A Study On The Dynamic Of Vegetation In Elm(Ulmus Pumila) Sparse Forest Grassland Ecosystem Based On The Machine Learning Algorithms And Unmanned Aerial Vehicles

Fractional vegetation coverage(FVC)is an important indicator to assess ecological environments and vegetation dynamics.It is also an important parameter to express vegetation dynamics in multiple land models and ecological models.Woody vegetation in dryland ecosystems are important components of vegetation carbon storage,also affect the ecological processes and land grazing capacity.Satellite-based remote sensing and ground investigation are the general methods to estimate the FVC.Satellite remote sensing provides an effective method for long-term dynamic detection of vegetation.However,even if the spatial resolution that reach the high-resolution satellite image with the accuracy of the meter level is difficult to accurately distinguish the herbaceous and woody in sparse forest steppe and evaluate the annual and inter-annual dynamic.Therefor,it is challenging to precisely estimate the FVC of woody and herbaceous in elm sparse forest grasslands.Unmanned aerial systems(UAS)provide a effective solution to bridge the gap between satellite-based remote sensing and field-based measurements.It is a new method of UAV-based vegetation monitoring platform to accurately monitor the vegetation dynamics of sparse forest steppe on the landscape scale.The purpose of the study was to propose an integrative tool for quickly,accurately distinguish the vegetation types and estimating FVC by coupling a UAS monitoring platform with decision tree algorithms.We applied this tool to observe the vegetation dynamics in elm(Ulmus pumila)sparse forest grassland ecosystem(ESFOGE)plot during a growing season in2016-2018.The results show:(1)The UAV-based vegetation remote sensing platform has flexible flight time that obtain centimeter-level image data of landscape scales in time.The spatial resolution of the digital orthophoto map(DOM)derived from UAS is 2.67 cm/pixel with UAV flights at a height of 100 m;(2)Image classification method based on machine learning can efficiently and accurately calculate sparse herbaceous and the woody vegetation coverage,the kappa coefficients for the classification accuracy respective in the early,middle and end of the growing season are 0.72,0.74 and 0.63;(3)The woody and herbaceous vegetation of theESFOGE plot were classified and their FVC were estimated as 14 ± 1% and 60 ± 4%,respectively,on the DOM by decision tree algorithms.The FVC variation of herbaceous vegetation was larger than that of woody plants during a growing season.The FVC of the ESFOGE plot was 75 ± 5%;(4)The contribution of woody and herbaceous vegetation to vegetation coverage was 19.5% and 80.5%,respectively.The FVC of the ESFOGE-plot was more influenced by herbaceous vegetation;(5)In 2016-2018 with annual precipitation of 426.9mm,297.2 mm and 392.3 mm,the vegetation of eucalyptus forest in the low grazing intensity did not change significantly.The high grazing intensity caused significant changes in the woody plants of the eucalyptus forest.Overall,this research proved that a UAS monitoring platform is an effective tool for observing the vegetation status at a landscape scale.It automatically and quickly classified the vegetation type and estimated the FVC by coupling with decision tree algorithms.To our knowledge,it is the first time that the FVC dynamics of woody and herbaceous vegetation derived from a UAS platform during a growing season in the ESFOGE;The grazing intensity is the main factor leading to the difference of different types of vegetation in the sparse forest steppe,and the coverage of woody vegetation is the most sensitive factor response to grazing intensity.