| Tidal vegetation is an important component of coastal tidal flat.Finely identify the types and aboveground biomass(AGB)of tidal vegetation,mastering the landscape patterns of tidal vegetation under multiple scales,revealing the evolution and development trends of tidal vegetation are the basis of multi-dimensional understanding and evaluation of coastal tidal flat.Due to the particularities in vegetation structure,community structure and growth environment etc.,there are still challenges in the remote sensing detection of tidal vegetation.Therefore,aiming at the typical tidal vegetation types Suaeda salsa,Spartina alterniflora and Phragmites australis in typical coastal areas of China,based on emerging remote sensing data including airborne high-resolution hyperspectral data and small-footprint Light Detection and Ranging(Li DAR)data,and integrating with multi-source data and geoscience knowledge,researches on fine classification method,AGB rapid estimation method and fine-scale dynamic monitoring for typical tidal vegetation were carried out in this study.This research will provide technical and data support for the fine management and protection of coastal tidal flat,and lay a foundation for the follow-up exploration in the temporal and spatial evolution mechanism of tidal vegetation under multi-scale conditions.(1)In the research on the tidal vegetation fine classification method based on high-resolution hyperspectral and small-footprint Li DAR data,a fine classification method for tidal vegetation based on remote sensing was proposed,which combined fine hyperspectral features of high-resolution hyperspectral data and a vegetation structure feature from small-footprint Li DAR data.The feasibility and utilities of high-resolution hyperspectral characteristics,high-resolution multispectral characteristics and a vegetation structure characteristic in the fine classification of complex tidal flat wetland ecosystem were further evaluated.Results showed that fusing fine hyperspectral features and a vegetation structure feature extracted from high-resolution hyperspectral data and small-footprint Li DAR data could achieve the fine classification of tidal vegetation,and the classification accuracy was higher than the conventional classification method based on high-resolution multispectral data.Based only on high-resolution multispectral features of tidal vegetation,it could achieve a relatively good classification level(Kappa 0.70,overall accuracy 84.47%),the introduction of fine hyperspectral features and a vegetation structure feature could improve the classification accuracy,but their effect was not significant,high spatial resolution was the leading factor in tidal vegetation fine classification at pixel scale.Integrating high-resolution hyperspectral features with a vegetation structure feature provided the best classification accuracy(Kappa 0.73,overall accuracy 86.48%).Among the three tidal vegetation types,the classification accuracy of Suaeda salsa was the highest,while those of Spartina alterniflora and Phragmites australis were relatively lower.Besides,the conventional classification accuracy evaluation method was improved by weighting the class area,and the bias of the number of classification verification points to classification accuracy evaluation was eliminated.(2)In the research on the tidal vegetation AGB estimation method based on high-resolution hyperspectral and small-footprint Li DAR data,an AGB estimation method of tidal vegetation was proposed,which combined fine hyperspectral features of high-resolution hyperspectral data and a vegetation structure feature from small-footprint Li DAR data.The feasibility and utilities of high-resolution hyperspectral characteristics,high-resolution multispectral characteristics and a vegetation structure characteristic in tidal vegetation AGB modeling were further evaluated.Results showed that fusing fine hyperspectral features and a vegetation structure feature extracted from high-resolution hyperspectral data and small-footprint Li DAR data could be used to estimate the AGB of tidal vegetation,and the estimates were better than the conventional method based on high-resolution multispectral data.AGB modeling based only on high-resolution hyperspectral features could obtain a relatively reliable AGB estimation results,the robustness and accuracy of the model were significantly better than the AGB estimation result based on high-resolution multispectral features.AGB modeling based only on high-resolution multispectral features was relatively poor.AGB estimates could be slightly improved with the assistance of a vegetation structure feature,and the optimal AGB model was built by integrating high-resolution hyperspectral features with a vegetation structure feature.In this research,a hyperspectral feature mining method based on the fractional order differential was also proposed.Compared with the integer order differential transformation,more sensitive hyperspectral features to tidal vegetation AGB could be extracted by the fractional order differential transformation.(3)In the research on homogeneous vegetation response unit-oriented tidal vegetation multi-scale AGB estimation method,a homogeneous vegetation response unit(HVRU)was constructed,which integrated multi-source data including high-resolution hyperspectral data,small-footprint Li DAR data etc.and geoscience knowledge.Based on multi-source information fusion,multi-scale segmentation technology and random forest machine learning algorithm,an HVRU-oriented multi-scale AGB estimation method was proposed,which could conveniently realize multi-scale tidal vegetation AGB estimates.Furthermore,comparative experiments were carried out to evaluate the effect and utilities of spectral factors/structural factors/geographical factors,linear parameter regression algorithm/nonlinear nonparametric regression algorithm and scale sizes in multi-scale AGB estimation modeling for short tidal vegetation with low AGB.The optimal research scale and optimal model configuration and maximum contribution factor in multi-scale AGB modeling were explored.Results showed that the HVRU-oriented multi-scale AGB estimation method avoided the the collection of multiple datasets under multiple spatial resolutions with advantages in flexible scale transformation and strong modeling robustness etc.HVRUs included geoscience knowledge and represented specific landscapes;they can not only quantify AGB,but also describe the shapes of AGB homogeneous areas,changing the previous presentation mode based on rectangular pixels with different spatial resolutions.Random forest algorithm performed better than partial least square algorithm in multi-scale modeling.Optimal AGB estimation model at each scale was constructed by fusing multi-source factors.The assistance of structural factors was helpful to improve AGB estimates at all scales,however,the introduction of geographical factors showed inconsistent utilities in multi-scale AGB modeling.Scale=15 was the optimal research scale in this study area,and the best AGB estimates were achieved by fusing spectral factors,structural factors and geographical factors,vegetation canopy height contributed most in modeling.(4)In the research on fine-scale dynamic monitoring of tidal vegetation by remote sensing,the constructed fine classification model and AGB estimation model in previous chapters were applied in a typical coastal area of China to classify and estimate the AGB of typical tidal vegetation based on remote sensing data in 2014 and2019.From the perspectives of cover area,spatial distribution and growth,the evolution of tidal vegetation in the past five years were revealed,and the direct impacts of human activities on the cover area of tidal vegetation were evaluated based on the changes of land use types.A simulation model of the evolution of vegetation types was constructed to predict the cover area and spatial distribution of tidal vegetation in 2024,and the development trends in the next five years.Results showed that,during 2014-2019,the cover area of tidal vegetation tended to be stable,the proportions of Suaeda salsa and Phragmites australis were the largest,the proportion of Spartina alterniflora was relatively smaller,and the cover areas of tidal vegetation in inland non-protected areas fluctuated significantly,the average AGB of tidal vegetation was basically unchanged.Human activities had a great influence on tidal vegetation.The changes of land use types directly leaded to the reduction of cover areas of Suaeda salsa and Phragmites australis respectively by 3.02 km~2 and 2.79km~2(more than 50%of total transferred area),which were mainly transformed into water areas and impervious areas.By 2024,the spatial distribution of tidal vegetation will be stable as a whole,the cover area of tidal vegetation will continue the evolution from 2014 to 2019,and it will decrease from 52.29 km~2 to 50.57 km~2.The development trend of each tidal vegetation type will also follow the dynamics from2014 to 2019,but the rate of increase or decrease will slow down. |
