Identification And Inversion Of Plants In Seriphidium Transiliense Desert Grassland Based On Multi-Source Remote Sensing Data

It is the basis for classifying grassland types,clarifying grassland degradation and restoration,and protecting grassland plant diversity to effectively understand the changes and fluctuations of grassland plant population.In this study,the main plants of the Seriphidium transiliense desert grassland concentrated in Xinjiang were studied.In April,June and September,the multi-source information of ground hyperspectral imager,UAV remote sensing and satellite remote sensing is used to analyze the variation law and trend of plant canopy spectral reflectance,extract characteristic bands,screen and identify sensitive bands,and identify the identified objects,so as to realize the inversion of plant coverage and biomass,and clarify the differences of plant canopy spectral characteristics under different spatial resolution remote sensing data under time sequence changes;The identification accuracy of different identification and inversion models is discussed in order to provide technical basis for the identification of main species of this kind of grassland and their effective monitoring and protection.The main results are as follows:(1)Using ground hyperspectral data to observe the two main plants of S.transiliense desert grassland,it generally shows the typical spectral characteristics,but the reflection amplitude in 400 ～ 1000 nm band are different in different periods;The spectrum of bare land tends to be linear,which is significantly different from that of vegetation.The identification parameters selected in different months are the same,and the OIF value is the largest in 638.64,789.49 and923.79 nm bands.The recognition accuracy between classifiers is SVM > CNN;The months are April > September > June,and the recognition objects are land > S.transiliense > C.arenarius.The overall recognition accuracy in April is 86.16%for CNN and 92.12% for SVM;Using the identified spatial position to extract the coverage of single species,the inversion accuracy is consistent with the identification accuracy;Using vegetation index to retrieve the biomass of Seriphidium and Chenopodium respectively,the best fitting vegetation index obtained has no difference among plants,but it is different between months.S.transiliense had the best fitting effect with NDVI in September,C.arenarius had the best fitting effect with DVI in June,and aboveground biomass had the highest fitting effect with NDVI in September.(2)The reflectance of ground objects in UAV spectral data decreases gradually with the increase of flight altitude,which is 15 m > 30 m > 60 m.S.transiliense and C.arenarius in different phenological periods showed the trend of "low high low" spectral reflectance in the visible wave band;The same as ground hyperspectral,the recognition accuracy among classifiers is SVM > CNN;The months are April > September > June;The identified objects are land > S.transiliense > C.arenarius;15 m > 30 m > 60 m at flight altitude;The total recognition accuracy is the highest at the flight altitude of 15 m in April,83.65% for CNN and 86.23% for SVM;The coverage extraction is consistent with the classification results.For the biomass inversion parameters,The fitting effect is the best at the height of 15 m.The biomass inversion of the two plants was consistent with the vegetation index of community biomass inversion,and the fitting effect was best with NDVI,DVI and RVI in April,June and September,respectively.(3)The measured ground biomass was correlated with GF-2 vegetation index.In April,RVI fitted best;In June,the fitting effect of DVI was the best;The prediction accuracy of April and June is 69.58% and 65.25% respectively.The UAV inversion biomass is used as the modeling sample of GF-2 inversion biomass,which improves the inversion accuracy of community biomass,and the fitting effect of NDVI in April is the best,and the fitting effect of DVI in June is the best,and the prediction accuracy is 79.47% and 75.78% respectively.