S. E. Franklin

Incorporating texture into classification of forest species composition from airborne multispectral images

Although research with digital airborne remote sensing data has been undertaken in different ecoregions to classify forested areas, the potential role of such imagery in deriving information to assist forest management has not yet been fully defined. The objective of this study was to determine the extent that the addition of texture could improve spectral classification of high spatial resolution images (pixel size 1m). These images represented pure and mixed wood forest stands from ecoregions in Alberta and New Brunswick, Canada. This study employed a judicious, selective application of texture to stands within a hierarchical classification framework. In Alberta, the addition of texture made a modest improvement in classification accuracy from 60% to 65%. In New Brunswick, the application of texture to selected land cover types resulted in an overall 12% improvement in classification accuracy. The addition of image texture increased classification accuracy for high spatial detail imagery relative to low spatial detail imagery. Incorporating texture into classification also improved classification accuracies for hardwood stands more so than for softwood stands, but greater attention to stand structure and composition will be needed in future work. Classification accuracies on the order of 60-65% were achieved with simple texture derivatives, maximum likelihood decision rules and conventional classification methods.

Texture analysis of IKONOS panchromatic data for Douglas-fir forest age class separability in British Columbia

This Letter presents the results of textural separability tests obtained by first- and second-order texture methods on different aged Douglas-fir stands in IKONOS panchromatic imagery acquired 3 June 2000 over the Sooke River watershed in British Columbia. The effects of different measures and window sizes on the textural separability are discussed. Small windows sizes were not as effective in separating stands as larger windows sizes. Second-order (spatial co-occurrence homogeneity) texture values were the most effective in distinguishing between age classes. A first-order (variance) texture measure, though still useful, provided less separability.

Mountain Pine Beetle Red-Attack Forest Damage Classification Using Stratified Landsat TM Data in British Columbia, Canada

ply of suitable pines is exhausted. Effective management of the The identification and classification of mountain pine beetle, mountain pine beetle is dependent upon rapid and accurate Dentroctonus ponderosa (Hopkins), red-attack damage patterns detection of population stage and trend (i.e., increasing or in a mature lodgepole pine (Pinus contorta) forest located in decreasing). Control or suppression is only feasible for the Fort St. James Forest District, British Columbia, was endemic or incipient populations (e.g., Carroll and Linton, accomplished using 1999 Landsat TM satellite imagery, 1999 2002). Most beetle detection programs involve a traditional mountain pine beetle field and aerial survey point data, and approach based on aerial surveys. In this approach, an observer GIS forest inventory data. Unrelated variance in the observed views the forest canopy in a fixed-wing aircraft and looks for spectral response at mountain pine beetle field and aerial signs of attacked trees, that is, dying trees whose foliage is turnsurvey points was reduced following image stratification with ing from green to red. The boundaries of foliage reddening are the GIS forest inventory data and removal of other factors then mentally averaged and delineated onto a sketch map. uncharacteristic of red-attack damage. Locations of known Ground surveyors are often used to verify the cause of the dismountain pine beetle infestation were used to train a maxi- turbance and assess the severity of forest damage. The drawmum-likelihood algorithm; overall classification accuracy was backs of using these surveying methods are the high operation 73 percent, based on an assessment of 360 independent val- costs and many hours of manpower required. Also, mountain idation points. If local stand variability is reduced prior to pine beetles often spread and colonize new areas before an signature generation, accuracies and map products can be infested area has been completely surveyed by ground or air. useful for those involved in active forest management decision- Mountain pine beetle attacks may first create a pre-visual making regarding mountain pine beetle infestations.

Estimation of forest Leaf Area Index using remote sensing and GIS data for modelling net primary production

Ecosystem models can be used to estimate potential net primary production (pNPP) using GIS data, and remote sensing input of actual forest leaf area to such models can provide estimates of current actual net primary production (aNPP) . Comparisons of pNPP and aNPP for a given site or regional landscape can be used to identify forest stands for different management treatments, and may provide new information on wildlife habitat, forest diversity and growth characteristics. Leaf area estimates may be obtained from satellite imagery through correlation with physiologically-based vegetation indices such as the Normalized Difference Vegetation Index (NDVI). However, in areas with high Leaf Area Index (LAI), vegetation indices usually saturate at leaf areas greater than about 4. In predominantly deciduous (hardwood) and mixedwood stands remote sensing estimates may be influenced by understory and other factors. We examined digital Landsat TM imagery and GIS data in the Fundy Model Forest of southeastern New Bru...

Empirical relations between Landsat TM spectral response and forest stands near Fort Simpson, Northwest Territories, Canada

Empirical relationships between forest stand variables, such as age and crown closure, and spectral response measured by the Landsat Thematic Mapper (TM) satellite sensor have long been suggested as an information source to support forest inventories in many regions of the world. Using regression and correlation techniques, the authors have identified the form and strength of these empirical relationships for a sample of forest stands near Fort Simpson, Northwest Territories. Models were strongest for pioneer forest species such as jack pine and trembling aspen as these relationships were characterized by reasonably consistent changes in stand structure and composition. White spruce, a secondary successional species, produced the statistically weakest models from Landsat TM spectral response patterns for all stand variables with the exception of crown closure. The authors attribute the differences in model strength to variable trends in stem growth and stand structural changes caused by differing successional pathways for white spruce stands in this region.

A Disturbance-Inventory Framework for Flexible and Reliable Landscape Monitoring

Remote sensing plays a key role in landscape monitoring, but our handling of these data in a multi-temporal time series is not yet fully developed. Of particular concern is the presence of spatial and thematic errors in independently created maps that distort measures of landscape pattern and constrain the reliability of change analysis. In addition, there is a need to incorporate continuous attributes of cover gradients for flexible map representations that support a variety of applications. In this paper, we present a framework for generating temporally and categorically dynamic land-cover maps that provide such a reliable and adaptable foundation. The centerpiece is a spatio-temporal disturbance-inventory database, created through semi-automated change detection and conditioned with boundary-matching procedures, which can be used to backdate and update both continuous and categorical reference maps. We demonstrate our approach using multi-annual Landsat imagery from a forested region in west-central Alberta, Canada, between the years 1998 and 2005.

Spatial and spectral variability due to Landsat image acquisition date: the impact on image classification

Large area land cover mapping with Landsat data is affected by differing image spectral characteristics caused by factors such as solar, radiometric, atmospheric, and phenologic conditions. For mapping land cover, especially forested land cover with single date satellite imagery, acquisition of imagery is ideally undertaken during the peak growing periods during the summer months. Yet, during the summer months cloud cover may preclude the collection of suitable imagery. To enable the collection of imagery to compile a large area coverage, the range of acceptable acquisition dates must be extended from the optimal mid-summer period. The question is how far can the acquisition dates range before the spectral information in the imagery is no longer appropriate for land cover mapping. To address this issue we have classified Landsat imagery representing a range of within year and between year dates and have assessed the variability in spatial features developed and the spectral information contained in the imagery. To ensure consistent radiometric characteristics between scenes we applied a top of atmosphere radiometric correction and a dark target subtraction. Our data set consists of 7 images. To assess within year variability we obtained 5 images over a range of dates within 1999.