Hyperspectral data processing for repeat detection of small infestations of leafy spurge

Leafy spurge (Euphorbia esula L.) is an invasive plant species in the north central and western U.S. and southern Canada. Idaho has established populations in the north and southeastern regions which are spreading into new sites. This study demonstrates the ability of high resolution hyperspectral imagery to provide high quality data and consistent methods to locate small and low percent canopy cover occurrences of leafy spurge. Locating leafy spurge in its early stages of invasion is critical for land managers in order to prioritize treatment, conservation, and restoration activities. Hyperspectral data were collected in 2002 and 2003 for the study area in southeastern Idaho. The imagery was classified with the Mixture Tuned Matched Filtering (MTMF) algorithm. Although classifications from single date images provided discrimination of leafy spurge at approximately 10% cover in one 3.5 m pixel, for repeatability and consistency purposes, the threshold for leafy spurge discrimination is approximately 40% cover. We hypothesize that georegistration errors, small differences in leafy spurge reflectance, training endmember selection, and image processing and field validation biases between years influence multi-date detection limits. Although hyperspectral imagery is costly, in some situations, the advantages of having reliable and repeatable mapping abilities for discrimination of economically damaging invasive species such as leafy spurge outweigh the image and processing costs.     

Evaluating the potential of multispectral imagery to map multiple stages of tree mortality

Insect outbreaks are major forest disturbances, causing tree mortality across millions of ha in North America. Resultant spatial and temporal patterns of tree mortality can profoundly affect ecosystem structure and function. In this study, we evaluated the classification accuracy of multispectral imagery at different spatial resolutions. We used four-band digital aerial imagery (30-cm spatial resolution and aggregated to coarser resolutions) acquired over lodgepole pine-dominated stands in central Colorado recently attacked by mountain pine beetle. Classes of interest included green trees and multiple stages of post-insect attack tree mortality, including dead trees with red needles (“red-attack”), dead trees without needles (“gray-attack”), and non-forest. The 30-cm resolution image facilitated delineation of trees located in the field, which were used in image classification. We employed a maximum likelihood classifier using the green band, Red-Green Index (RGI), and Normalized Difference Vegetation Index (NDVI). Pixel-level classification accuracies using this imagery were good (overall accuracy of 87%, kappa = 0.84), although misclassification occurred between a) sunlit crowns of live (green) trees and herbaceous vegetation, and b) sunlit crowns of gray- and red-attack trees and bare soil. We explored the capability of coarser resolution imagery, aggregated from the 30-cm resolution to 1.2, 2.4, and 4.2 m, to improve classification accuracy. We found the highest accuracy at the 2.4-m resolution, where reduction in omission and commission errors and increases in overall accuracy (90%) and kappa (0.88) were achieved, and visual inspection indicated improved mapping. Pixels at this resolution included more shadow in forested regions than pixels in finer resolution imagery, thereby reducing forest canopy reflectance and allowing improved separation between forest and non-forest classes, yet were fine enough to resolve individual tree crowns better than the 4.2-m imagery. Our results illustrate that a classification of an image with a spatial resolution similar to the area of a tree crown outperforms that of finer and coarser resolution imagery for mapping tree mortality and non-forest classes. We also demonstrate that multispectral imagery can be used to separate multiple postoutbreak attack stages (i.e., red-attack and gray-attack) from other classes in the image.     

Mapping nonnative plants using hyperspectral imagery

Nonnative plant species are causing enormous ecological and environmental impacts from local to global scale. Remote sensing images have had mixed success in providing spatial information on land cover characteristics to land managers that increase effective management of invasions into native habitats. However, there has been limited evaluation of the use of hyperspectral data and processing techniques for mapping specific invasive species based on their spectral characteristics. This research evaluated three different methods of processing hyperspectral imagery: minimum noise fraction (MNF), continuum removal, and band ratio indices for mapping iceplant (Carpobrotus edulis) and jubata grass (Cortaderia jubata) in California's coastal habitat. Validation with field sampling data showed high mapping accuracies for all methods for identifying presence or absence of iceplant (97%), with the MNF procedure producing the highest accuracy (55%) when the classes were divided into four different densities of iceplant.     

Wetland monitoring using classification trees and SPOT-5 seasonal time series

Multiseason reflectance data from radiometrically and geometrically corrected multispectral SPOT-5 images of 10-m resolution were combined with thorough field campaigns and land cover digitizing using a binary classification tree algorithm to estimate the area of marshes covered with common reeds (Phragmites australis) and submerged macrophytes (Potamogeton pectinatus, P. pusillus, Myriophyllum spicatum, Ruppia maritima, Chara sp.) over an area of 145,000 ha. Accuracy of these models was estimated by cross-validation and by calculating the percentage of correctly classified pixels on the resulting maps. Robustness of this approach was assessed by applying these models to an independent set of images using independent field data for validation. Biophysical parameters of both habitat types were used to interpret the misclassifications. The resulting trees provided a cross-validation accuracy of 98.7% for common reed and 97.4% for submerged macrophytes. Variables discriminating reed marshes from other land covers were the difference in the near-infrared band between March and June, the Optimized Soil Adjusted Vegetation Index of December, and the Normalized Difference Water Index (NDWI) of September. Submerged macrophyte beds were discriminated with the shortwave-infrared band of December, the NDWI of September, the red band of September and the Simple Ratio index of March. Mapping validations provided accuracies of 98.6% (2005) and 98.1% (2006) for common reed, and 86.7% (2005) and 85.9% (2006) for submerged macrophytes. The combination of multispectral and multiseasonal satellite data thus discriminated these wetland vegetation types efficiently. Misclassifications were partly explained by digitizing inaccuracies, and were not related to biophysical parameters for reedbeds. The classification accuracy of submerged macrophytes was influenced by the proportion of plants showing on the water surface, percent cover of submerged species, water turbidity, and salinity. Classification trees applied to time series of SPOT-5 images appear as a powerful and reliable tool for monitoring wetland vegetation experiencing different hydrological regimes even with a small training sample (N = 25) when initially combined with thorough field measurements.     

Mapping two Eucalyptus subgenera using multiple endmember spectral mixture analysis and continuum-removed imaging spectrometry data

Successful discrimination of a variety of natural and urban landscape components has been achieved with remote sensing data using multiple endmember spectral mixture analysis (MESMA). MESMA is a spectral matching algorithm that addresses spectral variability by allowing multiple reference spectra (i.e., endmembers) to represent each material class. However, materials that have a high-degree of spectral similarity between classes, such as similar plant-types or closely related plant species, and large variations in albedo present an ongoing challenge for accurate class discrimination with imaging spectrometry. Continuum removal (CR) analysis may improve class separability by emphasizing individual absorption features across a normalized spectrum. The spectral and structural characteristics common to most Eucalyptus trees make them notoriously difficult to discriminate in closed-canopy forests with imaging spectrometry. We evaluated whether CR applied to hyperspectral remote sensing data improved the performance of MESMA in classifying and mapping nine eucalypt tree species according to the two major Eucalyptus subgenera, Eucalyptus (common name “monocalypt”) and Symphyomyrtus (common name “symphyomyrtle”). Mixed-canopies comprised of monocalypts and symphyomyrtles are common in Australia, although their spatial distribution is not random. The ability to map these functional types on a landscape-scale could provide important information about ecosystem processes, landscape disturbance history and wildlife habitat. We created a spectral library of 229 pixels from 37 symphyomyrtle tree canopies and 406 pixels from 62 monocalypt tree canopies selected from HyMap imagery and verified with field data. Based on these reference data, we achieved overall classification accuracies at the subgenera-level of 75% (Kappa 0.48) for non-CR spectra and 83% (Kappa 0.63) for the CR spectra. We found that continuum-removal improved the classification performance of most endmember-models, although a larger portion of pixels remained unmodeled with the CR spectra (2%) compared to the non-CR spectra (0%). We utilized a new method for model optimization and created maps of monocalypt and symphyomyrtle distribution in our study area based on our best performing endmember-models. Our vegetation maps were largely consistent with our expectations of subgenera distribution based on our knowledge of the region.     

Optimizing unsupervised classifications of remotely sensed imagery with a data-assisted labeling approach

The quality of remotely sensed land use and land cover (LULC) maps is affected by the accuracy of image data classifications. Various efforts have been made in advancing supervised or unsupervised classification methods to increase the repeatability and accuracy of LULC mapping. This study incorporates a data-assisted labeling approach (DALA) into the unsupervised classification of remotely sensed imagery. The DALA-unsupervised classification algorithm consists of three steps: (1) creation of N spectral-class maps using Iterative Self-Organizing Data Analysis Technique Algorithm (ISODATA); (2) development of LULC maps with assistance of reference data; and (3) accuracy assessments of all the LULC maps using independent reference data and selection of one LULC map with the highest accuracy. Classification experiments with a composite image of a Landsat Thematic Mapper (TM) image and an Enhanced Thematic Mapper Plus (ETM+) image suggest that DALA was effective in making unsupervised classification process more objective, automatic, and accurate. A comparison between the DALA-unsupervised classifications and some conventional classifications suggests that the DALA-unsupervised classification algorithm yielded better classification accuracies compared to these conventional approaches. Such a simple, effective approach has not been systematically examined before but has great potential for many applications in the geosciences.     

Boosted decision tree classifications of land cover over Turkey integrating MODIS,climate and topographic data

This study investigates the impact of using different combinations of Moderate Resolution Imaging Spectroradiometer (MODIS) and ancillary datasets on overall and per-class classification accuracies for nine land cover types modified from the classification system of the International Geosphere Biosphere Programme (IGBP). Twelve land cover maps were generated for Turkey using boosted decision trees (BDTs) based on the stepwise addition of 14 explanatory variables derived from a time series of 16-day MODIS composites between 2000 and 2006 (Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and four spectral bands) and ancillary climate and topographic data (minimum and maximum air temperature, precipitation, potential evapotranspiration, aspect, elevation, distance to sea and slope) at 500-m resolution. Evaluation of the 12 BDTs indicated that the BDT built as a function of all the MODIS and climate variables, aspect and elevation produced the highest degree of overall classification accuracy (79.8%) and kappa statistic (0.76) followed by the BDTs that additionally included distance to sea (DtS), and both DtS and slope. Based on an independent validation dataset derived from a pre-existing national forest map and Landsat images of Turkey, the highest overall accuracy (64.7%) and kappa coefficient (0.58) among the 12 land cover maps was achieved by using MODIS-derived NDVI time series only, followed by NDVI and EVI time series combined; NDVI, EVI and four MODIS spectral bands; and the combination of all MODIS and climate data, aspect, elevation and distance to sea, respectively. The largest improvements in producer's accuracies were observed for grasslands (+50%), barrenlands (+46%) and mixed forests (+39%) and in user's accuracies for grasslands (+53%), shrublands (+30%) and mixed forests (+28%), in relation to the lowest producer's accuracy. The results of this study indicate that BDTs can increase the accuracy of land cover classifications at the national scale.     

Automated tree recognition in old growth conifer stands with high resolution digital imagery

Automated individual tree isolation and species determination with high resolution multispectral imagery is becoming a viable forest survey tool. Application to old growth conifer forests offer unique technical issues including high variability in tree size and dominance, strong tree shading and obscuration, and varying ages and states of health. The capabilities of individual tree analysis are examined with two acquisitions of 70-cm resolution CASI imagery over a hemlock, amabilis fir, and cedar dominated old growth site on the west coast of Canada. Trees were delineated using the valley following approach of the Individual Tree Crown (ITC) software suite, classified according to species (hemlock, amabilis fir, and cedar) using object-based spectral classification and tested on a tree-for-tree basis against data derived from ground plots.Tree-for-tree isolation and species classification accuracy assessment, although often sobering, is important for portraying the overall effectiveness of species composition mapping using single tree approaches. This accuracy considers not only how well each tree is classified, but how well each automated isolation represents a true tree and its species. Omissions and commissions need to be included in overall species accuracy assessment. A structure of rules for defining isolation accuracy is developed and used. An example is given of a new approach to accuracy analysis incorporating both isolation and classification results (automated tree recognition) and the issues this presents.The automated tree isolation performed well on those trees that could be visually identified on the imagery using ground measured stem maps (approximately 50-60% of trees had a good match between manual and automated delineations). There were few omissions. Commission errors, i.e., automated isolations not associated with a delineated ground reference tree, were a problem (25%) usually associated with spurious higher intensity areas within shaded regions, which get confused in the process of trying to isolate shaded trees. Difficulty in classifying species was caused by: variability of the spectral signatures of the old growth trees within the same species, tree health, and trees partly or fully shaded by other trees. To accommodate this variability, several signatures were used to represent each species including shaded trees. Species could not be determined for the shaded cases or for the unhealthy trees and therefore two combined classes, a shaded class and unhealthy class with all species included, were used for further analysis. Species classification accuracy of the trees for which there was a good automated isolation match was 72%, 60%, and 40% for the non-shaded healthy hemlock, balsam, and cedar trees for the 1996 data. Equivalent accuracy for the 1998 imagery was 59% for hemlock, 80% for balsam, with only a few cedar trees being well isolated. If all other matches were considered an error in classification, species classification was poor (approximately 45% for balsam and hemlock, 25% for cedar). However, species classification accuracies incorporating the good isolation matches and trees for which there was a match of an isolations and reference tree but the match was not considered good were moderate (60%, 57%, and 38% for hemlock, balsam, and cedar from the 1996 data; 62%, 61%, and 89%, respectively, for the 1998 imagery).Automated tree isolation and species classification of old growth forests is difficult, but nevertheless in this example useful results were obtained.     

Improving tropical forest mapping using multi-date Landsat TM data and pre-classification image smoothing

The present study explores the possibility of using Landsat imagery for mapping tropical forest types with relevance to forest ecosystem services. The central part in the classification process is the use of multi-date image data and pre-classification image smoothing. The study argues that multi-date imagery contains information on phenological and canopy structural properties, and shows how the use of multi-date imagery has a significant impact on classification accuracy. Furthermore, the study shows the value of applying small kernel smoothing filters to reduce in-class spectral variability and enhance between-class spectral separability. Making use of these approaches and a maximum likelihood algorithm, six tropical forest types were classified with an overall accuracy of 90.94%, and with individual forest classes mapped with accuracies above 75.19% (user's accuracy) and above 74.17% (producer's accuracy).     

Mapping reedbed habitats using texture-based classification of QuickBird imagery

Many organisms rely on reedbed habitats for their existence, yet, over the past century there has been a drastic reduction in the area and quality of reedbeds in the UK due to intensified human activities. In order to develop management plans for conserving and expanding this threatened habitat, accurate up-to-date information is needed concerning its current distribution and status. This information is difficult to collect using field surveys because reedbeds exist as small patches that are sparsely distributed across landscapes. Hence, this study was undertaken to develop a methodology for accurately mapping reedbeds using very high resolution QuickBird satellite imagery. The objectives were to determine the optimum combination of textural and spectral measures for mapping reedbeds; to investigate the effect of the spatial resolution of the input data upon classification accuracy; to determine whether the maximum likelihood classifier (MLC) or artificial neural network (ANN) analysis produced the most accurate classification; and to investigate the potential of refining the reedbed classification using slope suitability filters produced from digital terrain data. The results indicate an increase in the accuracy of reedbed delineations when grey-level co-occurrence textural measures were combined with the spectral bands. The most effective combination of texture measures were entropy and angular second moment. Optimal reedbed and overall classification accuracies were achieved using a combination of pansharpened multispectral and texture images that had been spatially degraded from 0.6 to 4.8 m. Using the 4.8 m data set, the MLC produced higher classification accuracy for reedbeds than the ANN analysis. The application of slope suitability filters increased the classification accuracy of reedbeds from 71% to 79%. Hence, this study has demonstrated that it is possible to use high resolution multispectral satellite imagery to derive accurate maps of reedbeds through appropriate analysis of image texture, judicious selection of input bands, spatial resolution and classification algorithm and post-classification refinement using terrain data.     

The role of digital bathymetry in mapping shallow marine vegetation from hyperspectral image data

Hyperspectral remote sensing is a proven technology for measurement of coastal ocean colour, including sea‐bed mapping in optically shallow waters. Using hyperspectral imagery of shallow (<15 m deep) sea bed acquired with the Compact Airborne Spectrographic Imager (CASI‐550), we examined how changes in the spatial resolution of bathymetric grids, created from sonar data (echosounding) and input to conventional image classifiers, affected the accuracy of distributional maps of invasive (Codium fragile ssp. tomentosoides) and native (kelp) seaweeds off the coast of Nova Scotia, Canada. The addition of a low‐resolution bathymetric grid, interpolated from soundings by the Canadian Hydrographic Service, improved the overall classification accuracies by up to ～10%. However, increasing the bathymetric resolution did not increase the accuracy of classification maps produced with the supervised (Maximum Likelihood) classifier as shown by a slightly lower accuracy (2%) when using an intermediate‐resolution bathymetric grid interpolated from soundings with a recreational fish finder. Supervised classifications using the first three eigenvectors from a principal‐components analysis were consistently more accurate (by at least 27%) than unsupervised (K‐means classifier) schemes with similar data compression. With an overall accuracy of 76%, the most reliable scheme was a supervised classification with low‐resolution bathymetry. However, the supervised approach was particularly sensitive, and variations in accuracy of 2% resulted in overestimations of up to 53% in the extent of C. fragile and kelp. The use of a passive optical bathymetric algorithm to derive a high‐resolution bathymetric grid from the CASI data showed promise, although fundamental differences between this grid and those created with the sonar data limited the conclusions. The bathymetry (at any spatial resolution) appeared to improve the accuracy of the classifications both by reducing the confusion among the spectral classes and by removing noise in the image data. Variations in the accuracy of depth estimates and inescapable positional inaccuracies in the imagery and ground data largely accounted for the observed differences in the classification accuracies. This study provides the first detailed demonstration of the advantages and limitations of integrating digital bathymetry with hyperspectral data for the mapping of benthic assemblages in optically shallow waters.     

Mapping invasive wetland plants in the Hudson River National Estuarine Research Reserve using quickbird satellite imagery

The National Estuarine Research Reserve (NERR) program is a nationally coordinated research and monitoring program that identifies and tracks changes in ecological resources of representative estuarine ecosystems and coastal watersheds. In recent years, attention has focused on using high spatial and spectral resolution satellite imagery to map and monitor wetland plant communities in the NERRs, particularly invasive plant species. The utility of this technology for that purpose has yet to be assessed in detail. To that end, a specific high spatial resolution satellite imagery, QuickBird, was used to map plant communities and monitor invasive plants within the Hudson River NERR (HRNERR). The HRNERR contains four diverse tidal wetlands (Stockport Flats, Tivoli Bays, Iona Island, and Piermont), each with unique water chemistry (i.e., brackish, oligotrophic and fresh) and, consequently, unique assemblages of plant communities, including three invasive plants (Trapa natans, Phragmites australis, and Lythrum salicaria). A maximum-likelihood classification was used to produce 20-class land cover maps for each of the four marshes within the HRNERR. Conventional contingency tables and a fuzzy set analysis served as a basis for an accuracy assessment of these maps. The overall accuracies, as assessed by the contingency tables, were 73.6%, 68.4%, 67.9%, and 64.9% for Tivoli Bays, Stockport Flats, Piermont, and Iona Island, respectively. Fuzzy assessment tables lead to higher estimates of map accuracies of 83%, 75%, 76%, and 76%, respectively. In general, the open water/tidal channel class was the most accurately mapped class and Scirpus sp. was the least accurately mapped. These encouraging accuracies suggest that high-resolution satellite imagery offers significant potential for the mapping of invasive plant species in estuarine environments.     

Improved classification of conservation tillage adoption using high temporal and synthetic satellite imagery

Conservation tillage management has been advocated for carbon sequestration and soil quality preservation purposes. Past satellite image analyses have had difficulty in differentiating between no-till (NT) and minimal tillage (MT) conservation classes due to similarities in surface residues, and may have been restricted by the availability of cloud-free satellite imagery. This study hypothesized that the inclusion of high temporal data into the classification process would increase conservation tillage accuracy due to the added likelihood of capturing spectral changes in MT fields following a tillage disturbance. Classification accuracies were evaluated for Random Forest models based on 250-m and 500-m MODIS, 30-m Landsat, and 30-m synthetic reflectance values. Synthetic (30-m) data derived from the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) were evaluated because high frequency Landsat image sets are often unavailable within a cropping season due to cloud issues. Classification results from a five-date Landsat model were substantially better than those reported by previous classification tillage studies, with 94% total and ≥ 88% class producer's accuracies. Landsat-derived models based on individual image scenes (May through August) yielded poor MT classifications, but a monthly increase in accuracy illustrated the importance of temporal sampling for capturing regional tillage disturbance signatures. MODIS-based model accuracies (90% total; ≥ 82% class) were lower than in the five-date Landsat model, but were higher than previous image-based and survey-based tillage classification results. Almost all the STARFM prediction-based models had classification accuracies higher than, or comparable to, the MODIS-based results (> 90% total; ≥ 84% class) but the resulting model accuracies were dependent on the MODIS/Landsat base pairs used to generate the STARFM predictions. Also evident within the STARFM prediction-based models was the ability for high frequency data series to compensate for degraded synthetic spectral values when classifying field-based tillage. The decision to use MODIS or STARFM-based data within conservation tillage analysis is likely situation dependent. A MODIS-based approach requires little data processing and could be more efficient for large-area mapping; however a STARFM-based analysis might be more appropriate in mixed-pixel situations that could potentially compromise classification accuracy.     

Comparison and integration of spaceborne optical and radar data for mapping in Sudan

The purpose of this study was to determine how different procedures and data, such as multiple wavelengths of radar imagery and radar texture measures, independently and in combination with optical imagery influence land-cover/use classification accuracies for a study site in Sudan. Radarsat-2 C-band and phased array L-band synthetic aperture radar (PALSAR) L-band quad-polarized radar were registered with ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) optical data. Spectral signatures were obtained for multiple landscape features, classified using a maximum-likelihood decision rule, and thematic accuracies were obtained using separate validation data. There were surprising differences between the thematic accuracies of the two radar data sets, with Radarsat-2 only having a 51% accuracy and PALSAR 73%. In contrast, the optical ASTER overall accuracy was 81%. Combining the original radar and a variance texture measure increased the Radarsat-2 to 78% and PALSAR to 80%, whereas the two original radar bands together had an accuracy of 87%. Sensor fusion of optical and radar obtained an accuracy of 93%. Based on these results, the use of multiwavelength quad-polarized radar imagery combined or integrated with optical imagery has great potential in improving the accuracy of land-cover/use classifications. In tropical and high-latitude regions of the world, where persistent cloud cover hinders the use of optical satellite systems, land management programmes may find this research promising.     

Use of textural measurements to map invasive wetland plants in the Hudson River National Estuarine Research Reserve with IKONOS satellite imagery

At this point, models, and accompanying field data, that could be used to predict the likely response of estuaries and tidal marshes to future environmental change are lacking. To improve this situation, monitoring efforts in these complex ecosystems need to be intensified, and new, efficient monitoring techniques should be developed. In this context, our research assessed the use of IKONOS satellite imagery to map plant communities at Tivoli Bays, in the Hudson River National Estuarine Research Reserve (HRNERR). Tivoli Bays, a freshwater tidal wetland, contains a unique assemblage of plant communities, including three invasive plants (Trapa natans, Phragmites australis, and Lythrum salicaria). To study the effects of textural information on the accuracy of land cover maps produced for the HRNERR, seven different 11-class land cover maps were produced using a maximum-likelihood classification on seven combinations of spectral and textural data derived from an IKONOS image. Conventional contingency tables served as a basis for an accuracy assessment of these maps. The overall classification accuracies, as assessed by the contingency tables, ranged from 45% to 77.7%. The maximum-likelihood classification relying on four spectral and four 5-by-5 filter textural bands (created by superposing a textural filter separately on each band of the IKONOS image) had the lowest overall accuracy, whereas the one based on four spectral and four 3-by-3 filter textural bands associated with all segments, identified by an object-based classification of the IKONOS image, had the highest accuracy. Results suggest that a combination of per-pixel classification and incorporation of texture for segments generated through an object-based classification slightly increases classification accuracy from 76.2% for the maximum-likelihood classification of the four spectral bands of the IKONOS image to 77.7% for the combination of spectral and textural information produced for selected segments. Further analysis indicates that better results may be obtained by using other types of data within the segments and that the traditional approach to the selection of training and accuracy sites may negatively bias the results for a combination per-pixel and object-based classification.     

Chaotic maps based on binary particle swarm optimization for feature selection

Feature selection is a useful pre-processing technique for solving classification problems. The challenge of solving the feature selection problem lies in applying evolutionary algorithms capable of handling the huge number of features typically involved. Generally, given classification data may contain useless, redundant or misleading features. To increase classification accuracy, the primary objective is to remove irrelevant features in the feature space and to correctly identify relevant features. Binary particle swarm optimization (BPSO) has been applied successfully to solving feature selection problems. In this paper, two kinds of chaotic maps—so-called logistic maps and tent maps—are embedded in BPSO. The purpose of chaotic maps is to determine the inertia weight of the BPSO. We propose chaotic binary particle swarm optimization (CBPSO) to implement the feature selection, in which the K-nearest neighbor (K-NN) method with leave-one-out cross-validation (LOOCV) serves as a classifier for evaluating classification accuracies. The proposed feature selection method shows promising results with respect to the number of feature subsets. The classification accuracy is superior to other methods from the literature.     

Modelling and mapping potential hooded warbler (Wilsonia citrina) habitat using remotely sensed imagery

Modelling and mapping of hooded warbler (Wilsonia citrina) nesting habitat in forests of southern Ontario were conducted using Ikonos and Landsat data. The study began with an analysis of skyward hemispherical photography to determine canopy characteristics associated with nest sites. It showed that nest sites had significantly less overhead canopy cover and larger maximum gap size than in non-nest areas. These findings led to the hypothesis that brightness variability in high to moderate resolution remotely sensed imagery may be greater at nest sites than in non-nest areas due to larger shadows and greater shadow variability related to these gap characteristics. This was confirmed when, in addition to some spectral band brightness variables, several image texture and spectrally unmixed fraction (shadow, bare soil) variables were found to be significantly different for nest and non-nest sites in Ikonos and Landsat imagery. These significantly different variables were used in maximum likelihood classification (MLC) and logistic regression (LR) to produce maps of potential nesting habitat. Mapping was conducted with Ikonos and Landsat in a local area where most known nest sites occur, and regionally using Landsat data for almost all of the hooded warbler range in southern Ontario. For the local area mapping using Ikonos data, a posteriori probabilities for both the MLC and LR methods showed that about 62% of the nest sites set aside for validation had been classified with high probability (p > 0.70) in the nest class. MLC mapping accuracy was 70% for the validation nest sites and 87% of validation nest sites were within 10 m of classified nesting habitat, a distance approximately equivalent to expected positional error in the data. LR accuracy was slightly lower. Nest site MLC mapping accuracy in the local area using Landsat data was 87% but the map was much coarser due to the larger pixel size. Regional mapping with Landsat imagery produced lower classification accuracy due to high errors of commission for the habitat class. This resulted from a poor spatial distribution and low number of observations of nest sites throughout the region compared to the local area, while the non-nest site data distribution was too narrow, having been defined and assessed (using standard accepted methods) as areas with no ground shrubs. If either of these problems can be ameliorated, regional mapping accuracy may improve.     

Land cover mapping of large areas using chain classification of neighboring Landsat satellite images

Satellite imagery is the major data source for regional to global land cover maps. However, land cover mapping of large areas with medium-resolution imagery is costly and often constrained by the lack of good training and validation data. Our goal was to overcome these limitations, and to test chain classifications, i.e., the classification of Landsat images based on the information in the overlapping areas of neighboring scenes. The basic idea was to classify one Landsat scene first where good ground truth data is available, and then to classify the neighboring Landsat scene using the land cover classification of the first scene in the overlap area as training data. We tested chain classification for a forest/non-forest classification in the Carpathian Mountains on one horizontal chain of six Landsat scenes, and two vertical chains of two Landsat scenes each. We collected extensive training data from Quickbird imagery for classifying radiometrically uncorrected data with Support Vector Machines (SVMs). The SVMs classified 8 scenes with overall accuracies between 92.1% and 98.9% (average of 96.3%). Accuracy loss when automatically classifying neighboring scenes with chain classification was 1.9% on average. Even a chain of six images resulted only in an accuracy loss of 5.1% for the last image compared to a reference classification from independent training data for the last image. Chain classification thus performed well, but we note that chain classification can only be applied when land cover classes are well represented in the overlap area of neighboring Landsat scenes. As long as this constraint is met though, chain classification is a powerful approach for large area land cover classifications, especially in areas of varying training data availability.     

Evaluation of MK-4 multispectral satellite photography in land cover classification of eastern Ontario

Russian MK-4 multispectral satellite photography has been investigated for potential in land cover classification. Thematic maps were generated using maximum likelihood, neural network and context classifiers. Classifications of the raw spectral data, of spectral transforms, and of combined spectral/textural data were evaluated. Low point-based class accuracies resulted for land cover types exhibiting high spatial variability at the given pixel spacing of 7.5m, while more spatially homogeneous cover types were well classified. Several issues arose which need to be addressed for effective future use of high-resolution satellite sensors in regional land cover mapping. They include the need for further research in techniques for classification and accuracy assessment which are sensitive to the spatial variance of such high resolution imagery, and optimization of class attribute definitions.     

Subpixel abundance estimates in mixture-tuned matched filtering classifications of leafy spurge (Euphorbia esula L.)

Two demonstration sites in southeast Idaho, USA were used to extend remote sensing of leafy spurge research to fine-scale detection for abundance mapping using matched filtering (MF) scores. Linear regression analysis was used to quantify the relationship between MF estimates and calibrated ground estimates of leafy spurge abundance. The two sites had r ² values of 0.46 and 0.64. Both the slope of the regressions and the scaling behaviour of MF scores indicate that the technique consistently underestimated true abundance (defined here as percentage canopy cover) by roughly one-third. This underestimation may be influenced by field estimation bias and algorithm confusion between target and background signal. Further results indicate that MF exhibits linear scaling behaviour in six locations containing dense, uniform infestations. At these locations, where canopy cover was held relatively constant, high spatial resolution (3 m) estimates were not significantly different from coarser spatial resolution estimates (up to 16 m). Given the mathematically unconstrained nature of the estimation technique, MF is not a straightforward method for estimating leafy spurge canopy cover.