Timothy A. Warner

Detection and analysis of individual leaf-off tree crowns in small footprint, high sampling density lidar data from the eastern deciduous forest in North America

Leaf-off individual trees in a deciduous forest in the eastern USA are detected and analysed in small footprint, high sampling density lidar data. The data were acquired February 1, 2001, using a SAAB TopEye laser profiling system, with a sampling density of approximately 12 returns per square meter. The sparse and complex configuration of the branches of the leaf-off forest provides sufficient returns to allow the detection of the trees as individual objects and to analyse their vertical structures. Initially, for the detection of the individual trees only, the lidar data are first inserted in a 2D digital image, with the height as the pixel value or brightness level. The empty pixels are interpolated, and height outliers are removed. Gaussian smoothing at different scales is performed to create a three-dimensional scale-space structure. Blob signatures based on second-order image derivatives are calculated, and then normalised so they can be compared at different scale-levels. The grey-level blobs with the strongest normalised signatures are selected within the scale-space structure. The support regions of the blobs are marked one-at-a-time in the segmentation result image with higher priority for stronger blobs. The segmentation results of six individual hectare plots are assessed by a computerised, objective method that makes use of a ground reference data set of the individual tree crowns. For analysis of individual trees, a subset of the original laser returns is selected within each tree crown region of the canopy reference map. Indices based on moments of the first four orders, maximum value and number of canopy and ground returns, are estimated. The indices are derived separately for height and laser reflectance of branches for the two echoes. Significant differences (p<0.05) are detected for numerous indices for three major native species groups: oaks (Quercus spp.), red maple (Acer rubrum) and yellow poplar (Liriodendron tuliperifera). Tree species classification results of different indices suggest a moderate to high degree of accuracy using single or multiple variables. Furthermore, the maximum tree height is compared to ground reference tree height for 48 sample trees and a 1.1-m standard error (R 2 =68%

A Comparison of Multispectral and Multitemporal Information in High Spatial Resolution Imagery for Classification of Individual Tree Species in a Temperate Hardwood Forest

Multitemporal, small-format 35-mm aerial photographs were combined in a coregistered database to determine the relative value of spectral and phenological information for overstory tree crown classification of digital images of the Eastern Deciduous Forest. A one-hectare study site, located in a second-growth forest 15 km east of Morgantown, West Virginia, USA, was photographed from a light aircraft nine times from May to October 1997 using both true-color and false-color infrared film. Using this imagery, differences in the spectral properties and timing of phenologic events between tree species made it possible to discriminate four deciduous tree species, namely Liriodendron tulipifera, Acer rubrum, Quercus rubra, and Quercus alba, which made up nearly 99% of the trees at this study site. Optimally timed photography acquired during peak autumn colors provided the best single date of imagery, while photography from spring leaf-out was the second best. The best individual image band for tree species discrimination was the blue band. Classifications using all four spectral bands (blue, green, red, and infrared) and four dates (05/23/97, 06/23/97, 10/11/97, and 10/30/97) provided the best classification accuracies. Variable canopy illumination made digital classification of individual trees complex. A Likelihood Ratio test confirmed that the number of spectral bands included in the classification procedure (spectral resolution) and the number of dates (temporal resolution) significantly influenced the ability to identify tree species correctly. This study suggests that although multispectral data appear to be more valuable than multitemporal data, it may be possible to compensate for the limited spectral resolution of planned high-resolution sensors by combining multiple dates of low spectral resolution images.

Forest type mapping using object-specific texture measures from multispectral Ikonos imagery: segmentation quality and image classification issues.

This study investigated the use of a geographic object-based image analysis (GEOBIA) approach with the incorporation of object-specific grey-level co-occurrence matrix (GLCM) texture measures from a multispectral Ikonos image for delineation of deciduous, evergreen, and mixed forest types in Guilford Courthouse National Military Park, North Carolina. A series of automated segmentations was produced at a range of scales, each resulting in an associated range of number and size of objects (or segments). Prior to classification, the spatial autocorrelation of each segmentation was evaluated by calculating Moran’s I using the average image digital numbers (DNs) per segment. An initial assumption was made that the optimal segmentation scales would have the lowest spatial autocorrelation, and conversely, that over- and under-segmentation would result in higher autocorrelation between segments. At these optimal segmentation scales, the automated segmentation was found to yield information comparable to manually interpreted stand-level forest maps in terms of the size and number of segments. A series of object-based classifications was carried out on the image at the entire range of segmentation scales. The results demonstrated that the scale of segmentation directly influenced the object-based forest type classification results. The accuracies were higher for classification of images identified from a spatial autocorrelation analysis to have an optimal segmentation, compared to those determined to have over- and under-segmentation. An overall accuracy of 79 percent with a Kappa of 0.65 was obtained at the optimal segmentation scale of 19. The addition of object-specific GLCM multiple texture analysis improved classification accuracies up to a value of 83 percent overall accuracy and a Kappa of 0.71 by reducing the confusion between evergreen and mixed forest types. Although some misclassification still remained because of local segmentation quality, a visual assessment of the texture-enhanced GEOBIA classification generally agreeable with manually interpreted forest types.

Multi-scale GEOBIA with very high spatial resolution digital aerial imagery: scale, texture and image objects

This study used geographic object-based image analysis (GEOBIA) with very high spatial resolution (VHR) aerial imagery (0.3 m spatial resolution) to classify vegetation, channel and bare mud classes in a salt marsh. Three classification issues were investigated in the context of segmentation scale: (1) a comparison of single- and multi-scale GEOBIA using spectral bands, (2) the relative benefit of incorporating texture derived from the grey-level co-occurrence matrix (GLCM) in classifying the salt marsh features in single- and multi-scale GEOBIA and (3) the effect of quantization level of GLCM texture in the context of multi-scale GEOBIA. The single-scale GEOBIA experiments indicated that the optimal segmentation was both class and scale dependent. Therefore, the single-scale approach produced an only moderately accurate classification for all marsh classes. A multi-scale approach, however, facilitated the use of multiple scales that allowed the delineation of individual classes with increased between-class and reduced within-class spectral variation. With only spectral bands used, the multi-scale approach outperformed the single-scale GEOBIA with an overall accuracy of 82% vs. 76% (Kappa of 0.71 vs. 0.62). The study demonstrates the potential importance of ancillary data, GLCM texture, to compensate for limited between-class spectral discrimination. For example, gains in classification accuracies ranged from 3% to 12% when the GLCM mean texture was included in the multi-scale GEOBIA. The multi-scale classification overall accuracy varied with quantization level of the GLCM texture matrix. A quantization level of 2 reduced misclassifications of channel and bare mud and generated a statistically higher classification than higher quantization levels. Overall, the multi-scale GEOBIA produced the highest classification accuracy. The multi-scale GEOBIA is expected to be a useful methodology for creating a seamless spatial database of marsh landscape features to be used for further geographic information system (GIS) analyses.

Kernel-based extreme learning machine for remote-sensing image classification

This letter evaluates the effectiveness of a new kernel-based extreme learning machine (ELM) algorithm for a land cover classification using both multi- and hyperspectral remote-sensing data. The results are compared with the most widely used algorithms – support vector machines (SVMs). The results are compared in terms of the ease of use (in terms of the number of user-defined parameters), classification accuracy and computation cost. A radial basis kernel function was used with both the SVM and the kernel-based extreme-learning machine algorithms to ensure compatibility in the comparison of the two algorithms. The results suggest that the new algorithm is similar to, or more accurate than, SVM in terms of classification accuracy, has notable lower computational cost and does not require the implementation of a multiclass strategy.

Estimation of optimal image object size for the segmentation of forest stands with multispectral IKONOS imagery

The determination of segments that represents an optimal image object size is very challenging in object-based image analysis (OBIA). This research employs local variance and spatial autocorrelation to estimate the optimal size of image objects for segmenting forest stands. Segmented images are visually compared to a manually interpreted forest stand database to examine the quality of forest stand segmentation in terms of the average size and number of image objects. Average local variances are then graphed against segmentation scale in an attempt to determine the appropriate scale for optimally derived segments. In addition, an analysis of spatial autocorrelation is performed to investigate how between-object correlation changes with segmentation scale in terms of over-, optimal, and under-segmentation.

Apartheid Representations in a Digital Landscape: GIS, Remote Sensing and Local Knowledge in Kiepersol, South Africa

A GIS is currently being developed for the Kiepersol locality in the Eastern Transvaal which integrates conventional environmental and infrastructural data with nonconventional behavioral and cognitive information. Regional political ecology informs the GIS design in which socially differentiated knowledge sources are brought together. The GIS production process is undertaken with concern for the competing discourses associated with post-apartheid social transformation in South Africa and in full appreciation that geographic information systems are social constructions. The multiple realities of resource access and use represented within the Kiepersol GIS are intended to contribute to democratic decision-making for land and agrarian reform.

The Sage handbook of remote sensing

Remote Sensing Data Selection Issues - Timothy A. Warner, Duane Nellis, and Giles M. Foody PART ONE: INTRODUCTION Remote Sensing Data Selection Issues - Timothy A. Warner, Duane Nellis, and Giles M. Foody Remote Sensing Policy - Ray Harris PART TWO: ELECTROMAGNETIC RADIATION & THE TERRESTRIAL ENVIRONMENT Visible, Near-IR & Shortwave IR Spectral Characteristics of Terrestrial Surfaces - Willem van Leeuwen Interactions of Middle Infrared (3-5 m) Radiation with the Environment - Arthur Cracknell and D. S. Boyd Thermal Remote Sensing in Earth Science Research - Dale Quattrochi and Jeffrey C. Luvall Polarimetric SAR Phenomenology and Inversion Techniques for Vegetated Terrain - Mahta Moghaddam PART THREE: DIGITAL SENSORS AND IMAGE CHARACTERISTICS Optical Sensor Technology - John Kerekes Fine spatial resolution optical sensors - Thierry Toutin Moderate Spatial Resolution Optical Sensors - Samuel N. Goward, Terry Arvidson, Darrel L. Williams, Richard Irish and Jim Irons Coarse Resolution Optical Sensors - Chris Justice and Compton Tucker Airborne Digital Multispectral Imaging - Doug Stow, Lloyd L. Coulter and Cody A. Benkelman PART FOUR: REMOTE SENSING ANALYSIS: DESIGN AND IMPLEMENTATION Imaging Spectrometers - Michael Schaepma Active and Passive Microwave Systems - Josef Kellndorfer and Kyle McDonald Airborne Laser Scanning - Juha Hyyppa, W. Wagner, M. Hollaus and H. Hyyppa Radiometry and reflectance: From terminology concepts to measured quantities - Gabriela Schaepman-Strub, Michael E. Schaepman, John V. Martonchik, Thomas H. Painter and Stefan Dangel Pre-Processing of Optical Imagery - Freek van der Meer and Harald van der Werff and Steven de Jong Surface Reference Data Collection - Chris Johannsen and Craig S. T. Daughtry Integrating Remote Sensing and Geographic Information Systems - James Merchant and Sunil Narumalani Image Classification - John Jensen, Jungho Im, Perry Hardin, Ryan R. Jensen Quantitative Models and Inversion in Optical Remote Sensing - Shunlin Liang Accuracy Assessment - Steve Stehman, Giles Foody PART FIVE: REMOTE SENSING ANALYSIS: APPLICATIONS A. LITHOSPHERIC SCIENCES Making Sense of the Third Dimension Through Topographic Analysis - Yongxin Deng Remote Sensing of Geology - Xianfeng Chen and David Campagna Remote Sensing of Soils - Jim Campbell B. PLANT SCIENCES Remote sensing for studies of vegetation condition: Theory and application - Mike Wulder, Joanne C. White, Nicholas C. Coops and Stephanie Ortlepp Remote Sensing of Cropland Agriculture - M. Duane Nellis, Kevin Price and Don Rundquist C. HYDROSPHERIC & CRYSOPHERIC SCIENCES Optical Remote Sensing of the Hydrosphere: From the open ocean to inland waters - Samantha Lavender Remote Sensing of the Cryosphere - Jeff Dozier D. GLOBAL CHANGE AND HUMAN ENVIRONMENTS Remote Sensing for Terrestrial Biogeochemical Modeling - Greg Asner and Scott V. Ollinger Remote Sensing of Urban Areas - Janet Nichol Remote sensing and the social sciences - Kelley Crews and Stephen J. Walsh Hazard Assessment and Disaster Management using Remote Sensing - Richard Teeuw, Paul Aplin, Nick McWilliam, Toby Wicks, Matthieu Kervyn and Gerald Ernst Remote Sensing of Land Cover Change - Timothy A. Warner, Abdullah Almutairi and Jong Yeol Lee PART SIX:. CONCLUSIONS Remote Sensing: A Look to the Future - Giles M. Foody, Timothy A. Warner and M. Duane Nellis

Change Detection Accuracy and Image Properties: A Study Using Simulated Data

Simulated data were used to investigate the relationships between image properties and change detection accuracy in a systematic manner. The image properties examined were class separability, radiometric normalization and image spectral band-to-band correlation. The change detection methods evaluated were post-classification comparison, direct classification of multidate imagery, image differencing, principal component analysis, and change vector analysis. The simulated data experiments showed that the relative accuracy of the change detection methods varied with changes in image properties, thus confirming the hypothesis that caution should be used in generalizing from studies that use only a single image pair. In most cases, direct classification and post-classification comparison were the least sensitive to changes in the image properties of class separability, radiometric normalization error and band correlation. Furthermore, these methods generally produced the highest accuracy, or were amongst those with a high accuracy. PCA accuracy was highly variable; the use of four principal components consistently resulted in substantial decreased classification accuracy relative to using six components, or classification using the original six bands. The accuracy of image differencing also varied greatly in the experiments. Of the three methods that require radiometric normalization, image differencing was the method most affected by radiometric error, relative to change vector and classification methods, for classes that have moderate and low separability. For classes that are highly separable, image differencing was relatively unaffected by radiometric normalization error. CVA was found to be the most accurate method for classes with low separability and all but the largest radiometric errors. CVA accuracy tended to be the least affected by changes in the degree of band correlation in situations where the class means were moderately dispersed, or clustered near the diagonal. For all change detection methods, the classification accuracy increased as simulated band correlation increased, and direct classification methods consistently had the highest accuracy, while PCA generally had the lowest accuracy.

Spatial autocorrelation analysis of hyperspectral imagery for feature selection

Abstract The spatial information in a single spectral image can be estimated from the spatial autocorrelation, which is a measure of how the local variation compares with the overall variance in a scene. In images of random noise, the local variation tends to be similar to the overall variance. In contrast, scenes in which large features can be discerned have clusters of pixels with similar values, which cause the local variation to be much smaller on average than the overall scene variance. A comparison of the autocorrelation of images formed by the ratios of two spectral bands is an excellent way to determine which combinations provide the best spectral representation of objects greater in size than the spatial resolution of the sensor. This is because an image formed from the ratios of two nonredundant bands will enhance spectral objects and thus tend to have greater spatial autocorrelation than the ratio of two bands that are very similar. Ratios are a particularly effective method of combining images because this operation tends to reduce the effect of illumination differences and to enhance spectral features. Feature selection is the process of finding a subset of the original bands that provides an optimal trade-off between probability of error and classification cost (Swain and Davis, 1978). Three feature selection problems are addressed in this paper: (1) narrow band feature selection, which is the selection of a subset of individual bands; (2) broad band feature selection, in which groups of adjacent bands are selected, and (3) nonadjacent multiple band feature selection, in which selection. of the groups of bands is not limited to adjacent bands. Spatial autocorrelation is useful in all three feature selection problems. Narrow band feature selection is carried out by ranking the spatial autocorrelation of all possible combinations of ratioed bands. Broad band feature selection can be carried out by iteratively grouping adjacent bands that are the most similar. If the grouping is started from the previously identified best bands, it is possible to develop a metric to check that the incorporation of each additional band to the group enhances the spatial autocorrelation of all the groups of bands together. Nonadjacent multiple band feature selection is simply an extension of the broad band case, except any of the original bands can potentially be grouped in any of the features. Tests with simulated data indicate that the spatial autocorrelation based methods consistently identify the best bands or groups of bands. Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data of eastern Washington state are used to illustrate the technique on -real data. The results suggest that visible and near-infrared bands provide a large proportion of the spectral and spatial information in that scene. Adjacent bands in many cases provide similar information, but there are important exceptions such as on the red edge of the infra-red plateau.