• Bentley, M. L., Mote T. L. , and Thebpanya P. , 2002: Using Landsat to identify thunderstorm damage in agricultural regions. Bull. Amer. Meteor. Soc., 83 , 363376.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Burgess, D. W., Magsig M. A. , Wurman J. , Dowell D. , and Richardson Y. , 2002: Radar observations of the 3 May 1999 Oklahoma City tornado. Wea. Forecasting, 17 , 456471.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cheng, P., and Toutin T. , 1998: Unlocking the potential for IRS-1C data: Geometric correction and data fusion processes offer the key to new applications and users. Earth Observation Mag., 7 , 2426.

    • Search Google Scholar
    • Export Citation
  • Dyer, R. C., 1988: Remote sensing identification of tornado tracks in Argentina, Brazil, and Paraguay. Photogramm. Eng. Remote Sens., 54 , 14291435.

    • Search Google Scholar
    • Export Citation
  • Dymon, U. J., 1997: Geographic information system support after Hurricane Fran. Proc. GIS/LIS Annual Conf. and Exposition, Cincinnati, OH, American Congress on Surveying and Mapping, 545–548.

    • Search Google Scholar
    • Export Citation
  • Eastman, J. R., and Fulk M. , 1993: Long sequence time series evaluation using standardized principal components. Photogramm. Eng. Remote Sens., 59 , 991996.

    • Search Google Scholar
    • Export Citation
  • Fujita, T. T., 1978: Aerial survey of Grand Gulf Plant and vicinity after the April 17, 1978 tornado. Satellite and Mesometeorology Research Project Res. Paper 162, Department of the Geophysical Sciences, University of Chicago, 18 pp.

    • Search Google Scholar
    • Export Citation
  • Fujita, T. T., and Smith B. E. , 1993: Aerial survey and photography of tornado and microburst damage. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 479–493.

    • Search Google Scholar
    • Export Citation
  • Goodman, S. J., Buechler D. , Driscoll K. , Burgess D. W. , and Magsig M. A. , 2000: Tornadic supercells on May 3, 1999 viewed from space during an overpass of the NASA TRMM observatory. Preprints, 20th Conf. on Severe Local Storms, Orlando, FL, Amer. Meteor. Soc., 638–641.

    • Search Google Scholar
    • Export Citation
  • Hung, R. J., and Smith R. E. , 1983: Remote sensing of Arkansas tornadoes on 11 April 1976 from a satellite, a balloon, and an ionospheric sounder array. Int. J. Remote Sens., 4 , 617630.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jackson, R. D., Slater P. N. , and Pinter P. J. , 1983: Discrimination of growth and water stress in wheat by various vegetation indices through clear and turbid atmospheres. Remote Sens. Environ., 15 , 187208.

    • Search Google Scholar
    • Export Citation
  • Jensen, J., 1996: Introductory Digital Image Processing: A Remote Sensing Perspective. Prentice Hall, 318 pp.

  • Jensen, J., Saalfeld A. , Broome F. , Cowen D. , Price K. , Ramsey D. , and Lapine L. , 1998: Spatial data acquisition and integration. University Consortium for Geographic Information Science White Paper on Research Priority. [Available online at http://www.ucgis.org/research_white/data.html.].

    • Search Google Scholar
    • Export Citation
  • Klimowski, B. A., Hjelmfelt M. R. , Bunkers M. J. , Sedlack D. , and Johnson L. R. , 1998: Hailstorm damage observed from the GOES-8 satellite: The 5–6 July 1996 Butte–Meads storm. Mon. Wea. Rev., 126 , 831834.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Koch, M., and El-Baz F. , 1998: Identifying the effects of the Gulf War on the geomorphic features of Kuwait by remote sensing and GIS. Photogramm. Eng. Remote Sens., 64 , 739747.

    • Search Google Scholar
    • Export Citation
  • Kumar, R. M., Surekha K. , Sheshasai M. V. R. , and Hebbar K. R. , 1999: Paddy acreage estimation using IRS-1C LISS-III and WIFS data. Abstracts, Second Int. Symp. on Operationalization of Remote Sensing, Enschede, Netherlands, International Institute for Aerospace Survey and Earth Sciences. [Available online at http://www.itc.nl/ags/research/ors99/abstract/Kumar.htm.].

    • Search Google Scholar
    • Export Citation
  • Kwarteng, A. Y., and Chavez P. S. , 1998: Change detection study of Kuwait City and environs using multi-temporal Landsat Thematic Mapper data. Int. J. Remote Sens., 19 , 16511662.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsh, S. E., Walsh J. L. , Lee C. T. , Beck L. R. , and Hutchinson C. F. , 1992: Comparison of multi-temporal NOAA-AVHRR and SPOT-XS satellite data for mapping landcover dynamics in the West Africa Sahel. Int. J. Remote Sens., 13 , 29973016.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mukai, Y., and Hasegawa I. , 2000: Extraction of damaged areas of windfall trees by typhoons using Landsat TM data. Int. J. Remote Sens., 21 , 647654.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Okamota, K., Yamakawa S. , and Kawashima H. , 1998: Estimation of flood damage to rice production in North Korea in 1995. Int. J. Remote Sens., 19 , 365371.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Purdom, J. F. W., 1993: Satellite observations of tornadic thunderstorms. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 265–274.

    • Search Google Scholar
    • Export Citation
  • Raghavswamy, V., Pathan S. K. , Mohan P. R. , Bhanderi R. J. , and Priya P. , 1996: IRS-1C applications for urban planning and development. Current Sci., 70 , 582587.

    • Search Google Scholar
    • Export Citation
  • Ramsey, E. W,I. I. I., Chappell D. K. , and Baldwin D. G. , 1997: AVHRR imagery used to identify hurricane damage in a forest wetland of Louisiana. Photogramm. Eng. Remote Sens., 63 , 293297.

    • Search Google Scholar
    • Export Citation
  • Ramsey, E. W,I. I. I., Jacobs D. M. , Sapkota S. K. , and Baldwin D. G. , 1998: Resource management of forested wetlands: Hurricane impact and recovery mapped by combining Landsat TM and NOAA AVHRR data. Photogramm. Eng. Remote Sens., 64 , 733738.

    • Search Google Scholar
    • Export Citation
  • Rao, D. P., Gautam N. C. , Nagaraja R. , and Mohan P. R. , 1996: IRS-1C applications in land use mapping and planning. Current Sci., 70 , 575581.

    • Search Google Scholar
    • Export Citation
  • Singh, A., 1989: Digital change detection techniques using remotely sensed data. Int. J. Remote Sens., 10 , 9891004.

  • Singh, A., and Harrison A. , 1985: Standardized principal components. Int. J. Remote Sens., 6 , 883896.

  • Speheger, D. A., Doswell III C. A. , and Stumpf G. J. , 2002: The tornadoes of 3 May 1999: Event verification in central Oklahoma and related issues. Wea. Forecasting, 17 , 362381.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Srivastava, P. K., Krishna B. G. , and Majumder K. L. , 1996: Cartography and terrain mapping using IRS-1C data. Current Sci., 70 , 562567.

    • Search Google Scholar
    • Export Citation
  • Tucker, C. J., 1979: Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ., 8 , 127150.

  • Witt, A., Eilts M. D. , Stumpf G. J. , Mitchell E. D. , Johnson J. T. , and Thomas K. W. , 1998: Evaluating the performance of WSR-88D severe storm detection algorithms. Wea. Forecasting, 13 , 513518.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    The postevent IRS-1C LISS-3 image taken on 8 May 1999. This is a false color composite image. In general, red indicates vegetated area; blue suggests urban area, water, roads, and tornado tracks; and green corresponds to bare soil and tornado tracks

  • View in gallery

    Images of three principal components from the LISS-3 multispectral image at 23.5-m resolution. The Oklahoma City tornado tracks show distinct signatures in the PC2 image. For display purposes, areas with higher brightness values are shown in darker shades

  • View in gallery

    Association of PC2 signatures and tornado damage in rural and urban areas. For display purposes, areas of higher brightness values are shown in darker shades

  • View in gallery

    Tornado tracks lacking PC2 signatures. The thick lines in the lower-left corner of (c) and the upper-right corner of (b) represent F4 contours. For display purposes, areas of higher brightness values are shown in darker shades

  • View in gallery

    Lack of signatures around Choctaw, Canadian River, and Chickasha in the NDVI image transformed from the LISS-3 multispectral image taken on 8 May 1999

  • View in gallery

    Comparison of signatures from NDVI and PC2 images in urban and rural areas. NDVI signatures appear wider than PC2 signatures and are responsive to some F2 damage in urban areas and F3 damage in rural areas

  • View in gallery

    More signatures of tornado damage appear in the NDVI change image. Circled areas are signatures missing in both PC2 and NDVI images. Nevertheless, these signatures are weak, and knowledge of the track proximity is helpful in identifying these weak signatures.

  • View in gallery

    Overviews of tornado damage in (a) rural and (b) urban areas. Photographs reproduced with permissions from Aerial Oklahoma, Inc

  • View in gallery

    NDVI image (a) before the tornado and (b) after the tornado, and (c) the NDVI change. (d) Typical tree damage, debris, and scoured ground in creek areas

  • View in gallery

    An example of large scoured ground and its signatures in (a) PC2 and (b) NDVI change images. (c) Photograph of debris and scoured ground at the location marked by the red arrows

  • View in gallery

    Comparison of tornado damage in open grassland and a residential area. Widespread debris in the residential area results in a much wider signature in the NDVI change image

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 433 374 27
PDF Downloads 250 217 21

Analysis of Tornado Damage Tracks from the 3 May Tornado Outbreak Using Multispectral Satellite Imagery

View More View Less
  • 1 Department of Geography, The University of Oklahoma, Norman, Oklahoma
  • | 2 National Weather Service Warning Decision Training Branch and Cooperative Institute for Mesoscale Meteorological Studies, Norman, Oklahoma
© Get Permissions
Full access

Abstract

Remote sensing (RS) and geographic information systems (GIS) techniques are applied to high-resolution satellite imagery to determine characteristics of tornado damage from the 3 May 1999 tornado outbreak. Three remote sensing methods, including principal components analysis, normalized difference vegetation index (NDVI) analysis, and NDVI change analysis, elicit tornado damage paths at different levels of detail on the 23.5-m-resolution images captured by the Linear Imaging Self-Scanning III (LISS-3) sensor on the Indian Remote Sensing (IRS) satellite before and after the outbreak. Remote sensing results were spatially overlaid on F-scale contours compiled by the members of Oklahoma Weather Center. Spatial overlays reveal that results from the principal components analysis correlate well with F3 or greater damage. NDVI analysis shows signatures expanding to F2 damage, and NDVI change analysis is capable of detecting F1 damage in some instances. In general, results of these analyses correspond to more severe damage in rural areas than in urban areas. Comparison with detailed ground surveys shows that the spectral signatures of tornado damage are related to vegetation damage and large debris fields. Variations in spectral signatures with Fujita tornado damage intensity suggest that land cover characteristics may be just as important as tornado damage intensity in creating a track detectable by satellite. It is concluded that RS and GIS techniques on IRS LISS-3 imagery (an example of multispectral satellite imagery) can be useful in assessing tornado damage, particularly for extensive and intense events.

Corresponding author address: May Yuan, Sarkey Energy Center, Room 684, 100 East Boyd St., Norman, OK 73019. Email: myuan@ou.edu

Abstract

Remote sensing (RS) and geographic information systems (GIS) techniques are applied to high-resolution satellite imagery to determine characteristics of tornado damage from the 3 May 1999 tornado outbreak. Three remote sensing methods, including principal components analysis, normalized difference vegetation index (NDVI) analysis, and NDVI change analysis, elicit tornado damage paths at different levels of detail on the 23.5-m-resolution images captured by the Linear Imaging Self-Scanning III (LISS-3) sensor on the Indian Remote Sensing (IRS) satellite before and after the outbreak. Remote sensing results were spatially overlaid on F-scale contours compiled by the members of Oklahoma Weather Center. Spatial overlays reveal that results from the principal components analysis correlate well with F3 or greater damage. NDVI analysis shows signatures expanding to F2 damage, and NDVI change analysis is capable of detecting F1 damage in some instances. In general, results of these analyses correspond to more severe damage in rural areas than in urban areas. Comparison with detailed ground surveys shows that the spectral signatures of tornado damage are related to vegetation damage and large debris fields. Variations in spectral signatures with Fujita tornado damage intensity suggest that land cover characteristics may be just as important as tornado damage intensity in creating a track detectable by satellite. It is concluded that RS and GIS techniques on IRS LISS-3 imagery (an example of multispectral satellite imagery) can be useful in assessing tornado damage, particularly for extensive and intense events.

Corresponding author address: May Yuan, Sarkey Energy Center, Room 684, 100 East Boyd St., Norman, OK 73019. Email: myuan@ou.edu

1. Introduction

Acquiring complete damage survey verification datasets is a crucial component of both National Weather Service (NWS) verification of severe weather products and services, and severe weather research. Detailed and accurate ground surveys unfortunately can take trained personnel days to complete and may cover distances on the scale of a state. Although detailed surveys are conducted on some high-profile events such as that of 3 May 1999, many events are not surveyed in great detail because of numerous resource issues (Speheger et al. 2002). Thus, the current verification data contained in Storm Data contain many errors that require significant evaluation before they can be used (Witt et al. 1998). It follows that better understanding of damage extent has the potential to reduce time involved in surveying damage and improve the accuracy of surveys. As a first step in evaluating a new technique to supplement ground surveys, this study illustrates the use of multispectral satellite imagery to detect tornado damage from the 3 May 1999 tornado outbreak.

Tornado damage surveys have historically relied upon aerial photographs and ground observations. As early as 1965, the Fujita group at the University of Chicago pioneered aerial survey and photography projects that were applied to understanding multiscale airflows of tornadoes and associated damage on the ground (Fujita and Smith 1993). They used numerous low-altitude aerial photographs specifically obtained for tornado damage track analysis. Through aerial photography interpretation and mapping, they offered explanations of wind dynamics related to patterns of structural or tree damage in the aftermath of tornadoes (Fujita 1978). Aerial photography continues to be the most commonly used remote sensing technique in tornado damage surveys, but visual inspection remains the primary analytical method used for this purpose. In this approach, researchers must have prior knowledge about the location of tornado paths, and aerial photographs are taken along these paths to capture damage patterns. However, aerial photography and ground observations quickly reach their limits for tornado damage verification when the knowledge of the location of tornado damage is unsure, and time, resources, and staffing constraints make exhaustive search impractical.

Satellite remote sensing offers a complementary alternative to aerial photography surveys. Remote sensing satellites make routine acquisitions of the earth's surface, providing imagery that can be used for tornado track verification and damage analysis. Although fixed revisiting periods of satellites and cloud-cover constraints may hinder the use of satellite imagery on rapid damage assessment, satellite imagery has the potential to provide damage information in remote areas and over large land surfaces. Furthermore, satellite imagery, distinguished from analog aerial photographs, enables automatic detection of changes on the earth's surface and allows digital analysis of these changes. Although satellite remote sensing, especially the use of Geostationary Operational Environmental Satellite and Tropical Rainfall Measuring Mission imagery, has been widely used to provide aerial perspectives of tornadic storms (Hung and Smith 1983; Purdom 1993; Goodman et al. 2000), the use of satellite imagery for tornado damage assessment is not yet common.

The fact that the resolution of some satellite imagery is too coarse for the detection of highly localized tornado damage may have contributed to the lack of remote sensing applications in tornado damage surveys. For example, the resolution of National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) imagery is 1.1 km, and the resolution of the National Aeronautics and Space Administration (NASA) Landsat Multispectral Scanner (MSS) is 79 m. These images can only detect tracks of large tornadoes, that is, those wider than 1.1 km or 79 m, respectively. Nevertheless, in an early attempt at analyzing Landsat MSS imagery acquired in the early 1970s, Dyer (1988) resolved questions involving a series of anomalies found on 1965 aerial photos taken over eastern Paraguay. The anomalies were in the form of elongated swaths with a predominant northwest–southeast orientation, practically devoid of vegetation, and the patterns contrasted very sharply with the surrounding forest terrain. After close examination and comparison with meteorological records, Dyer concluded that the anomalies were caused by several tornadoes occurring between May and July of 1965. The tornado tracks were still visible on an image acquired in March of 1986 by the French Earth Observing Satellite, SPOT.

In addition to AVHRR and Landsat MSS, there are satellites, especially recently launched ones, that carry advanced instruments to capture images at high spatial and temporal resolutions. Hence they are suitable for applications in environmental assessment and monitoring at regional and local scales. Most notable among these instruments are the Landsat Thematic Mapper (TM; 28.5 m multispectral, 16 days), Indian Remote Sensing satellite IRS-1C (5.8 m panchromatic and 23.5 m multispectral, 24 days), SPOT (5 m panchromatic and 10 m multispectral, 26 days), IKONOS Carterra (1 m panchromatic and 4 m multispectral, 6 days), and Earthwatch QuickBird (0.8 m panchromatic and 3.28 m multispectral, 4 days). For example, imagery from these satellites is useful for monitoring changes and recovery from hurricane damage to the environments (Ramsey et al. 1997) or forest fires in vegetated areas (Ramsey et al. 1998). However, spatial resolution determines the minimum feature discernible from a satellite image. For example, Landsat TM imagery with a spatial resolution of 28.5 m will not resolve any feature smaller than 28.5 m in size. In addition, the repeat cycles of satellites may not be able to capture postevent imagery in a timely manner and, hence, restrict the usefulness of satellite imagery for damage verification and assessment.

In particular, regional coverage of satellite imagery allows rapid and regular monitoring of the effects of natural and anthropogenic disasters, such as environmental impacts of the Gulf War (Koch and El-Baz 1998). The archiving of today's satellite imagery can facilitate revealing patterns that may be transformed into quantitative determinations of the rates of development or recovery of natural resources and into qualitative judgments of external affects on those resources (Ramsey et al. 1997). Likewise, the usefulness of satellite remote sensing to verify and to assess tornado damage becomes much more promising with satellite imagery of high spatial and temporal resolutions. Although the applications of remotely sensed data in studies of tornado damage per se are rare, satellite remote sensing has been widely used in assessing damage caused by natural and anthropogenic disasters worldwide. Studies of recent events such as environmental impacts of the Gulf War (Koch and El-Baz 1998; Kwarteng and Chavez 1998), floods [Mississippi River in 1993; Albany, Georgia, in 1994; Okamoto et al. (1998)], hurricanes (Hugo in 1989; Andrew in 1991; Fran in 1996), hail/wind damage (Klimowski et al. 1998; Bentley et al. 2002), forest fires (Los Alamos, New Mexico, in 2000), earthquakes (Northridge, California, in 1994), and typhoons (Mukai and Hasegawa 2000) demonstrated the effectiveness of using remote sensing imagery for damage assessment.

In many cases, it is important to acquire precise, quantitative information regarding damage to housing stock, disrupted transportation arteries, the flow of spilled materials, and utility infrastructure on the ground. Such information was invaluable in assessing damage and allocating scarce cleanup resources during Hurricanes Hugo, Andrew, and Fran (Dymon 1997). To facilitate such effort on damage assessment requires the combination of satellite remote sensing and aircraft-based photography surveys. Once the general damage areas have been identified from satellite imagery, aircraft can be deployed to take high-resolution imagery targeting these damaged sites. Jensen et al. (1998) suggested that high-resolution (0.3–1 m) panchromatic and/or near-infrared imagery be acquired within 1–2 days to support such detailed damage surveys.

This study explores the use of high-resolution satellite imagery with analytical methods in remote sensing and geographic information systems (GIS) to detect tornado tracks and to correlate the results to tornado damage on the ground. Tornadic storms can significantly alter land cover characteristics through damage to vegetation and anthropogenic structures and the resulting distribution of debris. Our premise is that these changes also affect spectral reflectance characteristics of the surface to a degree that can be detected using analysis of multispectral imagery. The significance of the study contributes to a technological approach to assist tornado verification and assessment. Expanding on this study, future work will investigate combining satellite and airborne images with detailed tornado ground damage surveys over diverse geographical areas using remote sensing and GIS methods.

The next section introduces the characteristics of remote sensing and ancillary data used in this study. A methodology section follows to elaborate on an integration of remote sensing and GIS technologies to detect tornado damage tracks and to correlate the tracks to F-scale and ground observations. The final section provides conclusions drawn from the findings, identifies potential applications, and suggests directions for further study.

2. Data

This study utilizes IRS satellite imagery because it has area coverage, spatial resolution, and cloud conditions suitable for studying the 3 May 1999 tornado outbreak.

The IRS satellite carries the Linear Imaging Self-Scanning III (LISS-3) sensor for multispectral imagery with 23.5-m spatial resolution. It also carries a panchromatic sensor with 5.8-m spatial resolution. The first three LISS-3 bands are in the ranges of green (0.52–0.59 μm), red (0.62–0.68 μm), and near-infrared (0.77–0.86 μm) at a swath width of 142 km. The panchromatic sensor captures band wavelength 0.5–0.75 μm at a swath width of 70 km. (Technical documentation of IRS-1C and its imagery data was available online at the time of writing at http://www.spaceimaging.com/aboutus/sat_consetel.htm and http://www.euromap.de/doc_004.htm) IRS-1C LISS-3 and panchromatic images are suitable for urban planning, agriculture crop acreage and yield estimation, drought monitoring, flood mapping, forest resource surveys, and land cover mapping (Srivastava et al. 1996; Rao et al. 1996; Raghavswamy et al. 1996; Cheng and Toutin 1998; Kumar et al. 1999).

This study focuses on the tornadoes with the most complete damage surveys of the outbreak captured in one LISS-3 scene. Tornadoes from one storm are analyzed, including those producing damage near Chickasha, Oklahoma City, and Choctaw, Oklahoma. Corresponding panchromatic imagery was also utilized for cross-referencing. To detect land cover change before and after the event, imagery was acquired to show land cover before and after the 3 May 1999 event. Images prior to the event were taken on 18 May 1997, and images following the event were taken on 8 May 1999. Both sets of the images were georegistered in Universal Transverse Mercator Zone-14 coordinates according to the same set of ground control points with root-mean-square errors (rmse) of 0.38 pixel size (8.9 m) in the 1997 image and 0.35 pixel size (8.2 m) in the 1999 image. Although the time lag between pre- and postevent images is not optimal, they were sufficient to detect nonrandom difference inside the area of the tornado tracks clearly (see section 4c).

The postevent LISS-3 image clearly shows distinct tornado paths in the Oklahoma City area (Fig. 1). Weak signatures exist for other significant tornadoes in the analysis area, but the Oklahoma City tornado damage stands out as unique in size and magnitude. Two research questions related to this image are central to this study: 1) how much of the tornado damage can be identified in this image? and 2) how do the damage signatures on the image correspond to the damage on the ground? To this end, damage data from detailed ground surveys by members of the Oklahoma Weather Center were acquired to correlate spectral damage signatures and ground damage severity. Damage photographs from ground surveys also provided references to identification and understanding of spectral signatures resulting from tornado damage.

3. Methods

There are many techniques established for change detection using satellite imagery (Singh 1989). In this study, two image enhancement methods—principal components analysis (PCA) and normalized difference vegetation index (NDVI) analysis—were used to transform the original multispectral imagery to new images to assist in recognition of land cover patterns left by the tornadoes. NDVI change analysis was then used to examine further the difference in vegetation before and after the event. One major advantage of PCA and NDVI is that no prior knowledge about the study area is needed, and they are able to enhance spectral signatures that may be later related to land cover types or ground features (such as tornado tracks) in an image. PCA elicited the best linear spectral combination of three LISS-3 bands to delineate tornado damage paths based solely on the postevent LISS-3 image. NDVI analysis examined the distribution of greenness values after the outbreaks. Whereas PCA and NDVI analyses only used the postevent LISS-3 imagery, NDVI change analysis took both pre- and postevent images to detect changes in vegetation properties before and after the event. It assessed differences in greenness value to reflect locality of vegetation damage. All three methods were implemented using ERDAS Imagine proprietary software (ERDAS, Inc., Atlanta, Georgia) and the proprietary ArcINFO Geographic Information System (Environmental Systems and Research Institute, Inc., Redlands, California).

F-scale contours developed from ground surveys by members of the Oklahoma Weather Center were digitized and georegistered using ArcView GIS and were spatially overlain with results from all three analyses to correlate F-scale and spectral signatures. Ground photographs and video were combined with aerial photographs to identify the land cover characteristics of selected areas of satellite signatures.

a. Principal components analysis

PCA is a common remote sensing technique used to analyze multispectral remotely sensed images by transforming the raw image to new linear band combinations that may be more interpretable than the original image (Singh and Harrison 1985). The transformation is based on a covariance matrix and results in orthogonal principal components with ordered variance properties. The first principal component (PC1) captures the largest amount of spectral variation in the original image. The second principal component (PC2) explains the second largest amount of spectral variation in the original image. Consequent principal components explain decreasing amounts of dataset variance. Because of the ordered variance properties, image enhancement and classification can be performed on just the first few selected principal components rather than all spectral bands individually in the original image. In addition, each principal component may exhibit high correlation to certain ground properties that contribute to the spectral variation (Eastman and Fulk 1993).

LISS-3 imagery has green, red, and near-infrared bands. Each band is treated as a variable in the PCA. Hence, the result consists of three principal components. Each of the principal components forms an image layer of a weighted linear combination of green, red, and near-infrared bands indicated by the eigenvectors in Table 1. The PC1 image carries over 78% of the total variation in the original image and accounts primarily for image brightness, as can be seen by the positive loadings on all three bands. The PC3 image explains only about 1.2% of the total spectral variation and looks to primarily pull information from the regional reservoirs and urban areas as well as from atmospheric scattering and sensor noise. The PC2 image shows a distinct signature along the tracks of the Oklahoma City tornado, whereas only subtle damage patterns appear in the images of the first and third principal components (Fig. 2). The two distinct tracks have been documented to be related to the Oklahoma City tornado, and the break in the satellite signature correlates well with the decrease in size of the tornado and rotation as viewed by both Weather Surveillance Radar-1988 Doppler (WSR-88D) and Doppler on Wheels radars (Burgess et al. 2002). The eigenvector of the second principal component indicates a relatively weak spectral response to the green band (0.368 79), a relatively strong response to the red band (0.6572), and a relatively strong negative response to the near-infrared band (−0.657 33). The contrast between the positive loadings of visible (green + red) bands and the negative loading of near-infrared results in de-emphasizing surfaces that are bright in the near-infrared in the PC2 image. The near-infrared band is particularly responsive to the amount of vegetation biomass. Hence, in the PC2 image, vegetated area exhibits lower brightness values and areas of pavement, bare soil, spare vegetation, and water bodies correspond to higher brightness values. Damaged areas inside the tornado tracks, therefore, tend to have lower brightness values. For display purposes, cells of lower brightness values (background) are shown in lighter shades in Figs. 2–4 so that the tornado-damaged tracks appear in darker shades in the PC2 image.

The PC2 image was examined further to provide information on how well the track signatures relate to the level of tornado damage represented by the F scale. Members of the Oklahoma Weather Center have created a contoured F-scale analysis for parts of the Oklahoma City tornado. The contour map was digitized and spatially overlain with the PC2 image. Spectral signatures of the tornado tracks are generally associated with F4 contours in rural areas and with F3 contours in urban areas (Fig. 3). Much more scatter exists in urban areas for which isolating tornado damage signatures is more complicated. A further investigation on multiple tornado cases is necessary to draw meaningful conclusions on such correlations, but it is clear that the PC2 spectral signatures match the damage from strong to violent tornadoes in both rural and urban areas.

There are relatively large segments of the track that do not have any signal in PC2, likely due to the lack of significant change in land cover (Fig. 4). Two factors may contribute to the insufficient change in land cover. First, the damage may be too weak to be registered in the image. That is, the level of damage is too weak such that damage less than F3 in urban areas and damage less than F4 in rural areas is not detectable in the PC2 image. A second factor may be that coverage of severe damage (the distribution of damage) may be too small and sparse to be recognizable in the image. For example, the lack of a stronger signature in F4 areas shown in Fig. 4 is either related to a local change in land cover or to sparse areas of the most intense damage. A further examination is necessary to understand which factors contribute to the weakness of the PC2 damage signal along the Oklahoma City tornado track shown in Fig. 4. Evaluation of NDVI and NDVI change analyses along with surface damage analysis reveal information about land cover characteristics related to the origin of the PC2 signatures.

b. NDVI and NDVI change analysis

NDVI is an index for assessing vegetation biomass and vegetation health using the red and near-infrared bands in a multispectral image (Tucker 1979; Jackson et al. 1983; Marsh et al. 1992). The calculation is based on the following equation, where IR stands for the near-infrared band and R stands for the visible red band:
i1520-0434-17-3-382-eq1

The brightness of a pixel increases in proportion to the amount of photosynthesizing vegetation (Jensen 1996). In general, pixels representing vegetation will have high NDVI values; water-based elements, such as clouds and lakes, have negative values; and roads, bare soil, and buildings usually have near-zero values. Transformation of the postevent LISS-3 image (Fig. 1) to an NDVI image reveals two continuous clear dark paths (low NDVI values) correlated to the tornado tracks (Fig. 5a). When compared with the PC2 image, the postevent NDVI image shows wider signatures, which in some urban areas correspond to F2 contours and in some rural areas correspond to F3 contours (Fig. 6). In addition, it has better spectral signatures at the early portions of the Oklahoma City tornado (cf. the upper right-hand corner in Figs. 4b and 5b). It also has a weak signature around the Canadian River area, where no signature can be identified in the PC2 image (cf. Figs. 4c and 5c). How the NDVI signatures relate to ground observations is investigated by surface damage analysis (see discussion in the next section).

The NDVI image only shows the greenness distribution after the tornado outbreak. It is best for patterns of spatially continuous low NDVI values, such as large areas of debris coverage or removal of vegetation. Further detection of tornado tracks in areas with moderate or even minor vegetation damage may be achieved by examining the change in NDVI values from before and after the event. The premise of the NDVI change analysis is that a change in NDVI values implies a change in land cover or vegetation conditions (either on vegetation biomass or vegetation health). Tornado tracks can therefore be identified by decreases in NDVI values, for the decreases may indicate some reduction in vegetation biomass (removal of vegetation by the tornado), degradation in vegetation health (damaged leaves and grass, etc.), or coverage by nongreen debris (such as dirt, construction materials, etc.). The NDVI change analysis used LISS-3 images taken on 18 May 1997 (about two years before the outbreak) and 8 May 1999. The image from 18 May 1997 was chosen because it is the most recent cloud-free image with similar seasonality as the 3 May 1999 image. Images taken in 1998 or in April 1999 should serve as a better alternative, but the result of comparison supports that the change detected from the 1997 and 1999 images can be correlated to the tornado damage track. In Fig. 7, the NDVI change image is better able to reflect parts of the tornado tracks that are not displayed clearly in both PC2 and NDVI images. The boundaries of NDVI change signatures for these tornado tracks generally correspond to F2 or F1 contours.

Table 2 summarizes statistics about the values of PC2, NDVI, and NDVI change within each F-scale zone and in the background areas of 23.5 km in proximity (1000 pixels) to these tornado tracks. The background values are used as baselines for comparing the effectiveness of the PC2, NDVI, and NDVI change images for track detection. Percent change (%) is calculated by the ratio of background-zone difference to the background value. Hence, its absolute value relates to the contrast of each zone to the background. Higher contrast results in a more pronounced signal. In comparison, the F3 zone exhibits about 16% contrast to the PC2 background but shows about 45% contrast in the NDVI image. Therefore, F3 zones are generally more discernible in the NDVI image than in the PC2 image. Among the three images, damage zones in the NDVI change image have the highest contrast values, which suggests that tornado damage tracks generally should be most visible in the NDVI change image. Because all damage zones exhibit more than 100% contrast to the NDVI change background, even F1 and F2 zones are identifiable in the NDVI change image.

In addition to mean, coefficient of variation (CV) can provide information on how diverse the cell values are in each zone, which leads to interpretation of how well each F-scale zone can be delineated in the damage track. Coefficient of variation normalizes standard deviation by the mean. As compared with the standard deviation, CV is relatively stable in assessing variations among data from different populations (in this case, different F-scale zones). In both PC2 and NDVI change images, CV values decrease as the degree of damage increases. It suggests that, in these two images, brightness values of higher damage zones (F3, F4, and F5) appear less diverse than the other zones. This is consistent with the observations that F3, F4, and F5 zones are more discernible than F1 and F2 zones in PC2 and NDVI change images. On the other hand, the NDVI image exhibits a reversed CV trend across F-scale zones. The effect of increasing coefficient of variation (noise) as the F scale increases suggests that individual F-scale zones are less discernible in the detected damage track from the NDVI image. Although the NDVI change image also has very high CV values, implying that all F zones are very noisy, the significantly larger CV values are mostly due to the fact that they are derived from normalization by considerably smaller means. Nevertheless, among the three images, the NDVI change image shows the highest contrast to the background mean, and its absolute value of CV decreases dramatically as the F scale increases. Therefore, it likely is the image that exhibits the clearest track of the tornado damage at F1–F5 scales and facilitates best the delineation of different damage zones.

Although there are strong signals for tornado damage, there are portions of the tracks missing in all three images. The subsequent surface analysis correlates spectral signatures and ground observations for a further understanding of the spectral correspondence to tornado damage.

c. Surface damage analysis

Extensive visual observations (including photos, video, and aerial photos) exist throughout the damage track of the Oklahoma City tornado and have been made available by damage survey members, storm chasers, local media, private business, and the public. Visual manifestations of surface modification were significant and expansive in both rural (Fig. 8a) and urban (Fig. 8b) areas. In areas in which accurate locations can be determined, visual documentation can be used to illustrate surface characteristics of tornado damage and how they relate to PC2, NDVI, and NDVI change images.

Figure 8a shows widespread vegetation destruction in rural areas for early parts of the tornado track. A closer look at damage along a creek area, combining NDVI images before and after the tornado with visual documentation of damage, illustrates one characteristic of land cover associated with the NDVI change. The fingerlike pattern of high values of NDVI before the tornado damage (Fig. 9a) represents locally dense areas of active vegetation, a common pattern associated with forested areas along creeks or drainage areas. The NDVI image following the tornado damage (Fig. 9b) shows a continuous area of low values throughout the damage track in this view. The subtraction of these two images (i.e., change in NDVI values; Fig. 9c) reveals relatively isolated pockets of significant vegetation change along creek patterns in the center of the damage path. It demonstrates that NDVI change can enhance the damage signatures in the NDVI analysis. The lack of a strong creek signature in NDVI change outside violent tornado damage illustrates that the severe winds of the tornado, not seasonal effects, were related to the change in vegetation.

One specific location along the creek signature in the NDVI change images shows a large area of denuded trees, scoured ground, and clusters of debris (Fig. 9d). This kind of damage was documented to be widespread throughout violent tornado damage in rural areas and was well registered in all three analysis images (i.e., PC2, NDVI, and NDVI change). A common feature to damage in creeks and ponds (seen in Fig. 9d) was significant accumulations of debris. The Emergency Watershed Protection program at the Natural Resources Conservation Service has found that creeks and ponds are common debris accumulation areas following many significant tornados, including the tornadoes on 3 May (T. Funderbunk 2001, personal communication). Thus while vegetation removal by violent tornado damage is likely a significant component relating to NDVI change, debris accumulation may be another contributing factor.

Farther along the tornado damage track in rural areas, there is another location exhibiting primarily vegetation damage devoid of debris. This damage area is detectable by all three remote sensing methods used in this study. Figure 10a shows a large area of low brightness values in the PC2 image that is consistent with lesser vegetation than surrounding areas. NDVI change over the same area (Fig. 10b) illustrates a more continuous damage signature than earlier portions of the track with some embedded low values within creeks also indicating less healthy vegetation. One particular location that is illustrated well using both techniques in Figs. 10a and 10b is displayed in a photograph (Fig. 10c) in which a large area of scoured ground is visible in a field containing only a crushed automobile. Therefore, in addition to damage in creek areas, the satellite signatures also relate to scoured open grassland.

Surface damage analysis in urban areas reveals more land cover characteristics relating to signatures in the satellite data. Investigation of urban characteristics is enhanced because one of the authors (Magsig) participated in the Oklahoma City portion of the NWS damage survey. The NDVI change image shows a near-continuous pattern of low brightness values as the tornado moved between residential areas (Fig. 11a). The first area of interest is a large field containing grasses (∼0.3 m tall). Figure 11c shows much less scouring of ground than earlier portions of tornado damage over rural areas. The NWS damage survey team spent a significant amount of time investigating the sparse gouge marks in this field created by large pieces of train car transported long distances, one piece of which is shown in Fig. 11c. Despite impressive evidence of lofting large heavy objects, grass in most of the field was not scoured, but rather appeared to be damaged or simply covered with a brown coating commonly displayed in aerial photos. Thus, damaged and nonscoured grass shows a strong signature in the NDVI change image because dust and dirt deposited on the canopy by the tornado would account for a significant decrease in near-infrared reflectance and NDVI.

As the tornado moved into dense residential areas displayed in Fig. 11a, the NDVI change signature widened by a factor of 2. Enormous amounts of debris were generated as subdivisions were reduced to rubble. Figure 11b illustrates an expansive debris field in a residential neighborhood representative of damage throughout the widened NDVI change signature. Thus, significant debris fields overlaying vegetation surfaces are another surface land cover characteristic associated with signatures in the satellite data.

4. Conclusions

This study has demonstrated that multispectral satellite imagery with remote sensing and GIS methods is useful for tornado verification and damage assessment. Through applications of principal components analysis, NDVI analysis, and NDVI change analysis, significant portions of the Oklahoma City tornado track and the Choctaw tornado track can be identified on IRS-1C LISS-3 imagery of 23.5-m resolution. The principal components analysis and NDVI analysis used only the postevent image, and the NDVI change analysis used both post- and preevent images to differentiate changes in greenness values.

The three analytical methods have demonstrated the ability to facilitate detection of tornado tracks at various levels. The principal components analysis reveals the most severe damage. Its signatures generally correlate to F4 contours in rural areas and F3 contours in urban areas. The NDVI analysis is able to detect F2 or F3 damage in rural and urban areas. Furthermore, the NDVI change analysis is at times capable of illustrating damage that is as weak as F1 or F2. Surface damage analysis illustrates that severely damaged areas are sometimes related to denuded trees, scoured ground, or extensive debris coverage, and less damaged areas are sometimes associated with damaged grass or sparse debris distribution.

Principal components analysis results in uncorrelated linear band combinations with decreasing amounts of spectral variation. The second principal component of the IRS-1C LISS-3 image is able to show spectral variation as a result of tornado damage, which is attributable to significant debris coverage or scoured ground within the tornado tracks. NDVI analysis, on the other hand, reveals the greenness distribution after the outbreak. Because damage to trees, debris coverage, and scoured ground results in a reduction of photosynthetic vegetation coverage on the ground, the NDVI analysis is able to reveal distinct paths of low NDVI values corresponding to the tornado tracks. The area of low NDVI values results from scoured ground and significant debris coverage within the severe damage areas. In combination, NDVI analysis revealed severe to moderate tornado damage, so that it has wider spectral signatures than PC2 along the tornado tracks.

NDVI change analysis examines the changes in greenness values before and after the outbreak. The analysis is able to glean major and minor vegetation damage that reduces the healthy appearance of vegetation without significant damage or removal. Note that distinct fingerlike patterns along creeks or drainage areas suggest that patterns of vegetation destruction can be enhanced further by NDVI change analysis. Hence, NDVI change analysis is able to reveal additional tornado tracks correlating with F1 or F2 damage, which is not prominent in the results of principal components analysis and NDVI analysis. In combination, the three methods can facilitate differentiating some areas of F1–F5 damage zones caused by the Oklahoma City tornado. Although analysis of NDVI change can reveal many levels of damage, it requires both post- and preevent images and therefore encounters a higher data cost than the other two methods. NDVI and NDVI change analyses are particularly suitable for damage assessment in rural areas because they are designed to measure difference in photosynthetic activity.

This study has shown the promise of applying multispectral satellite imagery for tornado verification and assessment, but more case studies are necessary to develop a system that relates spectral signatures to tornado damage. The spectral bands used in this study are commonly available in other satellite imagery, so the methods used in this study are readily applicable to other multispectral satellite imagery. The IRS-1C LISS-3 imagery is costly ($2500 for each LISS-3 image used in the study), but the newly launched NASA Landsat-7 taking imagery at 28.5-m resolution can provide comparable results at a much lower price ($605 per scene), which enables practical and operational use of satellite imagery for tornado verification and assessment. Furthermore, there are advanced very high resolution satellite images, such as IKONOS images at 1- and 5-m resolution, useful for detecting narrow tornado tracks, but they incur a much higher cost ($12–$17 km–2 for North America with a minimal order of $1000; the area of this study is about 5000 km2). Hence, the use of very high resolution satellite imagery for tornado verification and assessment, at least at the current state, may not be cost effective. In addition, the revisit times of satellites may prevent them from being able to acquire site images in a timely manner. The IKONOS satellite revisits a site every 6 days, but Landsat-7 has a repeat cycle of 16 days. Therefore, on average, IKONOS will have a 3-day delay and Landsat-7 an 8-day delay in capturing the aftermath of a tornado event (or other natural hazards). Nevertheless, as a research tool, high-resolution satellite imagery has the potential for effective tornado damage detection.

Acknowledgments

The research is in part supported by a NASA EPSCoR research grant (NCC5-171) with additional support from the Oklahoma NASA Space Grant program and Space Imaging, Inc. Greg Stumpf provided damage survey maps. James LaDue and Greg Stumpf provided detailed damage survey video and information. Aerial photographs were taken by Aerial Oklahoma, Inc. The authors thank the editor and three anonymous reviewers who provided constructive comments to an earlier draft of the paper.

REFERENCES

  • Bentley, M. L., Mote T. L. , and Thebpanya P. , 2002: Using Landsat to identify thunderstorm damage in agricultural regions. Bull. Amer. Meteor. Soc., 83 , 363376.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Burgess, D. W., Magsig M. A. , Wurman J. , Dowell D. , and Richardson Y. , 2002: Radar observations of the 3 May 1999 Oklahoma City tornado. Wea. Forecasting, 17 , 456471.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cheng, P., and Toutin T. , 1998: Unlocking the potential for IRS-1C data: Geometric correction and data fusion processes offer the key to new applications and users. Earth Observation Mag., 7 , 2426.

    • Search Google Scholar
    • Export Citation
  • Dyer, R. C., 1988: Remote sensing identification of tornado tracks in Argentina, Brazil, and Paraguay. Photogramm. Eng. Remote Sens., 54 , 14291435.

    • Search Google Scholar
    • Export Citation
  • Dymon, U. J., 1997: Geographic information system support after Hurricane Fran. Proc. GIS/LIS Annual Conf. and Exposition, Cincinnati, OH, American Congress on Surveying and Mapping, 545–548.

    • Search Google Scholar
    • Export Citation
  • Eastman, J. R., and Fulk M. , 1993: Long sequence time series evaluation using standardized principal components. Photogramm. Eng. Remote Sens., 59 , 991996.

    • Search Google Scholar
    • Export Citation
  • Fujita, T. T., 1978: Aerial survey of Grand Gulf Plant and vicinity after the April 17, 1978 tornado. Satellite and Mesometeorology Research Project Res. Paper 162, Department of the Geophysical Sciences, University of Chicago, 18 pp.

    • Search Google Scholar
    • Export Citation
  • Fujita, T. T., and Smith B. E. , 1993: Aerial survey and photography of tornado and microburst damage. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 479–493.

    • Search Google Scholar
    • Export Citation
  • Goodman, S. J., Buechler D. , Driscoll K. , Burgess D. W. , and Magsig M. A. , 2000: Tornadic supercells on May 3, 1999 viewed from space during an overpass of the NASA TRMM observatory. Preprints, 20th Conf. on Severe Local Storms, Orlando, FL, Amer. Meteor. Soc., 638–641.

    • Search Google Scholar
    • Export Citation
  • Hung, R. J., and Smith R. E. , 1983: Remote sensing of Arkansas tornadoes on 11 April 1976 from a satellite, a balloon, and an ionospheric sounder array. Int. J. Remote Sens., 4 , 617630.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jackson, R. D., Slater P. N. , and Pinter P. J. , 1983: Discrimination of growth and water stress in wheat by various vegetation indices through clear and turbid atmospheres. Remote Sens. Environ., 15 , 187208.

    • Search Google Scholar
    • Export Citation
  • Jensen, J., 1996: Introductory Digital Image Processing: A Remote Sensing Perspective. Prentice Hall, 318 pp.

  • Jensen, J., Saalfeld A. , Broome F. , Cowen D. , Price K. , Ramsey D. , and Lapine L. , 1998: Spatial data acquisition and integration. University Consortium for Geographic Information Science White Paper on Research Priority. [Available online at http://www.ucgis.org/research_white/data.html.].

    • Search Google Scholar
    • Export Citation
  • Klimowski, B. A., Hjelmfelt M. R. , Bunkers M. J. , Sedlack D. , and Johnson L. R. , 1998: Hailstorm damage observed from the GOES-8 satellite: The 5–6 July 1996 Butte–Meads storm. Mon. Wea. Rev., 126 , 831834.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Koch, M., and El-Baz F. , 1998: Identifying the effects of the Gulf War on the geomorphic features of Kuwait by remote sensing and GIS. Photogramm. Eng. Remote Sens., 64 , 739747.

    • Search Google Scholar
    • Export Citation
  • Kumar, R. M., Surekha K. , Sheshasai M. V. R. , and Hebbar K. R. , 1999: Paddy acreage estimation using IRS-1C LISS-III and WIFS data. Abstracts, Second Int. Symp. on Operationalization of Remote Sensing, Enschede, Netherlands, International Institute for Aerospace Survey and Earth Sciences. [Available online at http://www.itc.nl/ags/research/ors99/abstract/Kumar.htm.].

    • Search Google Scholar
    • Export Citation
  • Kwarteng, A. Y., and Chavez P. S. , 1998: Change detection study of Kuwait City and environs using multi-temporal Landsat Thematic Mapper data. Int. J. Remote Sens., 19 , 16511662.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsh, S. E., Walsh J. L. , Lee C. T. , Beck L. R. , and Hutchinson C. F. , 1992: Comparison of multi-temporal NOAA-AVHRR and SPOT-XS satellite data for mapping landcover dynamics in the West Africa Sahel. Int. J. Remote Sens., 13 , 29973016.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mukai, Y., and Hasegawa I. , 2000: Extraction of damaged areas of windfall trees by typhoons using Landsat TM data. Int. J. Remote Sens., 21 , 647654.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Okamota, K., Yamakawa S. , and Kawashima H. , 1998: Estimation of flood damage to rice production in North Korea in 1995. Int. J. Remote Sens., 19 , 365371.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Purdom, J. F. W., 1993: Satellite observations of tornadic thunderstorms. The Tornado: Its Structure, Dynamics, Prediction, and Hazards, Geophys. Monogr., No. 79, Amer. Geophys. Union, 265–274.

    • Search Google Scholar
    • Export Citation
  • Raghavswamy, V., Pathan S. K. , Mohan P. R. , Bhanderi R. J. , and Priya P. , 1996: IRS-1C applications for urban planning and development. Current Sci., 70 , 582587.

    • Search Google Scholar
    • Export Citation
  • Ramsey, E. W,I. I. I., Chappell D. K. , and Baldwin D. G. , 1997: AVHRR imagery used to identify hurricane damage in a forest wetland of Louisiana. Photogramm. Eng. Remote Sens., 63 , 293297.

    • Search Google Scholar
    • Export Citation
  • Ramsey, E. W,I. I. I., Jacobs D. M. , Sapkota S. K. , and Baldwin D. G. , 1998: Resource management of forested wetlands: Hurricane impact and recovery mapped by combining Landsat TM and NOAA AVHRR data. Photogramm. Eng. Remote Sens., 64 , 733738.

    • Search Google Scholar
    • Export Citation
  • Rao, D. P., Gautam N. C. , Nagaraja R. , and Mohan P. R. , 1996: IRS-1C applications in land use mapping and planning. Current Sci., 70 , 575581.

    • Search Google Scholar
    • Export Citation
  • Singh, A., 1989: Digital change detection techniques using remotely sensed data. Int. J. Remote Sens., 10 , 9891004.

  • Singh, A., and Harrison A. , 1985: Standardized principal components. Int. J. Remote Sens., 6 , 883896.

  • Speheger, D. A., Doswell III C. A. , and Stumpf G. J. , 2002: The tornadoes of 3 May 1999: Event verification in central Oklahoma and related issues. Wea. Forecasting, 17 , 362381.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Srivastava, P. K., Krishna B. G. , and Majumder K. L. , 1996: Cartography and terrain mapping using IRS-1C data. Current Sci., 70 , 562567.

    • Search Google Scholar
    • Export Citation
  • Tucker, C. J., 1979: Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ., 8 , 127150.

  • Witt, A., Eilts M. D. , Stumpf G. J. , Mitchell E. D. , Johnson J. T. , and Thomas K. W. , 1998: Evaluating the performance of WSR-88D severe storm detection algorithms. Wea. Forecasting, 13 , 513518.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The postevent IRS-1C LISS-3 image taken on 8 May 1999. This is a false color composite image. In general, red indicates vegetated area; blue suggests urban area, water, roads, and tornado tracks; and green corresponds to bare soil and tornado tracks

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 2.
Fig. 2.

Images of three principal components from the LISS-3 multispectral image at 23.5-m resolution. The Oklahoma City tornado tracks show distinct signatures in the PC2 image. For display purposes, areas with higher brightness values are shown in darker shades

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 3.
Fig. 3.

Association of PC2 signatures and tornado damage in rural and urban areas. For display purposes, areas of higher brightness values are shown in darker shades

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 4.
Fig. 4.

Tornado tracks lacking PC2 signatures. The thick lines in the lower-left corner of (c) and the upper-right corner of (b) represent F4 contours. For display purposes, areas of higher brightness values are shown in darker shades

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 5.
Fig. 5.

Lack of signatures around Choctaw, Canadian River, and Chickasha in the NDVI image transformed from the LISS-3 multispectral image taken on 8 May 1999

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 6.
Fig. 6.

Comparison of signatures from NDVI and PC2 images in urban and rural areas. NDVI signatures appear wider than PC2 signatures and are responsive to some F2 damage in urban areas and F3 damage in rural areas

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 7.
Fig. 7.

More signatures of tornado damage appear in the NDVI change image. Circled areas are signatures missing in both PC2 and NDVI images. Nevertheless, these signatures are weak, and knowledge of the track proximity is helpful in identifying these weak signatures.

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 8.
Fig. 8.

Overviews of tornado damage in (a) rural and (b) urban areas. Photographs reproduced with permissions from Aerial Oklahoma, Inc

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 9.
Fig. 9.

NDVI image (a) before the tornado and (b) after the tornado, and (c) the NDVI change. (d) Typical tree damage, debris, and scoured ground in creek areas

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 10.
Fig. 10.

An example of large scoured ground and its signatures in (a) PC2 and (b) NDVI change images. (c) Photograph of debris and scoured ground at the location marked by the red arrows

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Fig. 11.
Fig. 11.

Comparison of tornado damage in open grassland and a residential area. Widespread debris in the residential area results in a much wider signature in the NDVI change image

Citation: Weather and Forecasting 17, 3; 10.1175/1520-0434(2002)017<0382:AOTDTF>2.0.CO;2

Table 1. 

Results of PCA on the IRS LISS-3 image taken on 8 May 1999. The three input layers are the green, red, and near-infrared bands

Table 1. 
Table 2. 

Statistics of PC2, NDVI, and NDVI change in F-scale zones and background area (CV: coefficient of variation = standard deviation/mean)

Table 2. 
Save