ASPRS

PE&RS August 2000

VOLUME 66, NUMBER 8
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
JOURNAL OF THE AMERICAN SOCIETY FOR PHOTOGRAMMETRY AND REMOTE SENSING

Peer-Reviewed Article Abstracts

961 A Study on the Epipolarity of Linear Pushbroom Images
Taejung Kim

Abstract
Download Full Article
Although epipolar geometry is a very useful clue in processing stereo images, it has not been thoroughly examined previously for linear pushbroom images. Some have assumed that epipolar geometry would be the same for pushbroom images as for perspective images. Some do not use this geometry at all because it is not fully understood. The purpose of this paper is to provide a theoretical basis for the epipolar geometry of linear pushbroom images and to discuss the practical implications of this geometry in processing such images. We show that epipolarity for linear pushbroom images is different from that for perspective images. We also derive an equation for epipolar curves of linear pushbroom images, which are not lines but hyperbola-like non-linear curves. Through analyses of the properties of these curves, we conclude that these curves can be approximated as piece-wise linear segments and that any closely located points on one epipolar curve are mapped onto a common epipolar curve.

967 Relative Radiometric Normalization   Performance for Change Detection from Multi-Date Satellite Images
Xiaojun Yang and C.P. Lo

Abstract
Download Full Article
Relative radiometric normalization (RRN) minimizes radiometric differences among images caused by inconsistencies of acquisition conditions rather than changes in surface reflectance. Five methods of RRN have been applied to 1973, 1983, and 1988 Landsat MSS images of the Atlanta area for evaluating their performance in relation to change detection. These methods include pseudoinvariant features (PIF), radiometric control set (RCS), image regression (IR), no-change set determined from scattergrams (NC), and histogram matching (HM), all requiring the use of a reference-subject image pair. They were compared in terms of their capability to improve visual image quality and statistical robustness. The way in which different RRN  methods affect the results of information extraction in change detection was explored. It was found that RRN  methods which employed a large sample size to relate targets of subject images to the reference image exhibited a better overall performance, but tended to reduce the dynamic range and coefficient of variation of the images, thus undermining the accuracy of image classification. It was also found that visually and statistically robust RRN methods tended to substantially reduce the magnitude of spectral differences which can be linked to meaningful changes in landscapes. Finally, factors affecting the performance of relative radiometric normalization were identified, which include land-use/land-cover distribution, water-land proportion, topographic relief, similarity between the subject and reference images, and sample size.

981 Landslide Hazard Mapping and its Evaluation Using GIS: An Investigation of Sampling Schemes for a Grid-Cell Based Quantitative Method
Amod Sagar Dhakal, Takaaki Amada, and Masamu Aniya

Abstract
Download Full Article
An application of GIS for landslide hazard assessment using multivariate statistical analysis, mapping, and the evaluation of the hazard maps is presented. The study area is the Kulekhani watershed (124 km2) located in the central region of Nepal. A distribution map of landslides was produced from aerial photo interpretation and field checking. To determine the factors and classes influencing landsliding, layers of topographic factors derived from a digital elevation model, geology, and land use/land cover were analyzed by quantification scaling type II (discriminant) analysis, and the results were used for hazard mapping. The effects of different samples of landslide and non-landslide groups on the critical factors and classes and subsequently on hazard maps were evaluated. Simple random sampling was used to obtain samples of the landslide group, and either an unaligned stratified random sampling or an aligned systematic sampling method generated the non-landslide group. For the analysis, one set of the landslide group was combined with each of five different sets of the non-landslide groups. Combinations of different samples yielded some minor differences in the critical factors and classes. The geology was found to be the most important factor for landslide hazard. The scores of the classes of the factors quantified by the five analyses were used for the hazard mapping in the GIS, with four levels of relative hazard classes: high, moderate, less, and least. The evaluation of five hazard maps indicated higher accuracy for the combinations in which the non-landslide group was generated by the unaligned stratified random sampling method. The agreements in the hazard maps, produced from different sample combinations using unaligned stratified random sampling for selecting non-landslide group, were within the acceptable range for the practical use of a hazard map.

991 A Method for Continuous Extraction of Multispectrally Classified Urban Rivers
Yun Zhang

Abstract
Download Full Article
The continuous extraction of linear objects, such as rivers, roads, or boundaries, from digital images can hardly be achieved using automatic methods. Line extraction algorithms as well as multispectral classification usually break down linear objects into segments with significant noise. In urban areas it is especially hard to continuously extract small rivers because of mixed pixels and other disturbing elements (bridges, ships, shadows, etc.). Therefore, methods for connecting broken linear segments and eliminating noise are important. For mapping urban water areas, in addition to connecting broken river segments, additional operations have to be carried out. In this study, a simple and effective automatic method was developed to connect river segments and eliminate noise. By applying this method, discontinuous river segments can be connected, small water areas can be separated from noise, and noise in large water areas can be eliminated. The method was tested using Landsat TM images in the urban area of Shanghai, China. The visible water bodies in the TM image were extracted. The result shows that the accuracy of urban water area extraction is increased up to over 95 percent. Rivers and canals broader than one image pixel can be completely extracted. By applying this method, discontinuous river segments can be connected, small water areas can be separated from noise, and noise in large water areas can be eliminated.

1001 Forecasts of Monthly Averaged Daily Temperature Highs in Bowling Green, Ohio from Monthly Sea Surface Temperature Anomalies in the Eastern Pacific Ocean During the Previous Year.
Robert K. Vincent

Abstract
Download Full Article
Step-wise linear regression methods are employed to yield a whole-year forecast of the monthly averaged daily maximum temperature (TMAX) in Bowling Green, Ohio, from Sea Surface Temperature Anomalies (SSTA) measured in the previous year by the AVHHR satellite sensor package over a region in the eastern Pacific Ocean. The forecasts for 1996-1999 were less accurate on an absolute basis than on a seasonal basis, predicting TMAX on the same side of the seasonal average as actual TMAX for 7 months in 1996, 10 months in 1997, 10 months in 1998, and 5 of 7 months for which data are available in 1999. Improvement in forecast results occurred when models were constrained to those that definitively passed the Durbin-Watson statistical test for autocorrelation. Models produced with this method can only be applied to the area from which the TMAX data were recorded, but they are much simpler than models that simulate the physics of the atmosphere.

1011 Quantification Error Versus Location Error in Comparison of Categorical Maps
R. Gil Pontius, Jr.

Abstract
Download Full Article
This paper analyzes quantification error versus location error in a comparison between two cellular maps that show a categorical variable. Quantification error occurs when the quantity of cells of a particular category in one map is different from the quantity of cells of that category in the other map. Location error occurs when the location of a category in one map is different from the location of that category in the other map. The standard Kappa index of agreement is usually not appropriate for map comparison. This paper offers alternative statistics: (1) proportion correct with perfect ability to specify location, (2) proportion correct with perfect ability to specify quantity, (3) Kappa for no ability, (4) Kappa for location, and (5) Kappa for quantity. These statistics can help scientists improve classification. This paper applies these theoretical concepts to the validation of a land-use change model for the Ipswich Watershed in Massachusetts.

1017 Accuracy Assessment of Automatically Derived Digital Elevation Models from SPOT Images
Amnon Krupnik

Abstract
Download Full Article
Digital elevation models (DEMs) are nowadays an important resource for many disciplines and are used for generating digital products such as contour maps, orthoimages, and perspective views. The main incentive for the increasing popularity of DEMs is the ability to generate them automatically from aerial and satellite images by image processing techniques. The research project described in this paper aims at examining the accuracy and reliability of automatically generated DEMs. Four test areas were selected, covering desert, agricultural, urban, and mountainous terrain characteristics. DEMs were extracted for these areas automatically, and were compared to the results of manually measured DEMs, obtained from aerial images. The results are accurate in general; however, there are certain reliability problems, especially in the agricultural and mountainous areas. The paper describes the project, presents the results, and discusses the issue of handling the reliability problem.
Top Home