Extraction Of Spatial Distribution Information And Estimation Of Aboveground Biomass Of Mangrove Forests Based On Multi-source Remote Sensing Data

Mangrove forests,as important coastal wetlands,play a crucial role in maintaining ecological balance of coastal zones and regulating global carbon cycle.However,in recent years,mangrove ecosystems have been one of most threatened ecosystems in the world due to disturbances of human activities and climate changes.Mangrove forests in China are faced with the situation including area loss,habitat quality decline and ecological function and biodiversity degradation,which have given the greater urgency to protection and restoration efforts.As an important manifestation of the functions and value of mangrove ecosystems,information of distribution,species composition and aboveground biomass(AGB)is closely related with biodiversity,carbon cycle and habitat environment.Therefore,accurate acquirement of these information can provide foundational data and decision-making basis for rational protection,management and restoration of mangrove forests.Remote sensing(RS)has served as an accurate and time-and labor-saving tool in mapping and monitoring mangrove forests,primarily because of the logistical and practical difficulties involved in field surveys of the muddy and intertidal environments.Previous studies have extracted spatial distribution information and estimated AGB of mangrove forests successfully.However,there are still some challenges in the following research areas.(1)How to realize the accurate mapping of the mangrove forests extent efficiently.(2)How to classify mangrove species,which shows similar spectral and textural characteristics,using free remote sensing data.(3)UAV-based light detection and ranging(Li DAR),Sentinel-2(S2)and domestic Gaofen-2(GF2)remote sensing data have emerged and developed currently.However,their performance in estimating AGB of mangrove forests needs to be more evaluated.According to the above analysis,taking the core zone of Fujian Zhangjiangkou National Mangrove Nature Reserve(FZNNR)as the studied mangrove site,this article aims to explore the application and feasibility of multi-source remote sensing data in spatial distribution information extraction and above ground biomass estimation of mangrove forests.Firstly,based on Google Earth Engine(GEE)platform,two S2 images in leaf-on and leaf-off seasons of Spartina alterniflora(S.alterniflora)were acquired,and then OTSU algorithm(OA),which could calculate the binary threshold of Normalized Difference Vegetation Index(NDVI)automatically,was applied to delineate mangrove forests extent.Secondly,a dense time series S2 image collection(from 2019 to 2020)in the GEE was used to build NDVI phonological trajectories of three mangrove species,and then penology-based NDVI features were introduced to perform species classification.Thirdly,Li DAR,S2 and GF2 data were employed to derive RS variables.Then two categories of models based on different data,including optical image-based model and integrated Li DAR and optical data model(in detail,including S2 model and GF2 model based on S2 and GF2 data,respectively,as well as Li DAR-S2 model built by integrating Li DAR with S2 data,and Li DAR-GF2 model built by integrating Li DAR with GF2 data),were built with random forest(RF)regression algorithm to estimate mangrove forests AGB.At last,the accuracy of different RS data-derived models were compared,and the importance of RS variables were evaluated.Major conclusions are as follows:(1)Based on GEE platform and S2 imagery,using OA to calculate NDVI threshold automatically can identify and extract mangrove forests extent rapidly and accurately.The overall accuracy of the resultant map was 93.72%,and the mapping accuracy of the mangrove forests category achieved 94.33%.The classification errors mainly came from the mixed pixels.The S2 data have certain advantages,including freely available,short revisit cycle and relative high spatial resolution.Meanwhile,the GEE platform have predominant efficiency in RS data processing and the OA have automatic ability of calculating binary threshold.Therefore,this method have great potential in finer mangrove forests extent mapping on a larger scale.(2)Utilizing dense time series S2 images,combined with phenology characteristics of the mangrove forests in the study area,can realize high-precision mapping of mangrove species.The resultant overall accuracy was 86.20%,which showed a significant increase compared with the result derived from single S2 imagery.This results demonstrate that the phenology-based RS features have great importance in mangrove species classification in the study area.However,phenology of mangroves in different regions is affected by natural and anthropogenic factors.Therefore,the applicability and applicable extent of the phenology-based mangrove species classification method need to be further explored.In the study area,the total area of Kandelia obovata(K.obovata),Aegiceras corniculatum(A.corniculatum)and Avicennia marina(A.marina)were 25.18 hm~2,19.08 hm~2 and 13.29hm~2,respectively.(3)Based on the characteristic variables derived from Li DAR and S2 data,comparing the accuracies of the AGB estimation models built by different data sources,using RF algorithm can achieve accurate estimation of mangrove AGB(with an accuracy of R~2=0.60,RMSE=21.75 t/hm~2).Integrating Li DAR and optical data to estimate mangrove AGB was superior to using only optical RS data,and S2 data was superior to GF2 data.Integration of Li DAR and S2 data can comprehensively reflect the abundant spectral information of vegetation canopy and vertical structure information,which have applicability in mangrove AGB estimation.GF2 data showed low estimation accuracies due to the lack of the important red edge and short wave infrared bands.The total AGB of the mangrove forests in the study area was 5200.67 t,and for K.obovata,A.corniculatum and A.marina it was 2216.50 t,1691.12 t and 1293.05 t,respectively.