Stand Density And Aboveground Biomass Extraction Based On VHR Imagery

Accurate forest resources monitoring and survey is an important part of national forestry planning and design,and it is also an indispensable part of ecological forestry research.Traditional survey methods are time-consuming and labor-intensive and have low accuracy.Remote sensing,surveying,and geographic information system have become important research technique.In this paper,we developed the scheme of individual and plot absolute positioning in mountainous area in the Beijing Jiufeng and Fujian Jiangle forest farm and we also estimate the stand density and aboveground biomass based on the Local maximum method and empirical model by taking the QuickBird and Worldview-2 high spatial resolution remote sensing images as the data source.The main conclusions include:1.Image preprocessing process is proposed,mainly including orthorectification,atmospheric correction,topographical correction,image fusion using QuickBird imagery for example.Total station and handheld differential GPS was working together to locate the plot position in mountain area based on QuickBird imagery.After that,the method of fast absolute positioning of individual tree in mountainous area and also the correction scheme of the error coordinate were proposed.The method can lay foundation for the extraction of forestry vegetation parameters for further studies.2.The paper mainly studied the local maximum(LM)value filtering to extract forest stand density,focusing on analyzing the effect of five different window sizes(3×3,5×5,7×7,9×9,11×11)and NDVI threshold value(NDVI>0.2,0.3,0.4,0.5,0.6)to the amount of the local maximum points.A linear regression model was established between the spectral local maximum number of points and the actual stand density of the forest in two different study areas.The results show that:(1)Firstly when the same data source(QuickBird)is used in different study area in the stage of individual tree detection based on the local maximum spectral filtering.3×3 window size was mainly used in Jiufeng plantation area via verifying,while Fujian plantation area mainly uses a 7×7 window size.The NDVI threshold range used for QuickBird image or Worldview image was basically about NDVI>0.5 for Fujian research area.NDVI threshold is basically the same in the same study area and different data source.It concludes that the influence of stand conditions and lighting conditions on the estimation results is greater than that of remote sensing data source images.(2)Secondly,our research also indicate that the classification of the study object contributes to the improvement of the accuracy.Accuracy of the coniferous forest is 80.86%with R2=0.7933,RMSE=12.60 and it is higher than that of whole forest 75.92%with R2=0.7017,RMSE=11.20 and is also higher than that of broadleaf forest 71.522%with R2=0.44,RMSE=9.02 when the sample plots were classified into coniferous forest and broadleaf forest when the 2008 QuickBird panchromatic band is used for example.The same situation in Fujian study area,when using 2013 Worldview-2,the classification of study object could improve the estimation of the stand density.The accuracy of stand density in Cunninghamia lanceolata forest 72.5%with R2=0.502,RMSE=35.77 and Pinus massoniana forest 78.35%with R2=0.754,RMSE=41.46 are both higher than that in whole forest with R2=0.2907.(3)Thirdly,by comparing the extraction result of the stand density using QuickBird and Worldview-2 panchromatic band and near-infrared band after fused,found that it can achieve better results when near-infrared band was used to extract the stand density.The result of QuickBird near-Infrared image R2=0.6264 is higher than the panchromatic band R2=0.5455 using all forest sample plot while the Worldview image of the red edge band(705-745nm),NIR I(770-895nm)and NIR?(860-1040nm)are all higher than the panchromatic band R2=0.2351.Finally,this paper choosed the NIR2 to estimate the stand density in Jiangle study area and choosed NIR to estimate the stand density in Jiufeng study area.3.Biomass inversion research based on the high spatial resolution imagery found that:(1)it can improve the accuracy of estimation of biomass by dividing the forest type into coniferous forest and broadleaf forest.Taking Jiufeng coniferous forest farm for example,the highest coefficient of determination of coniferous model which contains stand density factor in Jiufeng is 0.5471 with RMSE 34.3 t/ha with accuracy=71.8%.Accuracy improves greatly compared to the unclassified forest stand.The estimation accuracy of the forest biomass estimation model based on texture information in Jiangle Forest Farm is 81.5%,R2=0.8481,RMSE=19.78,which is 18%higher than that of the unclassified texture biomass estimation model,which indicates that the forest stand type classification is helpful to improve the biomass estimation accuracy.(2)Texture-stand density-spectral-based biomass estimation model in Jiangle unclassified forest 63.33%is higher than that 61.54%in stand density-based biomass estimation model.Stand density contributes little to the estimation of broadleaf.(3)Compare the QuickBird imagery to Worldview-2 imagery,the factors from QuickBird have low correlation with AGB which is not sufficient to establish the models.The best fit factor is NDVI with correlation coefficient r=0.3775.However,AGB estimation model based on the texture,stand density and spectral information using Worldview-2 imagery achieve better result(R2=0.3733;RMSE=59.09 t/ha with accuracy=63.33%).The study in Fujian study area illustrates that the additional four bands of Worldview-2 band has more potential to extract AGB than QuickBird image.