Using Lidar and ArcGIS to Predict Forest Inventory Variables Dr. Kevin S. Lim kevin@limgeomatics.com P.O. Box 45089, 680 Eagleson Road Ottawa, Ontario, K2M 2G0, Canada Tel.: 613-686-5735 Fax: 613-822-5145
Presentation Outline Background Methodology Results Conclusions
Background
Purpose of Research To determine the accuracy and precision that forest inventory variables can be predicted using airborne lidar remotely sensed data. - Gross total volume - Gross merchantable volume - Basal area - Density - Quadratic mean DBH - Average height - Top height - Aboveground biomass - Diameter distributions (in progress)
Forest Inventory Variables Variable Abbrev. Definition Top Height (m) TOPHT Calculated as the average of the largest 100 stems per hectare. Average Height (m) AVGHT Calculated as the average height of all trees Density (stems/ha) Density Number of trees per hectare Quadratic Mean Diameter (cm) QMDBH Ø Œº ( DBH 2 n) ø œß Basal Area (m 2 /ha) BA DBH 2 * 0.00007854 Gross Total Volume (m 3 /ha) GTV Honer et al. (1983) equations Gross Merchantable Volume (m 3 /ha) Total Above Ground Biomass (Kg/ha) GMV SUMBIO Honer et al. (1983) equations Ter-Mikaelian and Korzukhin (1997) equations
Plot Data Fixed circular plots - 11.28 m radius 0.04 ha Plots were geo-referenced to sub-meter accuracy Trees with DBH 9.1 cm assessed - DBH - Species - Crown class - Height to base of live crown - Total tree height
Location of Study Site Located in the northeast portion of Ontario s Boreal Forest near Timmins, Ontario. Active forest management unit with approximately 532,000 productive forest hectares. Dominant species are: - black spruce, white birch, trembling aspen, jack pine, eastern white cedar, white spruce, eastern larch, and balsam fir Other minor species include: - black ash, yellow birch, soft maple and red and white pine
Stratification Four Forest Model Units Intolerant Hardwood white birch and poplar 70% Mixedwood conifer (Sb, Pj, Sw, Ce, Bf 40% and Po + Wb = 60%) Jack Pine Pj 70% Black Spruce Sb 70%
Coverage 630,000 ha (2,400 square miles) in Boreal forest 136 model calibration plots 138 model validation plots
Airborne Lidar Data Data acquired in summer of 2004 and 2005 Leica ALS sensor Point (pulse) density approximately 0.5 points/m 2
Methodology
Methodology Define Stratification Acquire LIDAR Data LIDAR Data Acquire Field Data For Sample Plots Classify Point Cloud Intersect With Sample Plots Vegetation Points Ground Points Create TIN Field Data Vegetation Points Per Plot Normalize Points To Terrain TIN Calculate Forest Variables Calculate LIDAR Predictors Normalized Vegetation Points Per Plot Normalized Vegetation Points Forest Variable Statistics Perform Statistical Analyses LIDAR Predictors LIDAR Predictor Surfaces Calculate LIDAR Predictors Regression Models Apply Models to Landscape Forest Inventory Surfaces
Data Processing Approach Divide and Conquer Strategy Custom code leveraging ArcObjects ArcGIS Desktop and Catalog
Normalize Points to Terrain z veg z grd = Z norm = Z veg - Z grd TIN TILE
Canopy Height Models
Lidar Predictors Statistical - Mean - Standard deviation Percentiles of height - Deciles (p10 p90) - Maximum height Canopy density - d1 d9 - Da: Number of first returns divided by all returns. - Db: Number of first and only returns divided by all returns.
Lidar Predictor Surfaces Each surface corresponds to a lidar predictor. Cell resolution of 20m (or 400m 2 in area). Apply a mask (optional).
Regression Modelling Jack Pine Dependent RMSE RMSE variable % BA (m 2 ha -1 ) 5.92 19.0 GTV (m 3 ha -1 ) 44.85 18.0 GMV (m 3 ha -1 ) 36.67 18.8 QMDBH (cm) 1.54 9.0 AvgHT (m) 1.05 6.5 TopHT (m) 0.76 3.8 Biomass (Kg ha -1 ) 24478 19.2 Black Spruce Dependent RMSE RMSE variable % BA (m 2 ha -1 ) 4.83 18.7 GTV (m 3 ha -1 ) 39.57 24.3 GMV (m 3 ha -1 ) 26.25 24.1 QMDBH (cm) 1.37 9.7 AvgHT (m) 1.13 8.8 TopHT (m) 1.24 7.4 Biomass (Kg ha -1 ) 19996 19.8 Intolerant Hardwood Dependent RMSE RMSE variable % BA (m 2 ha -1 ) 5.05 16.1 GTV (m 3 ha -1 ) 45.18 17.0 GMV (m 3 ha -1 ) 45.17 20.9 QMDBH (cm) 1.64 9.0 AvgHT (m) 1.02 6.1 TopHT (m) 0.88 3.9 Biomass (Kg ha -1 ) 29880 23.2 Mixedwoods Dependent RMSE RMSE variable % BA (m 2 ha -1 ) 5.58 17.1 GTV (m 3 ha -1 ) 48.66 18.4 GMV (m 3 ha -1 ) 42.72 19.1 QMDBH (cm) 2.01 10.3 AvgHT (m) 1.20 7.7 TopHT (m) 0.94 4.2 Biomass (Kg ha -1 ) 25998 18.8
Apply Models to Landscape
Methodology Define Stratification Acquire LIDAR Data LIDAR Data Acquire Field Data For Sample Plots Classify Point Cloud Intersect With Sample Plots Vegetation Points Ground Points Create TIN Field Data Vegetation Points Per Plot Normalize Points To Terrain TIN Calculate Forest Variables Calculate LIDAR Predictors Normalized Vegetation Points Per Plot Normalized Vegetation Points Forest Variable Statistics Perform Statistical Analyses LIDAR Predictors LIDAR Predictor Surfaces Calculate LIDAR Predictors Regression Models Apply Models to Landscape Forest Inventory Surfaces
Output
Spatially Explicit Predictions A prediction for every 20 m cell!
What s Next?
Advanced Forest Resource Inventory Decision Support System (AFRIDS) Lightning Talk: Advanced Forest Resource Inventory Decision Support System - LIDAR in Action
Conclusions
Concluding Remarks The science behind using airborne lidar to predict forest inventory variables has been published on extensively. Lidar data is affordable. GIS technology is well suited to handling the large lidar data volumes. A lidar enhanced forest inventory supports both tactical and strategic needs. Consider lidar as a complementary technology to traditional approaches instead of as a replacement.
Why settle for the traditional
When you can have predictions for every cell!
Acknowledgements