François A. Gougeon

Combined high-density lidar and multispectral imagery for individual tree crown analysis

Lidar technology has reached a point where ground and forest canopy elevation models can be produced at high spatial resolution. Individual tree crown isolation and classification methods are developing rapidly for multispectral imagery. Analysis of multispectral imagery, however, does not readily provide tree height information and lidar data alone cannot provide species and health attributes. The combination of lidar and multispectral data at the individual tree level may provide a very useful forest inventory tool. A valley following approach to individual tree isolation was applied to both high resolution digital frame camera imagery and a canopy height model (CHM) created from high-density lidar data over a test site of even aged (55 years old) Douglas-fir plots of varying densities (300, 500, and 725 stems/ha) on the west coast of Canada. Tree height was determined from the laser data within the automated crown delineations. Automated tree isolations of the multispectral imagery achieved 80%‐90% good correspondence with the ground reference tree delineations based on ground data. However, for the more open plot there were serious commission errors (false trees isolated) mostly related to sunlit ground vegetation. These were successfully reduced by applying a height filter to the isolations based on the lidar data. Isolations from the lidar data produced good isolations with few commission errors but poorer crown outline delineations especially for the densest plot. There is a complimentarity in the two data sources that will help in tree isolation. Heights of the automated isolations were consistently underestimated versus ground reference trees with an average error of 1.3 m. Further work is needed to test and develop tools and capabilities, but there is an effective synergy of the two high resolution data sources for providing needed forest inventory information.

Stand delineation and composition estimation using semi-automated individual tree crown analysis

Abstract Stand delineation and species composition estimation are cornerstones of forest inventory mapping and key elements to forest management decision making. Improved mapping techniques are constantly being sought in terms of speed, consistency, accuracy, level of detail, and overall effectiveness. Semi-automated analysis of high-resolution imagery at the individual tree crown level may offer such benefits. Methods, however, need to be developed and tested under a variety of forest conditions. High-resolution (60 cm) multispectral airborne imagery was acquired over a predominantly young conifer forest and plantation test area on the west coast of Canada. Automated tree isolation algorithms were applied to the data in order to delineate tree crowns or clusters of crowns. An object-oriented single tree classification was conducted using a maximum likelihood classifier. Stands of similar species composition, closure, and stem density were defined through a sequence that first generated images of these parameters from the automated delineation and classification, used these as input to an unsupervised classification, and then filtered and smoothed the resulting classification clusters. Because of the dense nature of the stands and small crowns on the site, the isolation process often delineated clusters of several trees. Species classification accuracy was determined by comparing the average stand composition from the automated technique to that derived from ground transects or plots. Species classification was good, with average composition error (difference between field measured and automated composition) over all 16 test stands being 7.25%. Most errors for individual species in stands were below 20%, but a few were up to 30%. The automatically generated stand boundaries mimicked well those of known plantation and interpreted inventory boundaries. The automated technique created a few larger stands and some additional small stands in areas of complex forest structure. Overall, for the young fairly uniform stands of the site, both stand delineation and species composition estimation were of a quality suitable for operational use in inventory and forest management. Further development and testing is needed to extend results to situations covering large areas, multiple flight lines, varied topography, and different forest conditions.

The Individual Tree Crown Approach Applied to Ikonos Images of a Coniferous Plantation Area

In forestry, the availability of high spatial resolution (� 1 m/pixel) imagery from new earth observation satellites like Ikonos favours a shift in the image analysis paradigm from a pixel-based approach towards one dealing directly with the essential structuring element of such images: the individual tree crown (ITC). This paper gives an initial assessment of the effects of 1 m and 4 m/pixel spatial resolutions (panchromatic and multispectral bands, respectively) on the detection, delineation, and classification of the individual tree crowns seen in Ikonos images. Winter and summer Ikonos images of the Hudson plantation of the Petawawa Research Forest, Ontario, Canada were analyzed. The panchromatic images were resampled to 0.5 m/pixel and then smoothed using a 3 � 3 kernel mean filter. A valleyfollowing algorithm and rule-based isolation module were applied to delineate the individual tree crowns. Local maxima within a moving 3 � 3 window (i.e., Tree Tops) were also extracted from the smoothed images for comparison. Crown delineation and detection results were summarised and compared with field tree counts. Overall, the ITC delineation and the local maxima approaches led to tree counts that were on average 15 percent off for both seasons. Visual inspection reveals delineation of clusters of two or three crowns as a common source of error. Crown-based species spectral signatures were generated for six classes representing conifer species, plus a hardwood class and a shrub class. After the ITC-based classification, classification accuracy was ascertained using separate test areas of known species. The overall accuracy was 59 percent. Important confusion exists between red and white spruces, and mature versus immature white pines, but post-classification regroupings into single spruce and white pine classes led to an overall accuracy of 67 percent.

Issues in species classification of trees in old growth conifer stands

Old growth temperate conifer forest canopies are composed of assemblages of tree crowns that vary by species, height, size, and intercrown distance. The challenge this complexity presents to species classification is formidable. In this paper we describe the exploration of spectral properties of old growth tree crowns as captured on two independent acquisitions of 0.7 m ground resolution compact airborne spectrographic imager (CASI) airborne multispectral imagery. Underlying spectral separability is examined, and classifications of manually delineated crowns are compared against field-surveyed ground truth. Technical issues and solutions addressing individual tree species classification of old growth conifer stands are discussed. The study site is a western hemlock, amabilis fir, and red cedar dominated old growth area on Vancouver Island on the west coast of Canada. Within-species spectral variability is large because of illumination and view-angle conditions, openness of trees, natural variability, shadowing effects, and a range of crown health. As well, spectral differences between species are not large. An object-oriented illumination and view-angle correction was effective at reducing the effect of view angle on spectral variability. Radiometric normalization between imagery from flight lines of the same site and time period was successful, but normalization with data from other sites and days was not. The use of spectral samples from sunlit areas of the tree crowns (mean-lit signature) produced the best spectral separability and species classification. Because of the wide within-species variability and spectral overlap among species, it was also found useful to create internal subclasses within a species (e.g., normal and bright crowns). It was not feasible to consistently classify species of shaded crowns or stressed trees, and it was necessary to create overall shaded tree and unhealthy classes to prevent these trees from corrupting the final species classification. Classification results also depend on the visibility of trees in the imagery. This was demonstrated by different visibility and shade conditions of trees between the two image dates. This effect is particularly strong in old growth stands because of variations in stem density and spacing and the range of tree heights and sizes. Old growth stands will have shaded, unhealthy, and visually or spectrally unusual trees. Excluding these and considering species classification of manually delineated trees of the normal and bright spectral classes, modest success was achieved, in the order of 78% accuracy. Hemlock and balsam were confused, and cedar classification was mostly confused by the presence of unhealthy trees. If speciation of shaded and unhealthy trees was required, overall species classification was weak. It was shown, however, that shaded and unhealthy trees could be identified well using a classification with shaded and unhealthy classes. Species classification of the remaining trees, including the unusual trees, was 67% for the 1996 imagery and 77% for the 1998 imagery. Despite the difficulties in classifying species in an old growth environment, practical solutions to issues are available and viable spectral classifications are possible. Fully automated species isolation and classification add further complications, however, and new approaches beyond simple spectral techniques need to be explored.

Comparison of six individual tree crown detection algorithms evaluated under varying forest conditions

In this article, six individual tree crown (ITC) detection/delineation algorithms are evaluated, using an image data set containing six diverse forest types at different geographical locations in three European countries. The algorithms use fundamentally different techniques, including local maxima detection, valley following (VF), region-growing (RG), template matching (TM), scale-space (SS) theory and techniques based on stochastic frameworks. The structurally complexity of the forests in the aerial images used ranges from a homogeneous plantation and an area with isolated tree crowns to an extremely dense deciduous forest type. None of the algorithms alone could successfully analyse all different cases. The study shows that it is important to partition the imagery into homogeneous forest stands prior to the application of individual tree detection algorithms. It furthermore suggests a need for a common, publicly available suite of test images and common test procedures for evaluation of individual tree detection/delineation algorithms. Finally, it highlights that, for complex forest types, monoscopic images are insufficient for consistent tree crown detection, even by human interpreters.

Forest inventory stand height estimates from very high spatial resolution satellite imagery calibrated with lidar plots

Many areas of forest across northern Canada are challenging to monitor on a regular basis as a result of their large extent and remoteness. Although no forest inventory data typically exist for these northern areas, detailed and timely forest information for these areas is required to support national and international reporting obligations. We developed and tested a sample-based approach that could be used to estimate forest stand height in these remote forests using panchromatic Very High Spatial Resolution (VHSR, < 1 m) optical imagery and light detection and ranging (lidar) data. Using a study area in central British Columbia, Canada, to test our approach, we compared four different methods for estimating stand height using stand-level and crown-level metrics generated from the VHSR imagery. ‘Lidar plots’ (voxel-based samples of lidar data) are used for calibration and validation of the VHSR-based stand height estimates, similar to the way that field plots are used to calibrate photogrammetric estimates of stand height in a conventional forest inventory or to make empirical attribute estimates from multispectral digital remotely sensed data. A k-nearest neighbours (k-NN) method provided the best estimate of mean stand height (R 2 = 0.69; RMSE = 2.3 m, RMSE normalized by the mean value of the estimates (RMSE-%) = 21) compared with linear regression, random forests, and regression tree methods. The approach presented herein demonstrates the potential of VHSR panchromatic imagery and lidar to provide robust and representative estimates of stand height in remote forest areas where conventional forest inventory approaches are either too costly or are not logistically feasible. While further evaluation of the methods is required to generalize these results over Canada to provide robust and representative estimation, VHSR and lidar data provide an opportunity for monitoring in areas for which there is no detailed forest inventory information available.

Automated Individual Tree Isolation on High-Resolution Imagery: Possible Methods for Breaking Isolations Involving Multiple Trees

One of the key problems in automated individual tree crown delineation, whether from multispectral or lidar data, is the grouping of several trees into a single tree crown outline (isol). Using airborne multispectral imagery, we explored four approaches to breaking such isols into multiple crowns: “core,” “tree top,” “template matching,” and “basin” breaks. Core breaks are made using only isol shape and morphological primitives. Tree top and template matching breaks utilize image maxima and pattern, and watershed drainage basins form the basis of basin breaks. The effectiveness of each of the four break types was assessed against the presence and position of the true boundary between multiple tree crowns and with reference to original isol shape. There was correspondence and differences between breaks of different types. A set of rules was developed to choose a single break when there was positional correspondence of several break types. The rules were based on isol shape type and the break types present. Despite being a complex and difficult issue, it was shown that the concept of identifying poor delineations, recognizing them as cases of multiple trees, and remediating the crown delineations is viable and worthy of further development.

Production of a Large-Area Individual Tree Species Map for Forest Inventory in a Complex Forest Setting and Lessons Learned

Abstract A large-area trial of automated individual tree crown isolation and species classification on a 228-km2 site in central-eastern Ontario using 40 cm multispectral imagery gave insight into the complications and effectiveness of such approaches when applied to a complex mixed forest setting. Tree isolation was reasonably effective with few omissions and 77% of automated tree isolations (isols) were considered a good correspondence to ground truth delineations. There were issues with grouping several trees within 1 delineation, poor isols at the edges of stands and minor splitting (multiple isols per tree). Spectral characteristics of 18 species indicated considerable variability within species and overlap of signatures. To circumvent this, several spectral subclasses were created for certain species and the site was divided into broad and localized zones, which for ecological reasons had a reduced number of species. Mapping accuracy and classification accuracy of manually delineated trees for a main classification zone of 14 species, both typically ranged from 40% to 85%. For a simpler zone of 8 species, manual tree class accuracy averaged 76%. Possible improvements are discussed. Regardless of sophistication of techniques used, production of individual tree species maps in complex forests will require judicious use of human judgment and intervention.

Recognition and Possible Remediation of Automated Tree Delineations with Multiple Isolations per Tree (Split Cases) on High-Resolution Imagery

Abstract Individual tree crown analysis from high-resolution imagery is gaining greater use in forest applications. Automated crown delineations (isols) that are poor can cause errors in species classification and inventory estimates. This study explores the issue of recognizing split cases (e.g., tree crowns oversegmented into several isols) and demonstrates 3 remediation procedures to improve delineations. Several methods for identifying split cases are proposed, but a conceptual framework for a template-matching approach is developed further. Candidate split cases are identified where there is a good match of a template model representing the appearance of trees with the imagery, and several isols are within the template. Candidates are further analyzed through evidence such as isol shape, species class, and match of templates centered on each isol. Procedures were demonstrated with a typical individual crown isolation on 40 cm multispectral imagery of a mixed species forest in northeastern Ontario. The process showed useful effectiveness in improving the isolation, with expected omission rates of 15%–20% and 25%–30% false alarms. Overall, almost all true split cases recognized had improved crown delineations. The work shows that approaches for recognizing and remediating split cases are possible, but will have to be complex and consider multiple evidence.

Fusion of High-Resolution Satellite and Lidar Data for Individual Tree Recognition

A major shift in the forest inventory and management paradigm toward the use of semi-automated analysis realized on an individual tree crown basis has been made possible by recent developments in high-resolution remote sensing. This paper discusses issues related to the fusion of high-resolution satellite imagery and LIDAR (light detection and ranging) data and their application in the classification of individual trees for precision forest management. The proposed methodological approach consists in the combination of spatial filtering object detection and reconstruction methods with a rule-based individual tree crown (ITC) system. Examples using QuickBird imagery combined with LIDAR data from an Alberta site (both boreal and mixed forest) demonstrate the advantages of the proposed fusion approach