Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Faculty of Forest Sciences
EXPERIENCES AND POSSIBILITIES OF ALS
BASED FOREST INVENTORY IN FINLAND
Professor Matti Maltamo
Faculty of Forest Sciences
Contents n Forests of Finland
n Forest inventories in Finland
n Experiences of ALS based forest inventory studies in Finland • Single tree detection • Area based methods
n Some calculation experiments of ALS based forest inventory approaches
n Conclusions
Faculty of Forest Sciences
Forests of Finland n Boreal forests
n Low number of tree species but often mixed stands
n Pine or spruce dominated
n Small trees, high density
n No shortrotation plantations
Faculty of Forest Sciences
National Forest Inventory (NFI)
n Field sample plots • Cannot be replaced by ALS • The role of terrestrial lasers? • Updating of permanent sample
plots? n Multisource NFI
• The role of ALS as auxiliary information
• National Laser Scanning n Coordinated by FFRI Copyright by Blom
Faculty of Forest Sciences
Inventory by compartments n For forest planning on small scale n Current field measurement based method ”slow”, expensive and inaccurate
• Aerial photographs • Field visits • Species specific stand mean
variables
n High potential for ALS based approaches
n ALS inventory almost in practise in Finland
Faculty of Forest Sciences
Specific inventories
n Wood procurement planning/ preharvest inventory n Nature conservation areas
• Target areas can be very small or large
• Detailed information • High point density
needed?
Faculty of Forest Sciences
ALS experiences in Finland n General
• History • Research started by 1990’s • Juha Hyyppä, FGI • Single tree detection
algorithms • DTM algorithms
• Research during last 3 years
Faculty of Forest Sciences
Single tree detection
Faculty of Forest Sciences
Change detection n Tree height growth at tree and plot level, stem volume growth and fallen trees by using multitemporal ALS data n Methods for tree height growth detection: (i) the difference of highest z value, (ii) difference between DSMs of tree tops and, (iii) difference of 85, 90 and 95% quantiles of the height histograms corresponding to a crown
n Yu, X., Hyyppä, J., Kaartinen, H. & Maltamo, M., 2004. Automatic Detection of Harvested Trees and Determination of Forest Growth using Airborne Laser Scanning. Remote Sensing of Environment, 90, pp. 451462. n Yu, X., Hyyppä, J., Kukko, A., Maltamo, M. & Kaartinen, H., 2006. Change detection techniques for canopy height growth measurements using airborne laser scanner data. Photogrammetric Engineering & Remote Sensing, 72, pp. 13391348
Faculty of Forest Sciences
Tree species classification
n Coniferous – deciduous n Tree level height quantiles and other variables
n Intensity n 3D textural variables n Leaf onoff conditions
Faculty of Forest Sciences
Identification of large aspens n Classification between ecologically important large aspens and other deciduous trees n Single tree recognition of ALS data
• Watershed segmentation
n Linear discriminant analysis • proportion of vegetation hits • standard deviation of pulse heights • accumulated intensity on 90th percentile • proportions of laser points reflected on 95th and
40th height percentiles
n Classification accuracy was about 79%. n Säynäjoki, R., 2007. Classification of aspen and other deciduous trees using high resolution remote sensing data. University of Joensuu, Faculty of Forest Sciences, master’s thesis.
Faculty of Forest Sciences
Prediction of tree variables n Tree volume, diameter, crown height, profile n Tree level height quantiles and other variables n Vector models n Intensity
Faculty of Forest Sciences
Crown height prediction
n Comparison of ALS based variables and field measurements in the prediction of crown height of boreal tree species
n ALS variables tree height and crown area, plot level height quantiles
n Accuracy as good as in the case of using field measurements of tree height and diameter
n Maltamo, M., Hyyppä, J. & Malinen, J., 2006b. A comparative study of the use of laser scanner data and field measurements in the prediction of crown height in boreal forests. Scandinavian Journal of Forest Research, 21, pp. 231 238
Faculty of Forest Sciences
Forest inventory
n No published studies in Finland about the prediction of general end products of forest resources (tree species specific estimates, timber sortiments per hectare) on large scale during recent years n Presentation by Ph.D. Ilkka Korpela in SILVILASER2007 n Specific inventories
• Pre harvest inventory
Faculty of Forest Sciences
Timber procurement planning n Estimation of timber sortiments
in a marked stand: (i) Lidarbased individual tree
detection (ii) Systematic field plot sampling
data (iii) Field inventory by compartments
and (iv) Area based canopy height
distribution approach.
G F G F
G F G F G F G F
G F
G F G F
G F G F
G F G F G F G F G F
G F G F G F G F
G F G F G F G F G F G F
G F G F G F
G F G F G F
G F
G F G F G F G F G F G F G F G F G F
G F G F G F
G F G F
G F G F
G F G F
G F G F G F G F G F G F
G F G F G F
G F
G F G F G F
G F G F
G F
G F G F
G F G F G F G F G F
G F
G F G F G F
G F G F
G F G F
G F G F G F G F
G F G F
G F G F G F G F
G F G F
G F G F G F
G F G F G F
G F G F G F G F
G F G F
G F G F G F G F G F
G F G F G F G F G F
0 10 20 30 40 5 Meters
Faculty of Forest Sciences
Timber procurement planning
n Ground truth data harvester measurements n The comparison of the methods was based on bucking simulations n Considerable advantage of lidar based single tree detection procedure compared to other studied methods in producing preharvest measurement information in this example marked stand
n Peuhkurinen, J., Maltamo, M., Malinen, J., Pitkänen, J. & Packalén, P., 2007. Preharvest measurement of marked stand using airborne laser scanning. Forest Science, In press
Diameter distributions produced with comparable methods
0
50
100
150
200
250
300
350
9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63
Diameter class
Num
ber o
f trees
Harvester measured
Laser based single tree detection
Theoretical (inventory by compartments)
Faculty of Forest Sciences
Area based methods
Faculty of Forest Sciences
Area based methods (canopy height distribution)
n Research started in 2004 n First studies considered regression models based prediction of stand total characteristics n Comparison of ALS and other methods n Young stands n Non parametric models
Faculty of Forest Sciences
One approach to predict species specific growing stock variables
n Plot level reference data
n Combine aerial photos and ALS data
n ALS: canopy height distribution variables
n Aerial photos: intensity and texture
n KMSN estimation by tree species for each grid
n Stand volume, basal area, stem number, mean diameter and height
n Compose stands from grids
Faculty of Forest Sciences
Species specific growing stock variables
n Measured versus predicted volumes by tree species and total volume n Accuracy by means of RMSE better than in conventional field measurements based inventory at least for stand main tree species
Faculty of Forest Sciences
Species specific growing stock variables n An example of the prediction of Norway spruce by using grid on large scale n Packalén, P. & Maltamo, M., 2006. Predicting the volume by tree species using airborne laser scanning and aerial photographs. Forest Science, 52, pp. 611 622. n Packalén, P. & Maltamo, M., 2007. The kMSN method in the prediction of species specific stand attributes using airborne laser scanning and aerial photographs. Remote Sensing of Environment, 109, pp. 328341
Faculty of Forest Sciences
Diameter distribution estimates
n Instead of stand mean and sum characteristics stem diameter distribution is usually of primary interest
n In general diameter distributions can be predicted by using ALS data: • Applying ALS based stand variables and existing diameter distribution parameter models or parameter recovery methods (presentation of Ph.D. Lauri Mehtätalo in SILVILASER2007)
• Predicting parameters or diameter distributions directly by using ALS variables
• Diameter model estimate in single tree detection
Faculty of Forest Sciences
Diameter distribution estimates an example
0 50 100 150 200 250 300 350 400 450 500
1 3 5 7 9 11 13 15 17 19 21 23
Diameter at breast height, cm
Num
ber of stems, ha
1
n Prediction of Weibull parameters and characteristics of growing stock simultaneously by using ALS data n Forming of diameter distribution n Calibration of diameter distribution estimate using predicted stand characteristics by means of calibration estimation n No need for basal area diameter distributions n Maltamo, M., Suvanto, A., and Packalén, P. 2007. Comparison of basal area and stem frequency diameter distribution modelling using airborne laser scanner data and calibration estimation. For. Ecol. Manage. 247: 26–34.
n Species specific distributions still to come!
Faculty of Forest Sciences
The usability of NFI sample plots as ground truth n A considerable amount of costs comes from the measurements of reference sample plot n The use of existing data? n NFI angle count sample plots n Both simulations and actual calculations showed that the accuracy of derived stand variables is close to that of fixed sized plots n However • Systematically located plots in stand borders • Sparse NFI data, large inventory area needed • GPS location of sample plots
n Maltamo, M., Korhonen, K.T., Packalén, P., Mehtätalo, L. & Suvanto A., 2007. A test on the usability of truncated angle count sample plots as ground truth in airborne laser scanning based forest inventory. Forestry, 80, pp. 7381.
Faculty of Forest Sciences
Prediction of coarse woody debris n ALS data was used to predict dead wood volume
• Standard deviation of laser heights, intensity
n Conservation area n Approach based on canopy gaps and growing stock variation n Accuracy was satisfactory n In managed forests accuracy worse, but still was able to classify plots n Pesonen, A, Maltamo, M., Eerikäinen, K. & Packalèn, P., 2007. Airborne laser scanningbased prediction of coarse woody debris volumes in a conservation area n Kotamaa, E., 2007. Prediction of CWD in managed forests by using ALS. University of Joensuu, Candidate’s thesis
Faculty of Forest Sciences
Calculation experiments n Comparison of single tree detection and area based canopy height distribution approaches n Spatial data provided by ALS
Faculty of Forest Sciences
Comparison of single tree detection and canopy height distribution approaches
n Simulation of single tree detection approach by using ground truth data • All trees were found • Tree heights without bias • All tree species detected correctly • Tree diameter predicted with tree height by using local model of 472 field sample plots
n The accuracy in volume about 23% in terms of RMSE n Corresponding figures for canopy height distribution approaches of actual studies by using same field data 1521%
Faculty of Forest Sciences
Comparison of single tree detection and canopy height distribution approaches
n Corresponding results were obtained for diameter distribution and timber sortiments n Although many restrictions in this simulation, it is expected that single tree detection will not provide more accurate results than canopy height distribution approach in boreal forests n Tree diameter prediction very difficult by using vertical characteristics • e.g. Site quality • Silvicultural history • Tree position in a stand n Even standwise field calibration needed for accurate general forest resources results?
Faculty of Forest Sciences
Spatial data provided by ALS – single tree detection
0
0.05
0.1
0.15
0.2
0.25
0.3
0 5 10 15 20 25 30 Tree height, m
Com
petition inde
x
ALS field
n Competition indices n Recognised trees and their heights n Neighbouring trees n Usability in growth and other spatial tree models n Excellent data compared to current field inventory system
Faculty of Forest Sciences
Spatial data provided by ALS – area based methods
0
5
10
15
20
25
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Basal area, m 2 ha 1
Relative prop
ortio
n, % Field measurement
ALS estimate
n Within stand variation n Basal area by using grid n Accurate ground truth data n Needed e.g., for timing of silvicultural operations n Excellent data compared to current field inventory system
Faculty of Forest Sciences
Conclusions
n ALS based applications are at the moment of very high interest in Finland with respect to forest inventory n Canopy height distribution approach already in practise in Finland n In single tree detection whole model chain should be taken into consideration • Tree detection, tree species AND tree diameter prediction n Canopy height distribution for general forest resource information, single tree detection for specific mature stands? n Many other forest characteristics are examined as well n This presentation considered forest aspect, technical aspect (lidar data types, algorithm combinations, etc.) as well primarily important
Faculty of Forest Sciences
Thank you