1 Best practices for generating forest inventory attributes from airborne laser scanning data using the area-based approach Joanne White Research Scientist Canadian Forest Service CIF Best Practices Workshop March 18, 2014
2 Best Practices Guide Released July 2013 Synthesizes 25 years of scientific research Available for download from CFS bookstore: http://cfs.nrcan.gc.ca/public ations?id=34887
3 Technology Laser altimetry, Light Detection And Ranging (LiDAR), Airborne Laser Scanning (ALS) Digital Surface Model Point Cloud Digital Terrain Model
4 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
5 Introduction Growing interest in using LiDAR data to develop enhanced forest inventories (EFIs) Synthesize best practices from scientific literature Scope: data acquisition to inventory attributes Provide information for implementation, RFPs, standards development Descriptive not prescriptive Importance of ground plots are emphasized
6 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
7 Area-based approach 1. Grid the point cloud 2. Calculate wall-to wall LiDAR metrics 3. Ground sample within the range of variability characterized by the LiDAR metrics 4. Clip the point clouds to the area corresponding to the ground plots 5. Develop model 6. Apply model
Area-based approach: Output example 8 ALS-derived wall-to-wall predictive surfaces for the Romeo Malette Forest in Ontario (Tembec). Average height and gross merchantable volume were estimated using models described in Woods et al. (2011). Forest-level averages ± 95% confidence intervals are reported.
9 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
10 Airborne Laser Scanning Data ALS data appropriate for ABA may be characterized by: Small scan angles (< ± 12 degrees) A minimum of 1 pulse per square metre and > 4+ pulses per square metre for dense forests on complex terrain A sensor capable of recording a minimum of 2 returns per pulse (4-5 returns/pulse is typical) 50% overlap of adjacent flight swaths A single survey, at the same time, with the same instrument Leaf-on or leaf-off, but not a mix of both Minimum ALS products required for the area-based approach are the bare earth DEM and the classified (unfiltered) ALS point cloud The point cloud should contain all valid returns Appendix 1 provides suggested listing of details required for a RFP for data acquisition
Airborne Laser Scanning Data: Quality assessment 11 LiDAR instrument(s) used (could be more than one) Acquisition parameters (and documentation) (e.g., date[s], altitude) Environmental conditions during acquisition (specifically fog and precipitation) Processing methods (and documentation), including software and specific procedures followed Projection/datum information Spatial coverage of the LAS files and DEMs provided (i.e., complete spatial coverage provided, no gaps in acquisition) Adherence to fundamental, supplemental, and consolidated vertical accuracy requirements Content of the LAS files provided (i.e., are the returns classified appropriately and consistently? Is scan angle provided?) Reported pulse density
12 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
13 Generation of Point Cloud Metrics LiDAR metrics = descriptive statistics from the point cloud over a unit area Canopy height and density metrics Hundreds of metrics are possible (e.g., mean height, 75 th percentile of height, canopy cover) Many are intercorrelated Generated on a unit area (i.e., a 25 m by 25 m grid cell) Scientific literature indicates that metrics related to height, coefficient of variation of height, and density of cover are most commonly used in predictive models Appendix 2 provides suggested elements for inclusion in an RFP for metric processing
14 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Generation of Point Cloud Metrics: Software 15 FUSION is a simple, robust freeware tool developed by the US Forest Service. (Also LASTools: www.rapidlasso.com) Tutorial for the FUSION software: http://www.fs.fed.us/eng/rsac/lidar_training/introduction_to_fusion/story.html There is also a more advanced module that actually steps through the full process required to go from data to modelled outcomes. http://www.fs.fed.us/eng/rsac/lidar_training/forest_inventory_modeling/story.html
16 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Generation of Point Cloud Metrics: Grid cell size 17 Must match the size of the ground plot Must be sufficiently large to enable the development of robust predictive models (i.e., 20 x 20 m, 25 x 25 m) 20 x 20 m; area = 400 m 2 ; a circular plot with a radius of 11.28 m 25 x 25 m; area = 625 m 2 ; a circular plot with a radius of 14.01 m UTM grid; divides evenly into 100
18 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Generation of Point Cloud Metrics: Tiling of area of interest 19 Processing of metrics is computing intensive and typically requires the area of interest to be divided into manageable size units for the sake of processing efficiency (i.e., 5 km by 5 km pieces) Processing then iterates by tile Output metrics are then mosaicked to create a single, wall-to-wall metric for the area of interest
20 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Generation of Point Cloud Metrics: Metrics 21 All returns or first and last returns separately? Metric selection height, coefficient of variation of height, and density of cover Principal Component Analysis Use of a minimum height threshold Separate canopy from non-canopy necessary for modelling > 2 m Output should be subject to a QC process build a common no data mask check for outliers
22 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Selection of ground plots Pre-existing Inventory Plots Stratified random sample Select ground plots to capture full range of structural variability
Ground Plot Data Characteristics: Size 24 Larger plots are needed to: Reduce perimeter-to-area ratio
Ground Plot Data Characteristics: Size 25 Larger plots are needed to: Reduce perimeter-to-area ratio Reduce likelihood of edge effects A B
Ground Plot Data Characteristics: Size 26 Larger plots are needed to: Reduce perimeter-to-area ratio Reduce likelihood of edge effects Minimize geolocation error
Ground Plot Data Characteristics: Size 27 Larger plots are needed to: Reduce perimeter-to-area ratio Reduce likelihood of edge effects Minimize geolocation error No universal optimum plots size 200 625 m 2 minimize edge effects minimize planimetric co-registration error maximize sampling efficiency, precision, and accuracy of target and explanatory variables
Ground Plot Data Characteristics: Shape 28 Fixed-area circular plots are preferred circular plots more common in the literature easier to establish 13% less perimeter than square plots of equal area
Ground Plot Data Characteristics: Other considerations 29 Timing of acquisition: Recommend ground data acquired after ALS acquisition; ALS metrics calculated and used to guide sample selection and location Need to minimize the time elapsed between ground and ALS data collection (within one growing season) Specific canopy conditions occurring at the time of both ground and ALS data collection (i.e., seasonal growth stages or leafon/leaf-off conditions where deciduous species are present).
30 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Ground Plot Data: Representativeness 31 Higher errors in modelled outcomes associated with ground calibration data that does not capture the full range of structural variability as captured by the ALS data Models perform best when operating within the bounds of their original calibration data.
32 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Ground Plot Data: Selection of ground plot locations 33 Structurally-guided sample Use a few key ALS metrics to stratify the area of interest (height, COV of height, canopy cover) Select the required number of samples within each strata
34 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
Ground Plot Data: Positioning 35 Accurate geo-referencing is fundamental to maximize the predictive power of the model Recall that larger plots can help mitigate the impact of geolocation error GPS positioning is challenging in forest environments Mapping-grade GPS receivers 500 points/location Post-processing correction
36 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
37 Tree measures Key response variables for model development Basic suite of attributes for direct measurement Species Status (live or dead) Crown class (dominant or co-dominant) Diameter at breast height Height (for a sub-sample representative of the dbh frequency distribution) Stem number Derived/compiled attributes Basal area (from dbh) Plot height (mean, Lorey s height, dominant height) Volume (gross, merchantable)
38 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
39 Modelling Parametric regression Random forests Advantages Transparent, easy to understand. Model is an equation that clearly quantifies the relationship between the predictors and the variable being predicted. Sample size determination is possible for given accuracy and precision requirements. Categorical variables may be predicted and (or) used as predictors. Faster and simpler to develop (does not require sophisticated statistical expertise). Does not require individual strata based models to be developed, provided calibration data represent the different strata involved. Does not require a pre-existing polygon-based inventory to implement strata-based models. Disadvantages Transformation of ALS metrics (X) or ground plot measures may be necessary to meet the assumptions of regression-based approaches, complicating interpretation and implementation. More statistical expertise and time are required to create the models. With strata-specific models, pre-existing stratification across the entire forest (i.e., an existing inventory layer) becomes prerequisite to implementation. Prediction errors will occur within polygons when individual grid cells do not match the overall strata assignment (e.g., pockets of aspen within a spruce polygon). Black box nature of the models. No equation output that is analogous to parametric regression. More critical to ensure that the full range of conditions are sampled, as this approach does not extrapolate like regression.
40 Parametric Alberti et al. (2013) J Biogeosciences and Forestry
41 Non-Parametric Racine et al. (2014) Forest Science
42 Best Practices Guide 1. Introduction 2. Area-based Approach to Attribute Estimation 3. Airborne Laser Scanning Data 4. Generation of Airborne Laser Scanning Point Cloud Metrics 4.1 Software 4.2 Grid Cell Size 4.3 Tiling of Area of Interest 4.4 Metrics 5. Ground Plot Data 6. Modelling 7. Mapping 5.1 Ground Plot Characteristics 5.2 Representativeness of Ground Plots 5.3 Selection of Ground Plot Locations 5.4 Ground Plot Positioning 5.5 Tree Measures Appendix 1. Airborne Laser Scanning Data Acquisition Appendix 2. Airborne Laser Scanning Point Cloud Metrics Appendix 3. A Sample FUSION Workflow for Metric Calculation
43 Mapping Once validated models can be applied to the entire area of interest using wall-to-wall metrics Common "no data" mask Models developed for specific forest types must be applied correctly Wall-to-wall rasters can be integrated into existing stand level inventories
44 Resources Best Practices Guide http://cfs.nrcan.gc.ca/publications?id=34887 Forestry Chronicle December 2013 Practitioner s Corner http://pubs.cif-ifc.org/doi/pdfplus/10.5558/tfc2013-132 BC Forest Professional Newsletter (Nov/Dec 2013) http://www.abcfp.ca/publications_forms/magazine.asp http://cfs.nrcan.gc.ca/publications?id=35300 CIF Enhanced Forest Inventory website http://cif-ifc.org/site/enhancedforestinventory
45 Thank you! Questions? joanne.white@nrcan.gc.ca Joanne White and Mike Wulder Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC Andrés Varhola and Nicholas Coops Integrated Remote Sensing Studio, Department of Forest Resources Management, Faculty of Forestry, University of British Columbia, Vancouver, BC Mikko Vastaranta Department of Forest Sciences, University of Helsinki, Helsinki, Finland Bruce Cooke NASA Goddard Space Flight Center, Biospheric Sciences Laboratory, Greenbelt, MD, USA Doug Pitt Natural Resources Canada, Canadian Forest Service, Canadian Wood Fibre Centre, Sault Ste. Marie, ON Murray Woods Ontario Ministry of Natural Resources, Southern Science & Information Section, North Bay, ON