PE&RS January 2006

Peer-Reviewed Articles

25 Effect of Alternative Splitting Rules on Image Processing Using Classification Tree Analysis
Michael Zambon, Rick Lawrence, Andrew Bunn, and Scott Powell

Abstract  Download Full Article
Rule-based classification using classification tree analysis (CTA) is increasingly applied to remotely sensed data. CTA employs splitting rules to construct decision trees using training data as input. Results are then used for image classification. Software implementations of CTA offer different splitting rules and provide practitioners little guidance for their selection. We evaluated classification accuracy from four commonly used splitting rules and three types of imagery. Overall accuracies within data types varied less than 6 percent. Pairwise comparisons of kappa statistics indicated no significant differences (p-value > 0.05). Individual class accuracies, measured by user’s and producer’s accuracy, however, varied among methods. The entropy and twoing splitting rules most often accounted for the poorest performing classes. Based on analysis of the structure of the rules and the results from our three data sets, when the software provides the option, we recommend the gini and class probability rules for classification of remotely sensed data.

31 Advanced Exploratory Data Analysis for Mapping Regional Canopy Cover
Yaguang Xu, John W. Prather, Haydee M. Hampton, Ethan N. Aumack, Brett G. Dickson, and Thomas D. Sisk

Abstract  Download Full Article
USGS Digital Orthophoto Quadrangles (DOQs) are a form of inexpensive, high spatial resolution (1 m ground resolution) imagery available for most regions of the United States. Typically, DOQs have been used in the construction of“basemaps” or as training datasets. In this paper we present a technical approach that uses DOQs as the primary data source to map regional forest canopy cover. This approach, a form of advanced exploratory data analysis (AEDA), can separate areas of crown, shadow, and non-crown vegetation using a single value threshold and a “value range” threshold obtained by analyzing the statistical plots built on a multi-fractal model. By applying AEDA, we can distinguish the crown, crown boundary zone, shadow areas, and non-crown areas within a DOQ mosaic by their distinctive multifractal properties. Over a period of two years, we mapped canopy cover of 20,000 km² of ponderosa pine-dominated forest across northern Arizona using this technique. We used two hundred ground plot measurements from an independent source to assess the accuracy of the canopy cover map across an 8,000 km² region.

39 Photogrammetric Techniques in Civil Engineering Material Testing and Structure Monitoring
Hans-Gerd Maas and Uwe Hampel

Abstract  Download Full Article
Civil engineering material testing includes a wide range of applications requiring the determination of the three-dimensional shape of an object and changes thereof. Large structure monitoring will often include the necessity of determining object deformations at a large number of points. Photogrammetric techniques offer a large potential for the solution of a wide range of measurement tasks in this field. A modular toolbox consisting of digital cameras, computer interfaces, illumination systems, calibration devices, combined with subpixel accuracy image measurement operators, multi-image matching techniques, and self-calibrating bundle adjustment in a suitable user interface, depicts a very powerful tool for tailoring custom-made solutions for material testing labs. Due to the wide range and the repetitive nature of measurements tasks in civil engineering, these applications could depict a significant future market for photogrammetry.

This paper will briefly discuss the major hardware and software modules of a toolbox for civil engineering material testing and large structure monitoring. Based on several sample applications covering object dimensions from 10 cm to 500 meters, the potential of photogrammetric deformation measurement techniques will be shown. The major advantage of photogrammetric techniques can often be seen in the fact that they allow for highly automated measurements at a large number of points simultaneously. In many cases, object deformations can be determined at a precision in the order of 1:100,000 of the object dimension, based on off-the-shelf hardware components.

47 Mapping Sagebrush Distribution Using Fusion of Hyperspectral and Lidar Classifications
Jacob T. Mundt, David R. Streutker, and Nancy F. Glenn

Abstract  Download Full Article
The applicability of high spatial resolution hyperspectral data and small-footprint Light Detection and Ranging (lidar) data to map and describe sagebrush in a semi-arid shrub steppe rangeland is demonstrated. Hyperspectral processing utilized a spectral subset (605 nm to 984 nm) of the reflectance data to classify sagebrush presence to an overall accuracy of 74 percent. With the inclusion of co-registered lidar data, this accuracy increased to 89 percent. Furthermore, lidar data were utilized to generate stand specific descriptive information in areas of sagebrush presence and sagebrush absence. The methods and results of this study lay the framework for utilizing co-registered hyperspectral and lidar data to describe semi-arid shrubs in greater detail than would be feasible using either dataset independently or by most ground based surveys.

55 Estimation of Inter-annual Crop Area Variation by the Application of Spectral Angle Mapping to Low Resolution Multitemporal NDVI Images
Felix Rembold and Fabio Maselli

Abstract  Download Full Article
This current work is aimed at developing and testing a methodology which can be applied to low spatial resolution satellite data to assess inter-annual crop area variations on a regional scale. The methodology is based on the assumption that within mixed pixels, such variations are reflected by changes of the related multitemporal Normalised Difference Vegetation Index (NDVI) profiles. This implies that low resolution NDVI images with high temporal frequency can be used to update land cover estimates derived from higher resolution cartography.

Specifically, changes in shape of NDVI profiles are quantified by Spectral Angle Mapping (SAM), which has the advantage of being nearly insensitive to inter-year NDVI differences caused by meteorological variability. A calibration phase is then necessary to convert the information derived from SAM into relevant area variations which is carried out by a regression estimator calibrated to the data of a training year for which both low-resolution NDVI data and higher-resolution land-cover references are available. The methodology can also cope with inter-annual differences in the crop phenological cycles through temporal shifting of the reference NDVI profiles.

The proposed methodology was applied in a study region in central Italy to estimate area changes of winter crops from low-resolution NDVI profiles. The accuracy of such estimates was assessed by comparison to official agricultural statistics. The method showed promise for estimating crop area variation on a regional scale with the accuracy of the final results dependent on the quality of the satellite data time series and of the reference high-resolution land-cover maps.

63 Improving Building Footprints in InSAR Data by Comparison with a Lidar DSM
Paolo Gamba, Fabio Dell’Acqua, Gianni Lisini, and Francesco Cisotta

Abstract  Download Full Article
The first aim of this paper is to show how the joint use of Digital Surface Models (DSMs) coming from different sources may improve the understanding of an urban environment. More specifically, we consider laser and radar three- dimensional data over the same urban area and show that they can be profitably combined to improve building extraction. We exploit the better vertical and horizontal accuracy of the laser DSM, assumed to be available only for a small area, to ease the deformation of Interferometric Synthetic Aperture Radar (INSAR) DSM with built structures. To achieve this, we propose a method based on subsequent steps of geometrical correction, and mainly on a simple“stretching step” that uses laser data as a reference to adjust INSAR-derived building footprints. We show quantitative results obtained from two different urban areas, using different laser and radar data sets, to assess advantages and drawbacks of the proposed method.

71 Mapping Structural Parameters and Species Composition of Riparian Vegetation Using IKONOS and Landsat ETM+ Data in Australian Tropical Savannahs
Kasper Johansen and Stuart Phinn

Abstract  Download Full Article
Government agencies responsible for riparian environments are assessing the utility of remote sensing for mapping and monitoring vegetation structural parameters. The objective of this work was to evaluate Ikonos and Landsat-7 ETM+ imagery for mapping structural parameters and species composition of riparian vegetation in Australian tropical savannahs for a section of Keelbottom Creek, Queensland, Australia. Vegetation indices and image texture from Ikonos data were used for estimating leaf area index (R² = 0.13) and canopy percentage foliage cover (R² = 0.86). Pan- sharpened Ikonos data were used to map riparian species composition (overall accuracy = 55 percent) and riparian zone width (accuracy within ±3 m). Tree crowns could not be automatically delineated due to the lack of contrast between canopies and adjacent grass cover. The ETM+ imagery was suited for mapping the extent of riparian zones. Results presented demonstrate the capabilities of high and moderate spatial resolution imagery for mapping properties of riparian zones.