François A. Gougeon,
Benoit A. St-Onge,
Mike Wulder,
and Donald G. Leckie.
Synergy of Airborne Laser Altimetry and Digital Videography for Individual Tree Crown Delineation.
.
Abstract: |
The obvious advantage of LIDAR data for forestry is as replacement for conventional stereoscopic methods or field measurements of dominant trees to obtain forest stand heights, an important factor in the inference of numerous other forest stand parameters (e.g., volume, biomass). However, when the density of LIDAR data becomes high enough, one can think in terms of individual tree crown heights. An earlier paper (St- Onge, 2000) reported good R2 (0.90) between laser-predicted heights and heights from field measurements of 36 trees. Part of the same dataset over the Lake Duparquet Research Forest, Québec (79.3 W, 48.5 N) is being used in this study. It consists in a multispectral video image and a LIDAR canopy model coregistered at 50 cm/pixel. However, the laser altimeter mean distance between two hits in about 1.5m. An ideal forest inventory system could incorporate individual tree crown (ITC) delineation and species recognition from multispectral imagery (Gougeon, 1999) with ITC-based height measurements from LIDAR data to produce more precise, accurate and timely ITC-based forest inventories. This could also permit the inference of volume and biomass to be calculated on an ITC-basis and would help in studying the height and crown diameter distributions for ecological or forest productivity studies. This study examines the possible synergy between airborne laser altimeter data and digital video imagery at two processing levels within an automatic image analysis system: (a) the elimination of non-forested or poorly forested areas from analysis, and (b), the possible improvement of individual tree crown delineation. The obvious synergistic effects of knowing height and species on an individual tree basis have been assumed and left to be more fully demonstrated in later works. Used at an appropriate spatial resolution, the individual tree crown delineation algorithm based on following valley of shade between the crowns works generally well on dense to medium dense coniferous forests. When the forest is more open, as is the case here, pre-processing can often be used to eliminate non-forested areas. A combination of masks generated from multispectral rules and by selecting a minimum height from the LIDAR data led to a very good separation of the forested areas and even of individual trees in open fields. The ITC delineation process was then applied to the unmasked areas, first on a smoothed version of the near infrared image and then, on a smoothed version of the LIDAR height image. Both cases produced numerous tree clusters rather than individual tree crowns, but for different reasons. With the video data, crown delineation is hampered by the lack of shade between tree crowns in a direction normal to the illumination angle. With the LIDAR data, crown delineation is hampered by the lack of effective spatial resolution. A post-processing combination of both results led to superior crown delineation, with very few tree clusters. |
@Article{gougeon01,
author = {François A. Gougeon and Benoit A. St-Onge and Mike Wulder and Donald G. Leckie},
title = {Synergy of Airborne Laser Altimetry and Digital Videography for Individual Tree Crown Delineation},
journal = {},
year = {},
volume = {},
pages = {},
number = {},
url = {},
keyword = {},
abstract = {The obvious advantage of LIDAR data for forestry is as replacement for conventional stereoscopic methods or field measurements of dominant trees to obtain forest stand heights, an important factor in the inference of numerous other forest stand parameters (e.g., volume, biomass). However, when the density of LIDAR data becomes high enough, one can think in terms of individual tree crown heights. An earlier paper (St- Onge, 2000) reported good R2 (0.90) between laser-predicted heights and heights from field measurements of 36 trees. Part of the same dataset over the Lake Duparquet Research Forest, Québec (79.3 W, 48.5 N) is being used in this study. It consists in a multispectral video image and a LIDAR canopy model coregistered at 50 cm/pixel. However, the laser altimeter mean distance between two hits in about 1.5m. An ideal forest inventory system could incorporate individual tree crown (ITC) delineation and species recognition from multispectral imagery (Gougeon, 1999) with ITC-based height measurements from LIDAR data to produce more precise, accurate and timely ITC-based forest inventories. This could also permit the inference of volume and biomass to be calculated on an ITC-basis and would help in studying the height and crown diameter distributions for ecological or forest productivity studies. This study examines the possible synergy between airborne laser altimeter data and digital video imagery at two processing levels within an automatic image analysis system: (a) the elimination of non-forested or poorly forested areas from analysis, and (b), the possible improvement of individual tree crown delineation. The obvious synergistic effects of knowing height and species on an individual tree basis have been assumed and left to be more fully demonstrated in later works. Used at an appropriate spatial resolution, the individual tree crown delineation algorithm based on following valley of shade between the crowns works generally well on dense to medium dense coniferous forests. When the forest is more open, as is the case here, pre-processing can often be used to eliminate non-forested areas. A combination of masks generated from multispectral rules and by selecting a minimum height from the LIDAR data led to a very good separation of the forested areas and even of individual trees in open fields. The ITC delineation process was then applied to the unmasked areas, first on a smoothed version of the near infrared image and then, on a smoothed version of the LIDAR height image. Both cases produced numerous tree clusters rather than individual tree crowns, but for different reasons. With the video data, crown delineation is hampered by the lack of shade between tree crowns in a direction normal to the illumination angle. With the LIDAR data, crown delineation is hampered by the lack of effective spatial resolution. A post-processing combination of both results led to superior crown delineation, with very few tree clusters.},
}
David Riano,
Erich Meier,
Britta Allgoewer,
Emilio Chuvieco,
and Susan L. Ustin.
Modeling airborne laser scanning data for the spatial generation of critical forest parameters in fire behavior modeling.
rse,
subm..
Keyword(s): Airborne laser scanning,
fuel modeling,
tree height,
crown base height,
surface canopy height,
tree cover,
surface canopy cover,
crown bulk density..
Abstract: |
Methods for using airborne laser scanning to retrieve forest parameters that are critical for fire behavior modeling are presented. A model for the automatic extraction of forest information is demonstrated to provide spatial coverage of the study area, making it possible to produce 3-D inputs to improve fire behavior models. The Toposys I airborne laser system recorded the last return of each footprint (0.30-0.38m) over a 2 km by 190 m flightline. Raw data were transformed into height above the surface, eliminating the effect of terrain on vegetation height, and allowing separation of ground surface and crown heights. Data were defined as ground elevation if heights were less than 0.6 m. A cluster analysis was used to discriminate crown base height, allowing identification of both tree and understory canopy heights. Tree height was defined as the 99 percentile of the tree crown height group, while crown base height was the 1 percentile of the tree crown height group. Tree cover was estimated from the fraction of total tree laser hits relative to the total number of laser hits. Surface canopy height was computed as the 99 percentile of the surface canopy group. Surface canopy cover is equal to the fraction of total surface canopy hits relative to the total number of hits, once the canopy height profile was corrected. Crown bulk density was obtained from foliage biomass estimate and crown volume, using an empirical equation for foliage biomass. Crown volume was estimated as the crown area times the crown height after a correction for mean canopy cover. |
@Article{riano02,
author = {David Riano and Erich Meier and Britta Allgoewer and Emilio Chuvieco and Susan L. Ustin},
title = {Modeling airborne laser scanning data for the spatial generation of critical forest parameters in fire behavior modeling},
journal = rse,
year = {subm.},
volume = {},
pages = {},
number = {},
url = {},
keyword = {Airborne laser scanning, fuel modeling, tree height, crown base height, surface canopy height, tree cover, surface canopy cover, crown bulk density.},
abstract = {Methods for using airborne laser scanning to retrieve forest parameters that are critical for fire behavior modeling are presented. A model for the automatic extraction of forest information is demonstrated to provide spatial coverage of the study area, making it possible to produce 3-D inputs to improve fire behavior models. The Toposys I airborne laser system recorded the last return of each footprint (0.30-0.38m) over a 2 km by 190 m flightline. Raw data were transformed into height above the surface, eliminating the effect of terrain on vegetation height, and allowing separation of ground surface and crown heights. Data were defined as ground elevation if heights were less than 0.6 m. A cluster analysis was used to discriminate crown base height, allowing identification of both tree and understory canopy heights. Tree height was defined as the 99 percentile of the tree crown height group, while crown base height was the 1 percentile of the tree crown height group. Tree cover was estimated from the fraction of total tree laser hits relative to the total number of laser hits. Surface canopy height was computed as the 99 percentile of the surface canopy group. Surface canopy cover is equal to the fraction of total surface canopy hits relative to the total number of hits, once the canopy height profile was corrected. Crown bulk density was obtained from foliage biomass estimate and crown volume, using an empirical equation for foliage biomass. Crown volume was estimated as the crown area times the crown height after a correction for mean canopy cover.},
}