Airborne LiDAR for EFI in the boreal: what s operational ; what s R&D today

Transcription

1 Airborne LiDAR for EFI in the boreal: what s operational ; what s R&D today For. Chron.: 87: SL SL SL Doug Pitt CIF-Manitoba Section Enhanced Forest Inventory Jan. 23, 213, Winnipeg, MB

2 Airborne LiDAR for EFI in the boreal: what s operational ; what s R&D today Background For. Chron.: LiDAR 11 87: Example of operational SL implementation SL Future SL directions Doug Pitt CIF-Manitoba Section Enhanced Forest Inventory Jan. 23, 213, Winnipeg, MB

3 Background Provincial Forest Inventories - limitations Typical FMU is > 1 M ha Strategic (forest-level) Serves multiple needs Based on manual interpretation of aerial images broad species, age, height, stocking, sometimes by ecosite. Timber volume is imputed. Delivery is a real bottleneck!

4 Background Needs today Volume of merchantable wood, sawlogs. Basal area, biomass, stems per ha Ave. tree size (diameter, height, volume). Tree-size distribution. Information to build better, smarter, roads. Information to reduce costs and maximize profits. Value chain optimization!

5 Operational Cruising Background ˆ y Height Volume (GTV, GMV) Basal area Density Quadratic mean DBH Biomass Diameter and Basal Area Distributions Not spatial!

6 CWFC National Program; Regional Delivery Background FPInnovations: CWFC, Forest Ops., Wood Prod., Pulp&Paper CFS: Luther, Leckie, Gougeon, Wulder, Beaudoin Universities: UBC, Queen s, Nipissing, Sherbrooke, UQAM, Laval, Sir Wilfred Grenfell Provinces: BCMoFR, ASRD, OMNR, MRNF-QC, NLDNR, NBDNR Industry: West Fraser, Tembec, Foothills G&Y, Corner Brook P&P, J.D. Irving Funding partners: GEOIDE, OCE, FQRNT, ACOA, AIFI, NSERC Inventory systems that support VCO by allowing us to send the right wood to the right markets, at the right time! Airborne LiDAR: big part of the Program!

7 Advanced An operational example Forest Resource Inventory Technologies Murray Woods - Doug Pitt Dave Nesbitt Dave Etheridge Kevin Lim Margaret Penner - Don Leckie François Gougeon Paul Treitz Jeff Dech

8 Team AFRIT Murray Woods Doug Pitt Dave Nesbitt Margaret Penner Forest Analysis Ltd. Kevin Lim Lim Geomatics Paul Treitz Jeff Dech Don Leckie Dave Etheridge François Gougeon Great bunch to work with!

9 Background What is Light Detection and Ranging? Active remote sensing technology; transmit & receive ~5, pulses of laser light per second Each pulse can produce multiple returns (up to 8) GPS provides the exact X-Y-Z position of each return

10 Background Basic Products Digital Surface Model (DSM) Digital Terrain Model (DTM) Canopy Height Model (CHM)

11 Background Huge array of uses Hazard mapping Floodplain/risk mapping Landform Classification-ELC Corridor/Right-of-way Mapping Woodlot Extraction Agricultural mapping Geological Mapping Urban Modeling Predictive Hydrology Transmission Line corridors Wetlands/Riparian areas Open pit mining Coastal/Shoreline Mapping Habitat modeling Forest Engineering Forest Inventory www. Saminc.biz Ambercore.com Opportunities for cost-sharing!

12 height Background Using the point clouds for area-based estimates D(%) LiDAR Data by elevation LiDAR Point Cloud P Cloud Statistics 1

13 Predicted stand volume - total (m 3 /ha) Predicted Background Using the point clouds for area-based estimates 35 3 Sb GTV (m 3 /ha) = (mean p9) tvol 1: Actual Actual stand volume - total (m 3 /ha)

14 Doing even better Model construction y = b o + b 1 x 1 + b 2 x 2 y 1 = b o + b 1 x 1 + b 2 x 2 y 2 = b o + b 1 x 1 + b 2 x 3 y p = b o + b 1 x 2 + b 2 x 3 y = e b x 1 b1 x 2 b2 x 5 b5 RandomForest (nonparametric approach based on an ensemble of regression tree models)

15 Doing even better Model construction RF one tree :

16 Doing even better Model construction RF one tree :

17 Model construction Doing even better RF many trees = a forest :

18 Predicted Predicted mean Tree tree Diameter diameter (cm) (cm) 2 m Background Using the point clouds for area-based estimates 45 2 m Actual Actual mean Tree tree Diameter diameter (cm) (cm) ˆ y Height Basal area Density Volume (GTV, GMV) Biomass Quadratic mean DBH Mean tree volume Size distribution Now, we have spatial data!

19 Both tactical AND strategic! Background

20 Two Forests ~ 2M ha Northeastern Ontario An operational example HEARST LiDAR data; relatively low-resolution ROMEO MALETTE Parameter Romeo Malette Hearst Sensor Leica ALS4 Leica ALS5 Platform King Air 9 Cessna 31 Pulse Rate 32,3 Hz 119, Hz Scan Rate 3 Hz 32 Hz FOV 2 deg. 3 deg. H 2,74 2,4 m Line spacing Vert. Accuracy Pulse density 9 m < 5 cm ~.5 /m 2 1, m < 3 cm ~1. /m 2 Relatively low pulse densities!

21 LiDAR predictive models An operational example Top and Dom/codominant Height QMDBH Volume (GTV, GMV) Basal area Biomass Density* Mean tree volume Sawlog volume Diameter/volume Distributions * Derived from DBHq & BA Proper calibration is essential! Murray Woods

22 An operational example LiDAR predictive models

23 Some results Feedback from the woods OBM 2m Old Lidar 5m New Old New Basic products = huge advantages!

24 Some results Feedback from the woods

25 Some results Feedback from the woods LiDAR Information can add more value to our basic image and inventory interpretation Virtually walk every square inch!

26 GMV m 3 /ha Some results Feedback from the woods PJ9 Sb1 17m 6.91 Site Class 2 CC=6% PJ9 Sb1 18m 6 1. Site Class 2 CC=6% Plonski's Normal Yield Table Jack Pine - Site Class m 3 /ha 15.8 cm PJ9 Sb1 18.6m 6 1. Site Class 2 CC=6% Stand Age

27 GMV m3/ha Stems/ha GMV m3/ha Stems/ha GMV m3/ha Stems/ha 5 Some results Feedback from the woods PJ9 Sb1 17m 6.91 Site Class 2 CC=6% PJ9 Sb1 18m 6 1. Site Class 2 CC=6% LiDAR Derived /-14.6 m 3 /ha /-.7cm / m 3 /ha 2.7 +/-.8 cm Mean Volume and Density by Size Class Stand: Size Class VOLUME DENSITY Mean Volume and Density by Size Class Stand Size Class VOLUME DENSITY PJ9 Sb1 18.6m 6 1. Site Class 2 CC=6% /-12.2 m 3 /ha /-.5 cm Mean Volume and Density by Size Class Stand: Efficiencies in planning! Size Class VOLUME DENSITY

28 Some results Feedback from the woods Clearcuts: e.g., Jack Pine, 35 ha FMP planned: 4,669 m 3 LiDAR predicted: 7,543 m 3 Scaled volume: 7,733 m 3 Image and data courtesy Tembec Inc. The ultimate validation!

29 Some results Advanced Forest Resource Inventory Decision Support System 12 1 Density (stems per ha) GMV (m 3 per ha) 3 Forest-Type strata present: Intolerant Hwd 3 2x2m cells Mixedwood 228 2x2m cells Jack Pine 54 2x2m cells 771 Prediction Units for 3.8 ha IH MW PJ Diameter class (cm)

30 AFRIDS System Some results

31 AFRIDS System Some results

32 AFRIDS System Some results

33 AFRIDS System Some results

34 AFRIDS System Some results

35 AFRIDS System Some results

36 AFRIDS System Some results

37 AFRIDS System Some results

38 AFRIDS System Some results

39 AFRIDS System Some results

40 AFRIDS System Some results

41 AFRIDS System Some results Forest Unit: SB1 Area: 45 ha

42 AFRIDS System Some results Forest Unit: SB1 Area: 45 ha

43 Operational Feedback Some results Cost-benefit the Romeo Malette example e.g., LiDAR Cost Savings 1) Inventory acquisition and processing (6 items) -$.1/m 3 2) Forest operations (2 items) $1.4/m 3 3) Mill operations (4 items) $.3/m 3 Total savings: $1.6/m 3 X 5, m 3 /year: $8,/year Payback 1.3 years Informed decision making pays off!

44 The future OK, so where is the research taking us? LiDAR 25 LiDAR 212 SGM 29 Advances in technology ~ HI-resolution!

45 Semi-auto Species ID The future Augment spectral classifiers with data from LiDAR DTM; CHM; SGM for crown resolution; as a classifier. Can LiDAR support ITC?

46 LiDAR Predicted GMV Raster The future Aiding photo interpretation Use LiDAR rasters to create polygons?

47 Actual (%) Semi-auto Species ID The future n = 346 plots; 86 used for validation: Predicted (%) H HC CH C H: HC: 1 CH: C: 4 96 Fraction correct Actual Bf Bw Ce Pb Pj Po Sb Sw Bf Bw 1. Ce Pb Pj Po Sb Species directly? Sw

48 Wet Areas Mapping The future First step in mapping productivity!

49 driving towards ecosite, The future Add slope, aspect, species preference

50 and soil productivity The future Participants: Clement Akumu, John Johnson, Peter Uhlig, Sean McMurray, David Etheridge (OMNR); Murray Woods (SSIS); Doug Pitt (CWFC); Doug Aspinall (OMAFRA); Paul Arp (UNB); < linked with work in AB, NL, and NB, Time to get excited?

51 GROWING the inventory The future EFI+productivity = view of the future (what mill infrastructure is needed; what impact will silvicultural investment have on the outcome?) Explore successive LiDAR/SGM to quantify change Spatially predict the future?

52 Adding FIBRE Attributes The future Participants: Jeff Dech (Nipissing U); Bharat Pokharel (Nipissing U); FPInnovations; Art Groot, Doug Pitt (CWFC); Murray Woods (SSIS) > linked with work in AB, QC, and NL. Predict wood density, MOE, MOR?

53 The business of managing our forests depends on sound forest inventory We need a spatial inventory, and we need it now! Thank you!

54 The business of managing our forests depends on sound forest inventory We need a spatial inventory, and we need it now! murray.woods@ontario.ca Thank you! dpitt@nrcan.gc.ca

55 What is Individual Tree Crown classification? Background François Gougeon and Don Leckie, CFS Semi-automated image analysis based on digital imagery Identify crown edges by following shadows Species training & automated classification of each crown Softcopy Interpretation/Training

56 Two Forests ~ 2M ha Northeastern Ontario HEARST An operational example ADS4 imagery; provincial coverage ROMEO MALETTE Parameter Romeo Malette Hearst Sensor DFC ADS4 Product Mono R,G,B Stereo R,G,B, CIR Resolution (GSD) 3 cm 25 cm pan; 3 cm colour DSM none 5 m SGM no yes