EXTRACTION OF FOREST INVENTORY PARAMETERS AND 3D MODELING FROM AIRBORNE LIDAR LiDAR point cloud Vinod Kumar Indian Forest Service Esri International User Conference, 2014 San Diego, USA LiDAR Technology (Light Detection And Ranging) 1.Laser scanner 2.Differential GPS 3.Inertial Measurement Unit (IMU) 4.On board computer to store data Trees in Point cloud Trees in Point Cloud Point cloud & normalised point cloud Rasterization of Point cloud DSM Single tree in point cloud & in CHM Single tree in normalized point cloud Single tree in CHM Canopy Height Model in 3D Research Objectives 1. Tree peak identification and crown delineation 2. Extraction of forest inventory parameters 3. Species classification 4. Carbon estimation 5. 3D Forest Modeling Study Area •Area = 1.5 KM2, ortho A=1.3 KM2 •92 % forest •Mainly coniferous forest Pinus uncinata •Slope 10° to 35° avg slope=14.5° •Undulating terrain •Frequent landslides DTM DSM Ortho-image with elevation from DTM LiDAR derived DTM LiDAR derived DSM Actual Picture DSM Tree Peak detection in enveloping canopy surface CHM Smooth CHM Smoothing CHM Filter size 0.45 m sCHM3 Filter size 0.75 m sCHM5 Filter size 1.05 m sCHM7 Smooth CHM Filter size 1.35 m sCHM9 Peaks identified in sCHM9=54,194 Peaks identified in sCHM7=92,867 Peaks identified in sCHM3=164,787 Final Peaks =128,918 Filter Size (m) Height cut off (m) 1.35 >20 1.05 >16 and <= 20 0.75 >11 and <=16 0.45 <=11 Tree peak identification in point cloud No. of sample trees 275 Filter size- 0.45 m Filter size- 0.75 m Filter size- 1.05 m Filter size- 1.35 m No. tree peaks identified 264 Accuracy % 96 Canopy Height Model SEGMENTATION Peak identification Binary Gap Mask Region Grow Segmentation Polygon Smoothing Thiessen Polygon Segmentation Thiessen Segmentation on Ortho- image Thiessen Polygons Thiessen Polygons updated with Gaps and applying polygon smoothing D Value of Segmentation (fraction of error) 1:1 correspondence Point cloud of 1000 random trees extracted to visually see the result of segmentation in 3D Tree Height Canopy Volume Tree canopy enveloping surface Canopy Volume contained within these two surfaces Tree Canopy base Touching surface Canopy Base Height CBH is calculated as average height of canopy base touching surface C Canopy base basetouching touchingsurface surface Canopy • INVENTORY_DATA Accuracy of extracted parameters Parameter R2 RMSE CPA (Region growing smooth) 0.87 3.67 m2 CPA (Thiessen smooth) 0.90 3.16 m2 Canopy base height (CBH) 0.73 0.86 m Canopy tilt 0.57 3.26 degree Descriptive Statistics Variable n Mean Minimum Maximum Std. deviation CHM height (m) 126,653 9.82 0.03 34.53 4.12 CPA (m²) 126,653 6.73 0.32 108.79 5.47 Avg CPA height (m) 126,653 7.69 2 26.53 3.39 Canopy Volume (m³) 126,653 37.91 0.05 1211.3 46.82 Crown diameter (m) 126,653 2.74 0.63 11.77 1.03 CBH (m) 126,653 2.76 2 14.85 1.63 CBH/Height 126,653 0.28 1.5 150.45 0.73 Tree_Inclination (degree) 126,653 84.03 8.24 89.98 4.68 SLOPE (degree) 126,653 21 1 87 11 ELEVATION (m) 126,653 1632 1403 2038 124 Tree Density (trees/ha) 126,653 1451 42 5260 611 LidarHits/m² 126,653 122 0 1606 90 LiDAR_Pts/tree 126,653 945 0 55654 1407.09 CanopyDensity(Pts/vol) 126,653 28.18 0 1609.1 26.34 Gap_Percent 126,653 17.41 0 100 21.67 CD/Height 126,653 0.32 0.04 95.88 0.55 Query Larix CPA>15 m² Height>12 m Canopy Volume>130 m³ P. uncinata: Elevation>1600m P.sylvestris: Elevation<=1600m No. of trees Species distribution Larix decidua 80000 70000 60000 50000 40000 30000 20000 10000 0 Series1 Larix Broad leaved P. sylvestris P. uncinata 3538 9724 45131 69790 Species classification accuracy Canopy tilt<70 degree in different Zones No of trees Landslide Zone 75690 Nonlandslide 45778 Zone Total 121468 Zone Tilt<70 degree 1303 1.7 673 1.5 1976 1.6 % Open Canopies in different Zones Zone No of trees Landslide Zone 75690 Gap%>50 % 7887 10.4 Nonlandslide Zone 45778 3512 7.7 Total 121468 11399 9.4 Gap % Close Canopies in different Zones Zone No of trees Landslide Zone 75690 Gap%=0 % 29358 38.8 Nonlandslide Zone 45778 17133 37.4 Total 121468 46491 38.3 Regression Models for Carbon • Model1: carbon= a+b*Height+c*CPA • Model2: carbon= a+b*Height+c*CanopyVolume • Model3:carbon=a+b*Height+c*tree_density*Height+d*CPA*Height =a+Height*(b+C*tree_density+d*CPA) Model R2 RMSE (Kg) Model 1 0.757 12.293 Model2 0.751 12.439 Model3 0.782 11.539 Carbon estimation of Pines in the study area No of Mean Pines Carbon(Kg) Carbon 114921 33.16 Total Carbon (Tons) 3811 Conclusion • Tree peaks can be detected with high accuracy 96 % Segmentation Thiessen was found comparable to Region growing 94.2 % 93.5% Thiessen Region growing Tree parameters can be extracted with significant accuracy from LiDAR. New approach •Canopy Volume •CBH •Canopy tilt •Canopy Orientation Carbon Models It is possible to make carbon models from extracted parameters Model R2 RMSE (Kg) Model 1 0.757 12.293 Model2 0.751 12.439 Model3 0.782 11.539 • Species classification is possible with structural and spatial information 97% 3D forest modeling from inventory data • Tree: Location, Species, Height, Inclination, Orientatioion. Actual Photo Model Photo Visualizing degraded forest areas Visualizing forest path & regeneration Visualizing inaccessible areas Visualizing landslide areas Visualizing species richness https://www.youtube.com/watch?v=dkfolP-e6Uo Uses of Forest Inventory • Working Plan – – – – • • • • • Growing stock Biomass, Stock Carbon Vegetation density Species distribution Forest Disturbance Production Forestry Forest growth modeling Predictive Forest modeling Plantation and regeneration monitoring Uses of Forest Inventory • Watershed Management Analysis • Fire Modeling • Eco-tourism [email protected]
© Copyright 2024 ExpyDoc
