which were located in nine coastal cities.The mangroves in Haikou city and Wenchang city situated on the northeast of Hainan Island accounted for 68.82%of the total area of Hainan Island's mangroves.There were three coastal cities?Changjiang,Ledong,and Wanning?that do not have obvious mangroves.?2?Base on Sentinel-2 imagery,the object-based imagery analysis method coupled with machine learning algorithms can precisely discriminated mangrove species overall.When using the Dongzhai Harbor National Nature Reserve as study area,the overall classification accuracies of Sentinel-2 and Landsat 8 data were 70.95%and 68.57%,individually.While,the overall accuracy of commercial Pléiades-1 imagery with spatial resolution of 0.5 m was 78.57%.Although the overall accuracy of Sentinel-2 was moderate,the majority of point/line/polygon-shaped mangrove species communities could be well demarcated and mapped.?3?In local scale,the combined use of UAV-LiDAR and World View-2 successfully discriminated mangrove species at individual tree level.The proposed individual tree-based inference method successfully estimated mangrove AGB at individual tree level,which is the first time that mangrove AGB are predicted at individual tree level.When using the high structurally complex mangrove forests in the core area of the Qinglan Harbor Provincial Nature Reserve as study area,the overall classification accuracy of six mangrove species at individual tree level was 86.08%with Kappa coefficient of 0.83.For AGB estimation,the R² of the individual tree-based inference method was 0.49 with an RMSE of 48.42 Mg ha^-1.The AGB map at individual tree level can portray more detail of spatial distribution of mangrove AGB,which is useful for elaborate and differential mangrove management.?4?The individual tree detection results showed that the total accuracy of tree spatial position was 87.43%and the total accuracy of tree number was 51.11%.The individual tree detection accuracy of B.sexangula was the highest among six species,while E.agallocha,L.racemosa,and Sonneratia spp.had obviously segmentation bias.When tree diameter at breast height?DBH?was lower than 10 cm,under-segmentation and high deviation appeared more often.While,when DBH was higher than 10 cm,the individual tree segmentation accuracy was more accurate.Furthermore,there was a negative correlation between the segmentation precision and community complexity,that was,the more complex the community,the lower the segmentation precision.?5?Based on a newly defined point-line-polygon framework,with field plots,UAV-LiDAR strip data and Sentinel-2 imagery,the proposed upscaling method successfully estimated mangrove AGB and tree height on Hainan Island.The results indicated that in the regional scale?Hainan Island?,the newly constructed G?LiDAR?S2_AGB model performed better than the traditional G?S2_AGB model,which directly relates field plots to Sentinel-2 data(R²=0.62>0.52,RMSE=50.36 Mg ha^-1<56.63 Mg ha^-1).The LiDAR?S2_H model for tree height estimation also performed well with R² of 0.67 and RMSE of 1.90 m.The average tree height of the island-wide mangroves on Hainan Island was 6.99 m±2.14 m,the total AGB was 474,199.31 Mg,and the AGB density was 128.27Mg ha^-1±45.87 Mg ha^-1.The AGB hot spots and high mangrove were mainly distributed in Qinglan Harbor and the south of Dongzhai Harbor.The mean height and mean AGB of the mangrove forests on Hainan Island were higher than those of the other mangroves in China and lower than the global mean height and AGB of mangrove forests.Through a trend extrapolation method,this study inferred that the G?LiDAR?S2_AGB model could decrease the number of field samples required by approximately 39%in comparison with those required by the G?S2_AGB model.The result implied that for AGB estimation of a mangrove forest,only approximately 1%of the mangrove area should be sampled in collecting representative and typical mangrove LiDAR point clouds.With further decline in the cost of LiDAR sensors and the simplification of UAV operation,the proposed method may pave the way for large-scale mangrove biophysical parameter estimation.?6?Both the introduced RFE and NRFE algorithms can effectively filter features from high-dimensional and high-correlated feature spaces.The optimal features screened by the two algorithms resulted slightly different modeling accuracy?<2%?.For the species classification,tree height and AGB estimation of mangroves,the importance of spectral bands was:red-edge>shortwave infrared>near infrared>other visible light bands.The texture of high spatial resolution imagery was useful for mangrove species classification,while the texture features of moderate spatial resolution images were not useful.The LiDAR metrics describing the canopy's thickness and its top and bottom characteristics were the most important variables for mangrove AGB estimation.?7?This study successfully builds a database for multi-scale mangrove observation on Hainan Island,and defines the required data,naming rules,spatial reference and data format for observations at each scale.The data mainly includes multi-source remote sensing data,field measurement data and related geospatial data.The main innovations of this study are as follows:?1?This paper combines UAV-LiDAR and satellite remote sensing data to successfully construct a multi-scale observation system of mangroves,forming three-scale observation methods at the regional,protected area,and local levels.?2?A new"point-line-polygon"framework is defined.Based on this framework,a novel upscaling method for estimating structure and functional parameters of mangroves is proposed with field plots,UAV-LiDAR strip data,and Sentinel-2 imagery.The efficiency of the proposed method in reducing the field sampling intensity was also quantitatively evaluated.The first mangrove height and AGB maps of Hainan Island at a spatial resolution of 10 m are produced.?3?A novel method for mangrove AGB estimation at individual tree level is proposed?named individual tree-based inference method?based on the combination of World View-2,UAV-LiDAR,and field survey data.?4?Using the random forest-based recursive feature elimination methods,the important multispectral and LiDAR metrics are selected from the high-dimensional and high-correlated feature space.The inherent differences of mangrove species and how they display in Sentinel-2 and UAV-LiDAR metrics are analyzed.