SP-399 SKYLAB EREP Investigations Summary

 

2

Land Use and Cartography

 

ROGER M. HOFFERa* R. E. JOOSTEN,b R. G. DAVIS,c AND F. R. BRUMBAUGH c

 

a Purdue University.
b NASA Lyndon B Johnson Space Center.
c Lockheed Electronics Company, Inc.
* Principal investigator.
 

[7] AN INCREASING POPULATION and a growing awareness of the finite nature of U.S. natural resources have created a demand for more effective land use planning and mapping programs. In the past decade, National, State, and local legislation has been enacted for better management of land and resources. Conflicts for alternate uses of the land are evident everywhere: urban expansion decreases land available for agriculture; increase in strip mining raises environmental concerns; forest timber production competes with recreational needs; and commercial developments can adversely affect residential communities. In examining these situations, effective land use planning requires accurate up-to-date information concerning the current use of the land and the potential capabilities of the land. Such information is often very difficult to obtain, particularly if large geographic regions are involved.

In discussing the application of the Earth Resources Experiment Package (EREP) data to land use and cartography, it is important to distinguish between these two discipline activities. The land use application includes natural resource inventories and planning relationships; the cartographic application is concerned with mapping activities. Therefore, this section has two major divisions, one devoted to land inventories, land use products, and related activities, and the other to the mapping potential of Skylab-acquired data.

The term "land use" is normally considered to include both the type of land cover and the actual use of the land as opposed to the potential use of the land or land suitability. The actual use of the land can only be inferred from remotely sensed data collected at any altitude; direct interpretation is not possible.

Most of the investigators who analyzed the Skylab EREP data for land use determination adopted a hierarchical classification scheme based on that proposed in the U.S. Geological Survey (USGS) Circular 671 (ref. 2-1). This scheme defined two major levels of land use classification (table 2-1). Level I contains nine categories that are closely related to different Earth surface features (e.g., water, urban buildup, etc.) or vegetative cover types (e.g., tundra, rangeland, forest land, etc.). In Level II, these nine categories are further subdivided, but these categories generally do not indicate the specific use of the land (e.g., deciduous forest does not indicate whether the land is being used for timber production, for recreation, or for some other purpose). Such specificity is introduced at Level III (e.g., recreational facilities) and Level IV (e.g., golf courses). To fit the varied conditions and the specific requirements involved in different geographical locations, most of the investigators modified the USGS system to meet their own particular situations and needs. Many factors were considered by the investigators in their analysis and interpretation of the EREP data for land use purposes. These factors are grouped into three major categories: the characteristics of the test site, the types of data, and [8] the interpretative and analytic techniques used. Some of these factors are discussed in the following paragraphs.

The spectral contrast and geometry of surface features within a given scene are extremely significant to the overall interpretability of a particular scene. Hannah et al. (ref. 2-2) and Stoeckeler et al. (ref. 2-3) pointed out that roads contrasted with the surrounding green vegetation distinctly and could be easily defined on the Earth Terrain Camera (S19OB) color photographs. Colwell et al. (ref. 2-4) found that, in one type of desert scene, tentative road locations that had been excavated by using a bulldozer had very high reflectance and, because of the lack of contrast of the roads in relation to the naturally occurring, highly reflective surrounding soils, were not easily detected. Thus, the ability to detect and identify roads or other features is often a function of the reflectance of the feature of interest in relation to the surrounding cover types. Spatial considerations also are important. Long linear features such as roads or powerline rights-of-way are more easily discerned than a small pinpoint feature such as an oil well, a small surface mine, or even a small water body. The contrast between the particular feature of interest and the surrounding cover types is of extreme importance in locating, identifying, and mapping such features.

Another site characteristic that must be considered involves temporal variation of data acquired in different seasons. For much of the United States, the Skylab 2 data were obtained during late spring, the Skylab 3 data during late summer and early fall, and the Skylab 4 data during winter. In many instances, the time of year at which the EREP data were collected became a critical factor in determining the effective use of such data. Two different investigations, Hoffer (ref. 2-5) and Poulton and Welch (ref. 2-6), found that the Skylab 2 data were less effective than the Skylab 3 photographic data for vegetative mapping because of the differences in the condition of the vegetation during the spring and the summer. Figure 2-1 illustrates, by color-infrared photographs, temporal variations in the San Juan Mountains, Colorado.

The level of detail in the mapping scheme to be used is a primary factor in land use classification. Many investigators found that the Level I land use categories could be mapped with a high degree of reliability. Some of the categories of Level II could be identified and mapped reasonably well; others could be mapped with....

.

 

TABLE 2-1.-A Land Use Classification System for Use With Remote-Sensor Data.

Level I

Level II

Category

Description

Category

Description

.

1.0

Urban and built-up land

1.1

Residential

1.2

Commercial and services

1.3

industrial

1.4

Extractive

1.5

Transportation, communications, and utilities

1.6

Institutional

1.7

Strip and clustered settlement

1.8

Mixed

1.9

Open and other

2.0

Agricultural land

2.1

Cropland and pasture

2.2

Orchards, groves, bush fruits, vineyards, and horticultural areas

2.3

Feeding operations

2.4

Other

3.0

Rangeland

3.1

Grass

3.2

Savannas (palmetto prairies)

3.3

Chaparral

3.4

Desert shrub

4.0

Forest land

4.1

Deciduous

4.2

Evergreen (coniferous and other)

4.3

Mixed

5.0

Water

5.1

Streams and waterways

5.2

Lakes

5.3

Reservoirs

5.4

Bays and estuaries

5.5

Other

6.0

Nonforested wetland

6.1

Vegetated

6.2

Bare

7.0

Barren land

7.1

Salt flats

7.2

Beaches

7.3

Sand other than beaches

7.4

Bare exposed rock

7.5

Other

8.0

Tundra

8.1

Tundra

9.0

Permanent snow and icefields

9.1

Permanent snow and icefields

 


[
9]

FIGURE 2-1.

FIGURE 2-1.-S193A color-infrared photographs showing seasonal differences in vegetation for the San Juan Mountains of southwestern Colorado. This type of seasonal scene variation in vegetative condition was significant in data interpretation for many Skylab investigations. (a) Photograph taken in June 1973 (SL2-09-17). The area shown in blue tones indicates the general absence of green vegetation. (b) Photograph taken in August 1973 (SL3-21-331). The areas shown in several different red hues indicate healthy vegetation. [For a larger picture, click here]

 

Only moderate success. It was also found that different definitions influenced the apparent results. In a few cases, similar or even identical terminology was being used by different investigators to indicate different levels of difficulty in the mapping process.

The data used in land use mapping were of prime importance, in terms of both the analytic techniques used and the types of results that could be achieved. For land use mapping, the primary data were collected by the Multispectral Photographic System (S190A), the Earth Terrain Camera (S190B), and the Multispectral Scanner System (S192). The inherently better spatial resolution of the S190A and S190B photographs as compared to the S192 imagery permitted flexibility for the enlargement of the Skylab photographs. Despite the relatively small scales of the original photographs (approximately 1:2 900 000 for the S190A and 1:950 000 for the S190B), more usable scales were readily obtained by enlarging the original photographs by using photographic processes or viewing devices. It was demonstrated that enlarged scales ranging from 1:250 000 to 1:50 000 were feasible and practical for most applications. Fidelity of interpretative detail was excellent for this scale range. For the higher resolution films of the S190B camera, a scale of 1:50 000 is probably the largest for maximum usefulness; however, projections of S190B transparencies to scales of 1:24 000 and larger were used in detailed land use analysis.

 

[10] General Land Use Results Obtained by Using S190/S192 Photographic Data

 

The types of EREP land use investigations ranged from academic research and theoretical examination of specific photographic products and analytic techniques to direct involvement with county planning agencies. The research-oriented investigators sought to determine why certain features on the Earth's surface appear as they do on certain types of film and how additional information can be derived through innovative interpretative methods. The direct-application investigations used the photographs to extract information that could be applied in the solution of specific problems.

In the evaluation of S19OA and S190B film types for land use analysis, most investigators preferred the S19OB color photographs because of the greater spatial resolution, but other investigators indicated the need for the better spectral discrimination provided by the color-infrared film for most land use categories (mainly cover-type features). For discrimination of urban categories, color film was most useful; but, for nonurban categories, the color-infrared data were better (fig. 2-2). These conclusions were substantiated by investigators who used the S19OB color films as the prime data source and the S19OA color-infrared data as complementary sources of information to improve the classification of natural features such as forest land, agricultural crops, rivers and lakes, and wetlands. Figures 2-3(a) and 2-3(b) illustrate how these two film types were used in land use planning and resource inventories.

Only one investigation (Hardy et al., ref. 2-7) involved the analysis of S19OB black-and-white photographs for land use activities. Although this type of photograph contains the best spatial information, effective separation of Level II land units was impossible because of a lack of tonal characteristics.

The two black-and-white film types (panchromatic and infrared) used in the S19OA camera system showed less information spectrally than either of the color-film types. The advantage of black-and-white multispectral coverage in the green, red, and infrared wavelengths is that different spectral responses of some land use categories can enhance feature identification from a comparative signature analysis. Additionally, each of the black-and-white bands can be combined with appropriate filters in the photoprocessing laboratory to produce a color-composite scene or can be enhanced by a variety of other additive-color techniques. Spatially, the red wavelength of the two black-and-white panchromatic films was judged best, but it lacked the tonal qualities required to separate some land units for identification.

General land use mapping with use of S190 data.- Most land use investigators produced some type of land use map as a final product by using a variety of photographic data and extraction techniques. Figure 2-4, derived by photointerpretation of imagery and photographs, illustrates the level of land use information that can be obtained from different types of platforms and sensors. Cooper et al. (ref. 2-8) compiled these land use maps at a common scale of 1:63 360, using Landsat, Skylab S19OA and S19OB color-infrared, and RB-57/RC-8 aircraft imagery as the data sources. The interpretations were accomplished primarily from an analysis of transparencies that verified the discrimination between tonal characteristics and mapping units. In terms of hours, the compilation of these land use maps for the same area coverage required 1.5 hours for Landsat, 4 hours for S19OA, 8 hours for S19OB, and 10 hours for aircraft data. When these compilation times are extrapolated to full-frame interpretation for each different format size, it is evident that satellite platforms offer a distinct advantage over aircraft coverage. It must be emphasized that although the time required to compile land use maps from Landsat and S19OA photographs is considerably less, the quantity and quality of information for the S19OB- and aircraft-data-derived land use maps are markedly superior (fig.2-4). The S19OB photographs, in particular, were found to be almost ideal for detailed as well as regional land use maps.

A broad assortment of land use products that vary in scale, type of data used, and extent of area involved is found in the reports of the individual investigators. One of these products is the detailed map of Alachua County in north-central Florida (Hannah et al., ref. 2-2), compiled from S19OB color-infrared photographs at a scale of 1 :40 000. This map was reviewed for completeness and accuracy by staff members of the North....

 


[
11]

FIGURE 2-2.

FIGURE 2-2. Comparison of S190B color and color-infrared film enlargements acquired over the Newburgh, New York, area. The higher spatial resolution of the color film (fig. 2-2(a)) is readily apparent. The color-infrared film (fig. 2-2(b)) is useful in discriminating vegetation (red) and bodies of water (black). (a) Color film (SL3-88-274). (b) Color-infrared film (SL3-87-300). [For a larger picture, click here]


[
12]

FIGURE 2-3.

FIGURE 2-3.-A comparison of spatial and spectral characteristics of S190B color film and S190A color-infrared film. This scene, showing the rapidly developing urbanized corridor between Washington, D.C., and Baltimore, Maryland, is representative of the variation in spatial resolution and spectral sensitivities between Skylab photographic systems for enhancing particular land use features. For example, urban detail is visible on the S19OB photograph (fig. 2-3(a)), whereas excellent enhancement of water features and improved discrimination within forested lands are provided by the S190A color-infrared photograph (fig. 2-3(b)). Photographs were taken in August 1973. (a) S190B color photograph (SL3-83-166). (b) S190A color-infrared photograph (SL3-15-172). [For a larger picture, click here]


[
13]

FIGURE 2-4.

FIGURE 2-4.-A comparison of Landsat-, Skylab- (S190A and S190B), and aircraft-derived land use maps. This figure represents Level I and Level II categories mapped in the vicinity of Newburyport, Massachusetts. The land use maps illustrate the type of mappable units that can be derived from imagery obtained by using three different platforms and four different sensing systems. The land use classification system used was modified from reference 2-1. Specific colors represent Level I categories. Numbers represent subcategories, or Level II Informations, within the Level I categories. Hachuring is used in a few areas where positive identification could not be made for a single Level I category. The close correlation between specific-category patterns and areal extent on the Skylab S190B and aircraft maps should be noted. (a) Political map. (b) Landsat color-composite land use map. (c) Skylab S190A color-infrared land use map. (d) Skylab S190B color-infrared land use map. (e) Aircraft color-infrared land use map. [For a larger picture, click here]


[
14]

Figure 2.4.- Continued (b) (c).

Figure 2.4.- Continued (b) (c). [For a larger picture, click here]


[
15]

Figure 2.4.- Concluded (d) (e).

Figure 2.4.- Concluded (d) (e). [For a larger picture, click here]

 

[16] ....Central Florida Regional Planning Council, who indicated that the mapping accuracy and the quality of detail were suitable for both regional and county planning endeavors. It was also compared with a recently released USGS land use map of the same area compiled from aircraft photographs, and only a few discrepancies were noted. Classification of the total county area varied by only 3 percent.

Coastal land use mapping from the S190B photographs was accomplished for Delaware and Maine. Klemas et al. (ref. 2-9), using S19OB color photographs and a zoom transfer scope, compiled a map with a scale of 1:125 000 that delineated 10 land use and vegetation categories. Color-infrared S19OA photographs of the Truk Watershed in Maine were used to provide additional definition of water boundaries and vegetative species.

Stoeckeler et al. (ref. 2-3) indicated that, in Maine, the S19OB color-infrared data were valuable in delineating vegetation cover types and cultural features even though the only coverage available for evaluation was a snow scene (January 1974; fig. 18 from Woodman and Farrell's report in ref. 2-3). Logging roads that were approximately 6 m wide were identified. Several vegetation-cover-type maps were prepared from the S19OA film types collected in September 1973. Vegetation types, especially forest stands, were best interpreted from color-infrared film, with black-and-white panchromatic film showing good delineations of most categories. The remaining film types contributed little additional information to this overall evaluation of general land use or vegetation cover-type mapping.

Simonett (ref. 2-10) also cited several examples that indicated that good-quality land use maps could be derived from EREP data. One such example was the regional land use map (Level II) prepared at a scale of 1:126 720 from S19OA (black-and-white and color-infrared) photographs for St. Marys County, Maryland. Interpreting the data, mapping the features, and preparing the final map of the 1088-km2 area required 14 hours. By comparison, a similar map prepared from high-altitude-aircraft color-infrared data at the same scale required 32 hours for completion. Additional detailed information was derived from the analysis but could not be adequately displayed on a map of this scale.

Hardy et al. (ref. 2-7) used a low-cost photographic enhancement process to convert black-and-white enlargements to color composites with use of the techniques discussed in section 6. Combinations of spectral bands, diazo hues, and exposures were selected to maximize the color contrast among the land use categories to be examined (figs. 2-5(a) and 2-5(b)). it should be noted that this process is used to achieve unique descriptions of classification units by means of color contrast; therefore, individual colors are not chosen to represent specific land uses. In a comparison of three test sites in New York, using the 1968 land use natural resource (LUNR) inventory as a data base, Hardy determined that more land use information was provided by the S19OA enhanced color composites than by the individual S19OA film types. In addition, the composites were nearly as effective as the individual S19OB film types for classification. Accuracy ranges related to the USGS Level I and Level II categories for a scale of 1:62 500 are shown in table 2-II.

Table 2-III is representative of specific Level I and Level II accuracies achieved by comparing S19OA color composites and S19OB film types to the existing LUNR base. Accuracies were determined in terms of square hectometers, which were inventoried by using film from Skylab camera systems. Random ground checks of the LUNR data base, which was compiled in 1968...

 

 

TABLE 2-II.-Accuracy Ranges for Level I and Level II Categories Calculated by Using Different Films and Techniques.

 

Film

Accuracy, percent

.

.

Level I

Level II

.

S190A enhanced color composites

88 to 93

71 to 78

S190B three film types

87 to 95

75 to 81

 


[
17]

FIGURE 2-5.

FIGURE 2-5.-S190A color composites of Tompkins County, New York, showing different color enhancements that assist in the interpretation of selected categories. These examples represent only two of several color composites that were generated for this test site. Lake Cayuga and the city of Ithaca are two prominent features in the scene. (Approximate scale, 1:250 000.) (a) S190A color composite designed to enhance natural features. (b) S190A color composite designed to enhance cultural features. [For a larger picture, click here]

 

[18] ...from 1:20 000-scale black-and-white aircraft photographs on overlays keyed to the 7.5' topographic map base, indicated an overall accuracy of 95 percent, based on the 1973 inventory for Tompkins and Suffolk Counties. However, for Orange County, the accuracy was 84 percent. If these adjusted figures were applied to the S190A data, table 2-III values would increase the Skylab inventory from 93 to 98 percent for Level I and from 78 to 83 percent for Level II, with comparable increases for the S190B data.

The numbers of classification units and the definitions used in the LUNR system are unrelated to the USGS classifications for Levels I and II. Several LUNR land use units had to be aggregated to provide the comparison presented in table 2-III. Difficulty was encountered in discriminating between light residential (Level III) and forested categories and between cropland and pasture. The reason for the difficulty was attributed, at least in part, to similar spectral responses for these categories. As previously mentioned, the season of data collection is significant in the inventory of natural resources. In this case, spring or late-fall coverage would have improved the overall classification results in the Level II categories. Hardy et al. (ref. 2-7) ranked the S190A and S190B sensors in terms of interpretation preferences as (1) S190B color, (2) S190A color composites, (3) S190B black and white for spatial patterns, and (4) S190B color infrared. The S190B black-and-white data had the best resolution properties, but tonal distinction for the various land use categories was often not sufficient to provide a high level of confidence for interpretability.

Agriculture, range, forest, and water-resource categories are discussed in detail in other sections of this report. To complete the representation of the major categories in land use, attention in the next part of this subsection is directed to the application of sample data to the urban category (with strip mining, wetlands, and other specific categories emphasized later).

Urban land use mapping with use of S190 data.-Primarily because of the high spatial resolution of the S190B system, almost all the investigators concerned with urban environment used this system for detection, identification, and mapping of urban categories. Table 2-lV shows a typical Level III classification unit within the "urban and built-up land" category. This table represents a compendium of the work of several investiga-....

 

TABLE 2-III.-Comparison of Skylab S190A (Multispectral) and S190B (Color) Errors to 1968 LUNR Data for the Riverhead-Southampton, Suffolk County Test Area. a

 

Category

1968 LUNR area hm2

Error, hm2

Relative error

S190A

S190B

S190A

S190B

.

Level I

.

1.0

9 239

-155

697

-0.02

0.08

2.0

12 818

2834

2369

.22

.18

4.0

17 737

3715

1527

.21

.09

5.0

18 575

55

-339

.00

-.02

6.0

1 142

-1054

- 145

-.92

-.13

7.0

484

368

743

.76

1.54

.

Aggregate error

0.07

0.05

.

1.1

5 065

-4693

-4285

-0.93

-0.85

1.2

261

7251

6123

27.78

23.46

1.4

381

-381

43

-1.00

.11

1.5

1 036

-348

-80

-.34

-.08

1.6

2111

- 1599

-859

-.76

-.41

1.7

384

-384

-344

-1.00

-.90

2.1

7 494

1626

2894

.22

.39

2 2

375

-367

-375

-.98

-1.00

2.4

4 949

-4093

-4888

-.83

-.99

4.2

16

-16

-16

-1.00

-1.00

4.3

17 721

3731

1543

.21

.09

5.1

111

-111

-87

-1.00

-.78

5.2

167

-147

93

-.88

.56

5.3

155

-155

-155

-1.00

-1.00

5.4

18 142

118

-190

.01

-.01

6.1

921

-837

-24

-.91

-.03

6.2

220

-216

-120

-.98

-.55

7.2

484

368

633

.76

1.3

7.4

0

0

80

0

.

Aggregate error

0.22

0.19

a From reference 2-7.

 

[19] Table IV.- Urban Features Discernible on Skylab S190B Photographs.

Level I

Level II

Level III

Qualitative evaluation of Level III categories

.

1.0 Urban and built-up land

1.1 Residential

1.1.1 Single-family household units

(a)

1.1.2 Multiple-family household units

(b)

1.1.3 Mobile home parks

(b)

1.1.4 Transient lodging

(c)

1.2 Commercial and services

1.2.1 Wholesale trade areas

(b)

1.2.2 Retail trade areas

(b)

1.2.3 Business, professional, and personnel services

(d)

1.2.4 Cultural, entertainment, and recreational facilities

(b)

1.3 industrial

1.3.1 Major manufacturing plants

(b)

1.3.2 Distribution centers

(c)

1.4 Extractive

1.4.1 Stone quarries

(c)

1.4.2 Sand and gravel pits

(c)

1.4 3 open-pit or strip mining

(a)

1.4.4 oil, gas, sulfur, salt, and other wells

(c)

1.5 Transportation, communications, and utilities

1.5.1 Highways and related facilities

(a)

1.5.2 Railroads and related facilities

(b)

1.5.3 Airports and related facilities

(a)

15.4 Marine craft facilities

(b)

1.5.5 Telecommunications and related facilities

(c)

1.5.6 Electric, gas, water, sewage disposal, solid waste, and related facilities

(c)

1.6 institutional

1.6.1 Educational facilities

(b)

1.6.2 Medical and health facilities

(c)

1.6.3 Religious facilities

(c)

1.6.4 Military areas

(b)

1.7 Strip and clustered settlement

(No further breakdown)

(b)

1.8 Mixed

(No further breakdown)

(b)

1.9 Open and other

1.9.1 improved

(a)

1.9.2 Unimproved

(b)

a Identification determined with ease
b Identification possible often enough to make data useful without collateral data.
c Recognizable by geometry, texture, color, and/or alinement but not positively identifiable unless correlated with aircraft underflight or ground-truth data.
d Cannot be recognized

 

.....-tors and offers a qualitative evaluation of the S190B photographs as they relate to urban features. Figure 2-6 is an S190B color enlargement of Jackson, Mississippi, that is representative of the urban-type information present in such photographs. Photographs of this type can he enlarged to show the spatial relationships of individual features with the adjacent environments (1:125 000 scale) or can be effectively enlarged to common data-base scales (e.g., 1:50 000 or 1:24 000) by photographic processes or viewing devices to enable more detailed urban analysis and general interpretation.

The type of urban land use interpretation accomplished by Alexander and Lins (ref. 2-11) for the Phoenix, Arizona, and New Haven, Connecticut, areas is shown in figures 2-7 and 2-8, respectively. A series of 20- by 20-km land use map sections was compiled by using 18- and 30-power microfiche viewers and transferring the land use polygon boundaries to scaled overlays....

 


[
20]

FIGURE 2-6.

FIGURE 2-6.-A portion of an S19OB color photograph (approximate scale, 1:50 000) showing the central city area of Jackson, Mississippi (SL4-92-295). Selected annotations show several Level III and Level IV land use categories. Transportation features and related components are visible but have not been annotated. [For a larger picture, click here]


 

[21] ...on which land parcels of 4 km2 or larger were mapped. Their evaluation indicated that the S19OB color photographs permitted the detection of higher levels of land use detail than any satellite imagery previously evaluated by using photointerpretative techniques. These areas are also part of the "Atlas of Urban and Regional Change" being produced in the USGS Geography Program.

The following paragraphs summarize the findings concerning the major Level II categories and several subcategories (Level III) for the Phoenix, Arizona, and New Haven, Connecticut, studies.

Residential: The residential land for both sites was identified on the S190B photographs and subdivided into single- and multiple-family categories. The individual signatures on the photographs differ somewhat between sites as a function of roof composition, quantity and type of vegetation, time of year, lot size, and block arrangement.

Commercial: Several types of commercial areas were identified. These types include central business districts, strip developments, and suburban shopping centers. Strip commercial developments extend along major transportation arteries and are usually no more than one block deep on either side of the road. Suburban shopping centers can be easily identified along major suburban highways by their large, brightly reflecting roofs and large, paved parking lots.

Industrial: Very large structures usually characterize the industrial category. Industrial activity commonly appears in groups or clusters as a mixture of crowded large buildings, fuel storage tanks, and numerous railroad sidings. These facilities resemble shopping centers except that many large buildings are present and parking areas are not oriented for convenient access to the buildings.

Transportation: The S190B photographs are valuable in delineating various types of transportation facilities. Major highways and interchanges are clearly visible and many lesser roads can be observed (although, in older residential areas, streets tend to be obscured by vegetation). Railroad yards, rights-of-way, and sidings have been identified. Airports with paved runways, aprons, and parking areas and with terminal buildings are readily evident. Many utility corridors (i.e., powerlines and pipelines) are visible, especially where they pass through forested land.

Institutions: institutions appear in both scenes and consist primarily of educational facilities and large medical complexes. They appear as groups of long, connected buildings surrounded by extensive vegetated areas, parking lots, and-in some cases-athletic fields.

Open space: The open-space category includes improved open spaces such as golf courses, cemeteries, parks, and vacant lots. These land uses would be classified as Level IV in this classification matrix. The fairway patterns of golf courses are the easiest to identify within this category.

Industrial and commercial complexes: Although not a listed category under Level III, these "industrial parks" were identified on S190B photographs. They usually consist of a mixture of industrial and commercial land use, including light assembly, regional distribution facilities, and research and development sites. These "parks" exhibit none of the features associated with heavy industry, such as fuel tanks, railroad sidings, and piles of raw materials. They are usually located along, or at junctions of major highways.

Additional, miscellaneous activities.-As part of the USGS Geography Program, Skylab S190B color photographs were compared to high-altitude-aircraft photographs collected in 1970 for selected urban scenes. The major objective of this study was to determine whether the S190B color photographs could be used effectively to determine the types of post-1970 land use changes that had occurred in the 3-year interval. This analysis was therefore one involving "change detection."

Several different urban scenes were studied to determine the level of detail that could be identified for very different environmental areas, extending from a desert setting to a forested setting. Only photointerpretation techniques were used to perform the analysis. A series of overlays containing different land use categories was prepared. When these overlays were superimposed on the 1970 aircraft-data-base land use maps, the areas of change were detected (fig. 2-9). The findings of this investigation were very positive. The resolution of the S190B photographs permitted detection of many Level III and some Level IV categories; it approached the quality of the high-altitude-aircraft color-infrared photographs, especially for the identification and mapping of urban changes. In figure 2-9, large parcels of agricultural land and rangeland that are undergoing changes were measured. From 1970 to 1973, many new....

 


[
22]

FIGURE 2-7.

FIGURE 2-7.-A portion of a land use map (fig. 2-7(b)) derived from an S190B color photograph (fig. 2-7(a)) for Phoenix, Arizona. This map is an example of a Level II and Level III land use map compiled at a scale of 1:100 000 for the Phoenix test site. The markedly different geographical settings (desert as opposed to vegetated landscapes) and the land use patterns should be noted and compared with figure 2-8. (a) S190B photograph (SL3-86-011). (b) Land use map. [For a larger picture, click here]


[
23]

FIGURE 2-8.

FIGURE 2-8.-A portion of a land use map (fig. 2-8(b)) derived from an S190B color photograph (fig. 2-8(a)) for New Haven, Connecticut. This example of Level II and Level III land use categories represents the type of information that can be useful for regional planning and land use inventories. This map was compiled by using a 1:100 000-scale S190B photograph taken September 19,1973. The overall scene contrast and land use patterns of this figure should be compared with those of figure 2-7. (a) S19OB photograph (SL3-88-276). (b) Land use map. [For a larger picture, click here]


[
24]

FIGURE 2-9.

FIGURE 2-9.-A portion of a Level Il and Level Ill land use change map for the Phoenix test site. A dot planimeter was used for area measurement to permit calculation (in percent) of total amount of change in the test site for each category. (Fig. 2-7(a) contains a photographic display of this area.) [For a larger picture, click here]

 

....residential areas were developed and were mapped on the S190B photographs. Areas that showed evidence of some activity but that could not be accurately classified were placed into the transitional category (symbol Y in fig. 2-9). Mapping of nonurban land use categories, however, was more difficult with the use of S190B color photographs. For example, forests tended to be uniformly green on the natural color film, and wooded residential areas were difficult to discriminate in some cases. Repetitive temporal coverage with use of a high-resolution color-infrared film in the S190B camera would improve the interpretability of these categories.

In a classification accuracy study, Alexander and Lins (ref. 2-11) used the S190B color-film data to produce a 1:24 000-scale land use map of Fairfax City, Virginia. This map product was evaluated against a similar map prepared from high-altitude-aircraft (U-2) color-infrared photography at the same scale and was field checked for accuracy. An accuracy of 83 percent was achieved with the EREP data for mapping Level III land units when compared to the aircraft data base. To quantify the results, Alexander used two methods to determine the accuracy of the resultant maps. First, a systematically alined sample (grid cells) of 69 sample points was examined; and second, the area measurements (square hectometers) of the land units were compiled for both the Skylab and the aircraft data maps. Of the 69 sample points, 57 were correctly classified. Of the units that were misclassified, nearly half the error was [25] attributable to the lack of spectral discrimination in the color Elm. Full foliage cover made detection of houses or other residential "keys" impossible. This difficulty is also true for high-altitude-aircraft data.

Another source of error was the difficulty of identifying and mapping the unimproved open-space category. These areas were confused with small plots of agricultural land. In a summary of the S19OB data, Alexander and Lins (ref. 2-11) indicated an ability "to distinguish and map with considerable confidence such structural urban details as the location and extent of most single-family residential areas, even some residential structures themselves, commercial and industrial areas, even individual commercial and industrial structures, streets and roads of moderate size and considerable detail in the use patterns of surrounding nonurban land. If color infrared film of comparable spatial resolution to that of color film used in this evaluation had been available the investigators are confident that even greater detail and reliability of detection of the various land use categories would have been obtained.... The Skylab S190B data here revealed a capability to distinguish Level III and in some cases, Level IV (trailer parks, tank farms, golf courses, drive-in theaters, cemeteries, etc.) in urban area land use analysis." Results of other investigators (Hannah et al., ref. 2-2; Simonett, ref. 2-10; and Baldridge et al., ref. 2-12) generally supported these findings.

When a suitable inventory data base (such as an aircraft mosaic) is available for an area, it is quick and easy to update that data base by using S19OB-type photographs. Although the S190B color photographs provided the most meaningful source of information to update inventory bases in most land use inventories, the multispectral characteristics of the S19OA also proved valuable. In several cases, the information provided by S190A film types was extremely beneficial to the analysis as a supplementary source for identification and delineation of specific nonurban categories. For example, in forested or generally vegetated scenes for which S19OB color film was the primary source of detailed information, the S19OA color-infrared film provided additional spectral information.

The pattern and extent of urban expansion into prime agricultural land were mapped for Columbus, Ohio, by Baldridge et al. (ref. 2-12), with the use of S19OA photographs (fig. 2-10). With use of the map data, percentage figures were calculated to determine the rate of encroachment. This type of analysis, when coupled with long-range population density and pattern projections, can be meaningful to urban and regional planners and to decisionmakers; it can also be used as an input to test and update projection-type models.

In other urban-related activities, Welby and Lammi (ref. 2-13) used Skylab photographs for studying environmental issues involved with the expansion of the Raleigh-Durham, North Carolina, airport. Pressure to expand airport facilities (runway extensions, orientation of new runways, etc.) created problems concerning effective planning for this expansion with the least amount of adverse effect on the local area. Although the decisionmaking process for this type of problem is complicated and is influenced by social, political, and economic factors, Welby and Lammi concluded that Skylab photographs, even when used by a relatively inexperienced photointerpreter, are useful for this type of environmental analysis.

Another urban environmental problem in which Skylab photographs were used by Welby and Lammi involved a "greenspace" study. In urban planning, a greenspace (or greenbelt) is defined as an area of land covered with some form of vegetation. Planners are concerned with this type of area in terms of how to protect and manage it effectively and how to obtain additional areas when needed. The S19OB color photographs, enlarged to a scale of approximately 1:62 500, provided the best source of and format for, relevant information for greenspace analysis.

Welby and Lammi summarized their use of S19OB color photographs with the following statements.

 

Land use, vegetative cover, and even the relative beauty and ugliness of the urban landscape can be seen or inferred from the photographs. The proximity and encroachment of commercial-industrial development into good quality residential areas, the tendency of many new residential developments to become open bulldozed biological deserts are examples of the environmental quality problems visible in Skylab Photography.

 

Another use of EREP data was explored by digitizing S19OA and S19OB photographs for generating land use classification maps. Hannah et al. (ref. 2-2) oriented their approach toward a direct application for urban planners, whereas Silva (ref. 2-14) directed his effort toward a quantitative, technical evaluation by comparing the digitized photographic results with Landsat and Skylab multispectral scanner results.

 


[
26]

FIGURE 2-10.

FIGURE 2-10.-Map of Franklin County, Ohio, showing urban encroachment on agricultural land in the Columbus, Ohio, area. This map graphically portrays the areal extent and pattern of urban encroachment into three predominantly agricultural townships surrounding Columbus. Conventional photographic techniques with the use of S190A photographs were used to plot the limit of the 1973 built-up area. [For a larger picture, click here]

 

[27] For the Gainesville, Florida, site, Hannah et al. generated a Level II classification map from three bands of digitized S19OB color-infrared multi-emulsion data. The result indicated that identification of various land classes on the basis of color tones can be accomplished more effectively by the human observer (photointerpretation); examples include identifying forest types and other vegetation types, such as waterhyacinths and marshlands, and relating land use units to their surroundings. By contrast, the maps produced with computer assistance reveal more details involving commercial-industrial classes because of the capability of machine processing to improve the degree of brightness of the spectral responses.

Silva digitized both S19OA color-infrared (multiemulsion) and S19OA multiband data, using the four-band black-and-white infrared films obtained over the Lake Monroe, Indiana, site. A classification performance analysis based on training fields for nine land use classes was performed. Table 2-V shows the results compared to the Skylab S192 and Landsat multispectral scanner data for the same area. Although the performance levels were lower than those of the best four bands of the Skylab multispectral scanner (bands 3, 7, 8, and 11) and those of the Landsat multispectral scanner bands, the multiemulsion data were better than the digitized multiband black-and-white data.

Both investigators indicated that, although the achieved results were undramatic, digitized photography is a technique for which research and development should be continued for land use resource and inventory programs.

In the following subsections, land use results obtained by using S192 imagery and computer-aided analysis techniques are discussed.

 

 

TABLE 2-V.-Classification Performance Results: Digitized S190A Photography Compared to Data From Two Multispectral Scanners for Selected Categories

[percent correct]

Category

Skylab S192

Landsat multispectral scanner

Skylab S190A

.

.

Bands

Bands

.

Color Infrared

Four-band black and white

3, 7, 8, 11

3, 5, 6, 8

.

Residential

97

81

97

91

84

Commercial-industrial

73

33

61

76

46

Extractive

51

59

61

32

34

Soil

87

78

83

67

78

Grass

95

86

93

82

69

Sparse woods and deciduous forest

81

80

86

84

77

Coniferous forest

99

68

95

85

43

River

87

27

77

16

64

Lake

89

86

86

98

93

.

Class average a

84

66

82

70

65

Overall performance b

87

80

88

83

76

a Arithmetic mean of the performance results of the nine classes
b Total number of points classified correctly divided by total number of points in test areas, times 100.

 

[28] General land use mapping with use of S192 data.- Several investigators (Hannah et al., ref. 2-2; Hoffer, ref. 2-5; Klemas et al., ref. 2-9; Simonett, ref. 2-10; Silva, ref. 2-14; Gilmer and Work, ref. 2-15; Higer et al., ref. 2-16; Polcyn et al., ref. 2-17; and Sattinger et al., ref. 2-18) used computer-aided analysis techniques for the study of Skylab S192 data obtained over many test sites to map a variety of different types of land cover. The results, in general, indicated a considerable potential for mapping various land cover types through use of these techniques, even though the procedures used to train the computer, the classification algorithms, and the methods used in evaluating the results varied considerably. Some of the investigators conducted detailed analyses to determine the wavelength bands that were most valuable for mapping different cover types. These results show the value of the spectral range of the S192 system. Because some studies also involved analysis of Landsat-l data obtained over the same test site as the EREP data, valuable comparisons of results obtained by using two very different satellite multispectral scanner systems were possible.

Most of the investigators used the USGS Circular 671 (ref. 2-1) system of land use classes in their analyses but modified it to meet their own particular requirements. In many studies, both Level I and Level II degrees of detail were mapped; and in some investigations, more detailed classifications were achieved for selected cover types. For the Level I cover types, the classification accuracies ranged from approximately 72 to 91 percent. For the Level II degree of detail, the results were more varied, with overall classification accuracies ranging from 43 to 89 percent. The exact reasons for these variations in results cannot be specifically determined. It is difficult to isolate the reasons for the variations because the different investigations involved diverse test sites that included a wide variation of cover types. Distinct differences in the analysis and evaluation techniques also must be considered.

The following paragraphs provide some insight into the similarities and differences among the many investigations by summarizing key aspects of the various studies. Only those investigations involving land use and land cover mapping by computer-aided analysis techniques are included here. Unless otherwise noted, each of these studies involved S192 data that had been digitally filtered and line-straightened at the NASA Lyndon B. Johnson Space Center data-processing facility.

A study by Simonett (ref. 2-10) involved the analysis of S192 data over a predominantly urbanized test site in the Washington-Baltimore region. The data were obtained on August 10, 1973, and all 13 wavelength bands were available for analysis. Extensive fieldwork provided the information for identifying many areas that could be located in the S192 imagery. A clustering technique was used to help define spectrally homogeneous training and test areas. The classification was performed by using a maximum-likelihood algorithm similar to that used by several other investigators (Polcyn, Silva, Hoffer, and Sattinger; refs. 2-17, 2-14, 2-5, and 2-18, respectively). A major difference in Simonett's approach was that a two-stage classification sequence involving different combinations of wavelength bands was used. This approach differed from that used by the other investigators because they generally defined a single optimal set of wavelength bands for classifying the data at the Level II degree of detail. The results of this Level II classification were then grouped into broader categories to display and tabulate Level I results.

Simonett defined five major Level I classes in the Washington-Baltimore test site: urban, agricultural, forest, water, and wetlands. The results were quantitatively evaluated by using a series of test areas that included approximately 13 600 picture elements (pixels), of which more than half (7000) were in the urban category. Simonett reported a 72-percent classification accuracy for the Level I classification of the test areas. This result compared with 73 percent for the training data sets and indicates that a fairly good statistical sample existed for both the training data sets and the test areas. Results obtained by using a combination of the training and test pixels (approximately half of each) are shown in table 2-VI. The Level II classification results yielded a classification accuracy of approximately 43 percent. Significant misclassification occurred at the Level II degree of detail in the agriculture and forest cover types, although other investigators reported reasonably good classification performances in similar categories of cover type. In some study sites, the number of training and test areas was insufficient to truly represent the various categories involved at this level of detail. As a result of this relatively poor classification performance at the Level II degree of detail, Simonett concluded that, in this particular study, only the Level I classification results would be of potential value for land use planning agencies. This general land use [29] investigation also included a detailed evaluation of spectral bands for land use studies. The results of similar evaluations are discussed in the subsection entitled "Urban land use mapping with use of S192 thermal-infrared data." Simonett's conclusions, together with those of other investigators, are discussed in the subsection entitled "Wavelength band evaluation."

The Brevard County Planning Department in Florida conducted a land use investigation that included computer-aided analysis of the S192 data (ref. 2-2). A supervised approach was used to develop the training statistics for the analysis; and, as in other investigations, a maximum-likelihood-ratio algorithm was used for the classification. To obtain output products for land we planners, the generalized boundaries of computer-generated land use classes were manually delineated. The investigators indicated that acceptable results were obtained and that multispectral scanner mapping was a useful tool. In general, the computer classification maps tended to have more precise quantitatively defined land use patterns, whereas results from photointerpretation tended to generalize some of the patterns. This difference resulted in an overestimation of forest areas on the photointerpreted results as compared to the computer-derived results.

In a study by Polcyn et al. (ref. 2-17), the S192 data were used to classify an area in southern Ontario, Canada, largely covered by vegetation. The cover types involved in the classification included marsh, conifer/hardwood, hardwood/orchard, undifferentiated vegetation (including brush, idle, etc.), suburban, bare soil/quarry, herbaceous vegetation, and water. Only four wavelength bands of the S192 data from the Skylab 2 mission (0.56 to 0.61, 0.62 to 0.67, 0.78 to 0.88, and 10.2 to 12.5 µm) were available for analysis. The training statistics were obtained by applying two successive clustering-step analyses. The classification was based on a maximum-likelihood algorithm, as incorporated into the software system. Because of the difficulty in developing an adequate set of test areas for evaluating the classification results, Polcyn developed a different approach for quantifying the results. This approach involved defining the "probability of correct classification" for the eight categories mapped, based on a statistical analysis of the training data. The results indicated that all categories except two had probabilities of correct classification above 90 percent. (Undifferentiated vegetation and suburban categories had probabilities of correct classification of approximately 78 percent.) The training statistics that were used as a basis for these quantitative evaluation figures represented the best classification statistics. Polcyn also noted that the analysis indicated a likely misclassification of the suburban category as either "brush" or "bare soil"; this conclusion is not surprising because a suburban area is often composed of a mixture of cover types that would include these spectral classes.

 

 

TABLE 2-VI. Skylab Classification Results for the Heavily Urbanized Baltimore-Washington Area

 

Ground-truth category

No. of pixels

Skylab classification results, percent

Percent classified correctly

1000: Urban

2000: Agricultural

4000: Forest

5000: Water

6000: Wetlands

.

1000:

Urban

12 993

71.3

19.3

7.4

0.5

1.5

71.3

2000:

Agricultural

7 451

17.3

69.4

11.3

.1

1.9

69.4

4000:

Forest

1 713

4.0

5.1

84.1

1.1

5.7

84.1

5000:

Water

1549

1.0

.3

.3

96.3

2.1

96.3

6000:

Wetlands

928

4.4

9.3

33.6

6.6

46.1

46.1

.

Total

.

24 634

-

-

-

-

-

a72.2

a percentage reflects correct classification of 17 796 pixels

 

 

[30] Silva (ref. 2-14), using unfiltered S192 data obtained on June 10,1973, over the Lake Monroe area in south-central Indiana, conducted a land use study that identified residential, commercial-industrial, extractive, soil, grass, deciduous forest, coniferous forest, river, and lake. (Note that, in some instances, the investigators were working with land cover characteristics rather than land uses, such as "soil," which would actually be in the agricultural cropland land use category.) A combination of supervised and clustering analysis techniques was used to develop the statistics for classification. A divergence processor was used to define the best combination of 4 of the 12 wavelength bands available for use. This process indicated that bands 3,7,8, and 11 represented the best combination. It appeared that these wavelength bands also generally had the best data quality among those available. Therefore, there was some concern as to whether those bands were being selected for their spectral characteristics or for their data quality or for a combination of both. The classification was conducted by using a maximum-likelihood ratio based on the multivariant Gaussian distribution as incorporated into the software. An overall classification accuracy of 87 percent was obtained, based on tabulation of the results in the test areas. In this particular set of results, the test areas selected did not represent a statistically defined array of test areas, and Silva (ref. 2-14) stated that these performance figures might have been somewhat biased in an upward direction. It is important to note that this result is based on a Level II degree of detail; therefore, it does indicate a potentially significant improvement over the classification accuracy reported by some of the other investigators for this level of detail. A map display of these classification results is shown in figure 2-11.

Additional analyses of the S192 data over south-central Indiana were performed by Silva, using a statistically defined set of test areas. Many classifications of the test areas were conducted; different numbers and combinations of wavelength bands were used, for both the interim unfiltered and digitally filtered data sets. The best classification results were obtained with six wavelength bands (2, 3, 7, 9, 1O, and 11) of the filtered data set and the weighted a priori probabilities (as opposed to equal a priori probabilities). For the six-wavelength-band combination, the overall classification performance accuracy was 92 percent for Level I and 90 percent for Level II. For the best four wavelength bands (3,7,8, and II), the Level I accuracy was essentially as good, 91 percent, and the Level II accuracy was 89 percent. Table 2-VII shows the results of the classifications for both Level I and Level II land use classes, based on the test-site data.

A study by Sattinger et al. (ref. 2-18) involved processing of the S192 data for application to recreational land analysis. The test site was of primary value for....

 

 

TABLE 2-VII.-Level I and Level Il Land Use Classification Results for an Area in Central Indiana.

 

Land use category

Percent of test area pixels classified correctly

Optimum six bands a

Optimum four bands b

Level I

Level II

Level I

Level II

. .

Urban

77.2

.

78.5

.

Residential

.

80.2

.

79.1

Commercial-industrial

.

71.9

.

75.0

Extractive

.

50.0

.

65.4

Agriculture

88.7

.

88.9

.

Soil

.

93.5

.

85.3

Grass

.

86.0

.

86.2

Forest

93.7

.

92.7

.

Deciduous

.

92.2

.

90.4

Coniferous

.

56.4

.

59.0

Water

98.1

.

96.1

.

River

.

89.2

.

73.0

Lake

.

98.9

.

98.5

Overall

91.6

89.7

91.1

88.8

a Bands 2, 3, 7, 9, 10 and 11.
b Bands 3, 7. 8. and 11.

 

 


[
31]

FIGURE 2-11.

FIGURE 2-11.-Color-coded classification using S192 multispectral scanner data for an area south of Bloomington, Indiana. Monroe Lake is at the top. Nine land use classes are represented. Bands 3, 7, 8, and 11 were used for this classification. (Scale of original, 1:240 000.) [For a larger picture, click here]

 

[32] ....wildlife habitat and involved a relatively small (5300 km2), mostly forested State game area in central Michigan. A maximum-likelihood algorithm was used for the classification. A total of 35 spectral training classes was initially defined; these classes were then combined to form 10 major informational categories. Six wavelength bands were used for the classification, based on a computer evaluation of the optimum combination for this particular analysis. The wavelength bands used, in order of preference, were 0.78 to 0.88, 1.55 to 1.75, 0.98 to 1.08, 0.68 to 0.76, 0.52 to 0.56, and 0.62 to 0.67 µm. The classified data were quantitatively evaluated by a summarization of the area of the various cover types over three 2.6-km2 sections of land, and by a detailed quantitative comparison of the classified data with the Michigan Department of Natural Resources cover-type maps. These evaluations indicated that the 10 informational classes had been mapped with only a 54-percent classification performance. Consolidation of the 10 classes into 5 more generalized categories resulted in an accuracy of 72 percent. This low level of accuracy is attributed to the complexity of the test site, as indicated by the measure of many small stands, a condition that resulted in many "edge" pixels, or pixels containing more than a single cover type. Such factors are important to consider in classification. Sattinger (ref. 2-18) stated that S192 data can be used for regional surveys of existing or potential recreational sites, for delineation of open-space areas, and for preliminary site evaluation of geographically extensive sites.

The Green Swamp in central Florida was investigated by Higer et al. (ref. 2-16); S192 data were used to prepare maps showing environmental categories. The Green Swamp is not a continuous expanse of swamp, but a composite of many swamps with interspersed low ridges, hills, and flatlands. The water, land, and vegetation in the area are undergoing rapid changes caused by logging; reforestation; alteration of natural drainage by channelization and ponding and burning and clearing for such purposes as sod farming, citrus farming, pasture improvement, and urban and industrial development. Improper planning and construction of new industrial and residential areas could clearly have a disastrous effect on such an environmentally sensitive area. For this reason, current cover-type maps are urgently needed for use in environmental appraisals to develop a rational basis for further planning and controlled development. Higer's study was directed at evaluating the usefulness of Skylab S192 data for timely interpretation, assessment, and mapping of environmental categories. These categories included wetlands, water, cypress, pine, pasture, and uplands. A computer-aided analysis technique, incorporating supervised classification, was used. After classification, a qualitative evaluation of the results was conducted that involved comparison of the classification maps and aerial photographs of the study area. These comparisons indicated that the categorized S192 data were "found to be truly representative land-water cover conditions in the Green Swamp area." These results are discussed further in the subsection entitled "Comparison of Skylab S192 and Landsat multispectral scanner analyses."

Klemas et al. (ref. 2-9) used the S192 data and a computer interactive system to perform a land use classification of part of the Delaware Bay area. The system allowed a maximum of only four wavelength bands and eight spectral classes to be used in a single classification. Bands 4, 6, and 8 were selected because they roughly correspond to three of the Landsat bands; band II was included because other investigators have suggested that the use of wavelengths further into the reflective infrared could increase land use classification accuracies. The eight classes identified were water, sand and bare sandy soil, saltmarsh cordgrass, forest land, built-up land, plowed fields, cropland (planted fields), and a class composed of cattails and giant reedgrass. The classification results were displayed one class at a time and evaluated qualitatively. The investigator showed that classification accuracies ranged from 100 percent for water to 44 percent for the built-up-land category. In the latter group, most of the errors in classification were due to confusion with the agricultural category. However, photointerpretation of EREP S190A/S190B data indicated an accuracy for built-up areas of 81 percent. Klemas indicated that the results represented "a conservative indication of S192 capabilities because of the somewhat unsophisticated classification algorithm used and the limited number of categories and band combinations available."

Most of the investigations that included computeraided analysis of S192 data were involved with test areas having little topographic relief in contrast, a study by Hoffer (ref. 2-5) involved a 77 354-km2 test site in the San Juan Mountains of southwestern Colorado-an [33] area of rugged topographic relief and complex patterns of cover types. This study site was in a particularly important region from the standpoint of many conflicting demands for use of the land; e.g., timber production, wildlife habitat, grazing, recreation, and mineral production. Increasing public pressure for summer- and permanent-home developments is causing much concern on the part of the U.S. Forest Service, the National Park Service, the Bureau of Reclamation, and other agencies responsible for management of these lands. As a result of this pressure, personnel from these agencies indicated considerable interest in obtaining reasonably accurate and up-to-date cover-type maps that could be used for inventory purposes and for monitoring environmental alterations.

Computer classification of the S192 data involved several different analysis sequences. First, a detailed study of the data quality indicated considerable variation among the 13 wavelength bands, from both a qualitative and a quantitative standpoint. A newly developed "modified clustering" technique (discussed in sec. 6) was used to obtain training statistics. A divergence algorithm was used to determine the optimum combination of wavelength bands for various numbers of bands, and the best four bands were identified as 0.46 to 0.51, 0.78 to 0.88,1.09 to 1.19, and 1.55 to 1.75 µm. The data were classified by using a maximum-likelihood algorithm. The cover types involved in this phase of the analysis were coniferous forest, deciduous forest, grassland, exposed rock and soil, water, and snow. Figure 2-12(a) illustrates the computer classification map of major land use classes for a sample portion of the entire test area as compared to a covertype map (fig. 2-12(b)) developed by manual interpretation of the aerial photographs that were obtained in support of this Skylab mission. The figure shows that these two maps qualitatively compare quite well.

To obtain a quantitative evaluation of this classification of land use cover types, a test area consisting of a statistically defined grid of four by four resolution elements was used. A summary of classification performance for each land use class is shown in table 2-VIII. In this case, the overall classification accuracy was 85 percent. If the coniferous forest and the deciduous forest were combined, the overall accuracy for forest cover would be 98 percent. Several other investigators also indicated that forest cover, as a major cover type, could be accurately mapped. The grassland category was the most difficult to classify; i.e., a considerable amount of grassland was misclassified as deciduous forest. Hoffer believed that this misclassification occurred primarily because foliation of much of the deciduous forest cover was not complete at the time of the Skylab 2 overpass (June 5,1973). Therefore, the training field statistics for grassland tended to overlap those of deciduous forest. The cover-type category of "exposed rock and soil" is not included in table 2-VIII because none of the statistical-grid test areas fell on a sufficiently large area (approximately 8 km2 minimum size) of exposed rock or soil.

Evaluation of the computer-developed classification maps indicated that the classification was reasonably accurate, and an estimate of the areal extent of each cover type was then tabulated (table 2-lX). This tabulation required 45 seconds of computer time to complete the entire 77 354-km2 area.

Areal estimates of the major cover types obtained by photointerpretation were compared, on a quadrangle-by-quadrangle basis, to the areal summary based on computer-aided analysis of EREP data. This comparison resulted in a correlation coefficient of 0.929. Hoffer (ref. 2-5) stated that this correlation was particularly significant because it indicated (together with similar results obtained from Landsat-l) that reliable areal estimates can be obtained by using computeraided analysis of satellite data, even for areas of rugged mountainous terrain. It was also noted that, for areal estimates obtained from computer classifications, commission and omission classification errors tended to balance, particularly as the geographic areas involved became larger. A similar comment concerning such areal estimates was made by Sattinger et al. (ref. 2-18).

Urban land use mapping with use of S192 thermal-infrared data.-Although relatively few detailed urban mapping studies with S192 data were undertaken, several investigators concentrated their efforts on the value of the thermal-infrared band because it provided a type of data that could not be obtained with the higher resolution EREP photographic systems.

The thermal-infrared band and a density-slicing technique were used by Hannah et al. (ref. 2-2) to map Orlando, Florida. The resultant thermal map indicated that the thermal radiance is highest for commercial-industrial regions, next highest for "modern" residential....

 


[
34]

FlGURE 2-12.

FlGURE 2-12.-Comparison between major cover types obtained by computer classification of S192 data (fig. 2-12(a)) and those obtained by manual interpretation of aerial photographs (fig. 2-12(b)) of the Vallecito Reservoir study area in southwestern Colorado. The reservoir is at the top. (a) Computer classification. (b) Manual interpretation. [For a larger picture, click here]

 

....areas having relatively few trees, and less for more wooded residential areas. The commercial-industrial sectors were identified and enhanced by computeraided analysis of the thermal-infrared data. In this geographical area, even the "modern" residential and wooded residential areas were distinctly defined by this density-slicing technique. Hannah concluded that this type of "a thermal radiance map might be used to classify sectors according to their environmental impacts (relative to amounts of trees and other vegetation and concrete) and certainly can be utilized by planners as a graphic indication of the value of landscaping,"

Alexander and Lins (ref. 2-11) studied the Skylab S192 thermal-infrared data as a source of information on urban climates and surface energy balance in urbanized areas. In their experiment, a combination of....

 

[35] TABLE 2- VIIl.-Classification Performance for Major Cover Types Using Four Wavelength Bandsa of Skylab S192 Data Obtained on June 5,1973.

 

Cover-type category

No of samples

No. of test-area samples classified as-

Percent classified correctly

Water

Snow

Grassland

Deciduous forest

Coniferous forest

Rock/soil

.

Water

96

91

0

0

0

5

0

94.8

Snow

112

0

112

0

0

0

0

100.0

Grassland

128

0

0

67

28

33

0

52.3

Deciduous forest

368

0

0

0

227

132

9

61.7

Coniferous forest

1696

0

2

1

132

1542

19

90.9

Totals

2400

91

114

68

387

1712

28

b85

a Bands 2, 7, 9, and 11 (optimum four).
b Overall Performance (2039/2400) = 85.0 percent

 

.....modeling and observational techniques was used to assess the application of remote sensors to improve understanding of the relationships between surface properties and the mesoclimates of urbanized regions. Because drastic changes in land use, such as land clearing and urbanization, often have a significant climatic impact on urbanized areas, this study was undertaken to derive information useful for land use planners who are required to consider energy use and climatological effects of proposed changes in their planning jurisdictions.

 

TABLE 2-lX.-Areal Estimates for Major Cover Types in the Colorado Granite Peaks Test Site Based on Computer Classifications of Skylab S192 Data.

 

Cover-type category

No. of scanner resolution elements

Area, hm2

.

Water

2 064

959

Snow

16 852

7 828

Grassland

3 397

1 578

Coniferous forest

109 975

51 086

Deciduous forest

31370

14 572

Exposed rock and soil

2 865

1 331

.

Totals

166 523

77 354

 

The climate of an urbanized region has been described simplistically as a group of heat islands set in a matrix of cooler, nonurban areas. The study of such heat islands clearly involves more than the analysis of the surface thermal state. The overall reflectance of an area will control the amount of energy being absorbed by the surface and thus affect the net radiation, which is the balance between energy absorbed and energy emitted. If one is limited to the use of ground-based data collection techniques, effective and accurate net radiation measurements over different locations and conditions are extremely difficult to obtain. The S192 scanner system, however, provided an effective method for obtaining a broad variety of data on cover types and conditions. The 10.2 to 12.5-µm band was calibrated by using a variety of reference data to correct for atmospheric effects. Alexander and Lins (ref. 2-11) then generated a map that showed the distribution of surface-radiation temperatures in the Baltimore, Maryland, area for the August 5,1973, Skylab overpass (fig. 2-13(b)). The corresponding portion of an S19OB photograph (fig. 2-13(a)) is shown for comparison. The radiation-temperature map shows the value of the synoptic view obtained from spacecraft altitudes for this type of study The hypothesized urban heat islands are easily identified on the map, and the relative coolness of nonurban land is effectively documented. Thermal patterns and absolute values of radiation levels were obtained, documented, and later used to test the simulated model of....

 


[
36]

FIGURE 2-13.

FIGURE 2-13.-An S190B photograph and computer-generated temperature maps of the Baltimore, Maryland, area. Chesapeake Bay protrudes from the right. The white patches in the photograph are clouds. (Original scale, 1:250 000.) (a) Small portion of an S190B photograph taken August 5, 1973, showing the surface features (SL3-83-166). (b) Computer-generated thermal-infrared surface-radiation-temperature map obtained from S192 dab band 13. (c) Computer-generated, simulated surface-radiation-temperature map. [For a larger picture, click here]

 

[37] ....the area (fig. 2-13(c)). These results indicate the value of such thermal-infrared data for observing the actual temperature characteristics of an urban area, despite the fact that the quality of these particular thermal-infrared data was less than optimal for the thermal-mapping purposes. Alexander and Lins (ref. 2-11) also noted that the conical scan pattern of the S192 scanner system was very desirable for this type of work because it enables maintenance of an optimal path of constant length for the scanner data.

A simulation experiment using the thermal-map information was conducted on the basis of the concept that each land use type has a particular mix of surface cover and building configuration associated with it. Therefore, if a land use map with an appropriate classification scheme is available, it should be possible to use the distribution of significant surface-cover characteristics in modeling the energy balance and distribution. The surface characteristics used by Alexander and Lins (ref. 2-11) as input for the energy balance model studies were (1) the building configuration, which provides information related to the surface roughness and solar radiation calculations; (2) the surface "wet fraction"; (3) the substrate thermal diffusivity and conductivity; (4) the surface albedo; and (5) the surface emissivity. The values of these factors for each different land use category were calculated and used as input to the model on the basis of the land use map of the test area.

A comparison of the S192-derived and the simulated surface-temperature maps (figs. 2-13(b) and 2-13(c), respectively) shows that the general heat distribution within the city of Baltimore is similar to that of outlying residential and commercial areas. Although both maps show the same generalized patterns of temperature distribution, differences are apparent in the shape of individual features. The map obtained from the S192 data has much more complex and intricate patterns than those on the simulated map. Also, the simulated map temperatures are as much as 6 to 8 K lower than the temperatures on the S192 map for the area in the center of the city heat island. Alexander and Lins stated that there is a considerable potential for further improvements in simulation-map studies through the use of such thermal-infrared scanner data. The results also indicated that the input from the land use map used for the simulation model did not have sufficient detail concerning phenomena that affect surface temperature.

The land use classes mapped in the simulation study tend to generalize many features such as houses, streets, lawns, forests, and fields.

Although the results of the S192 data analysis showed a potential urban application of a simulated surface-radiation-temperature map, the investigation indicated a need for further definition of those phenomena that affect surface temperature. In essence, the land use categories mapped may not represent the distribution of the features that actually affect the surface-temperature regime or the surface-energy-exchange phenomenon with sufficient realism. Consequently, the investigators recommend further work to refine and develop a more detailed classification for land cover that can be effectively used for climatological purposes.

An integral part of the land use investigation by Silva (ref. 2-14) over an area in northeastern Indiana was the analysis of S192 data obtained on January 25, 1974, for the city of Fort Wayne, Indiana. The land use classes involved were residential, commercial-industrial, hard surfaces (parking lots and runways), grass-covered areas (pastures, winter wheat, and golf courses), bare land, forest/snow, water, and snow. The analysis procedures Silva used were basically the same as those discussed in the preceding subsection. The classification performance of the scanner resolution elements, or pixels, used to train the computer is shown in table 2-X. The optimum combination consists of four wavelength bands-one band in the visible wavelength region (band 4, 0.56 to 0.61 µm), one in the near infrared (band 8, 0.98 to 1.08 µm), one in the middle infrared (band 11, 1.55 to 1.75 µm), and one in the thermal infrared (band 13, 10.2 to 12.5 µm). This combination resulted in an overall training-pixel classification performance of 90 percent. When the thermal-infrared band was excluded, the accuracy of the same training pixels decreased to 80 percent. When the middle-infrared band was excluded, a visible band (the only one available in this data set, two near-infrared bands, and the thermal-infrared band were selected for use; and the resultant classification performance was 79 percent. This comparison indicates the value of both the thermal-infrared and the middle-infrared wavelength bands in this land use classification sequence. One of the most significant results is seen in the extremely poor classification performance (even for the training data) of the residential and commercial-industrial land use classes when the thermal-infrared wavelength band is

 

 

[38] Table 2-X.- Training-Pixel Classification Performance.

 

Land use category

No. of training pixels

Percent classified correctly

Optimum four bands overall (4, 3, 8 13)

Optimum four bands exclusive of middle Infrared (4, 7, 8, 11)

Optimum four bands exclusive of middle Infrared (4, 3, 9, 13)

.

Residential

175

92

65

82

Commercial-industrial

61

95

18

90

Hard surface

52

73

71

52

Grass

141

90

89

81

Bare land

459

95

93

89

Forest/snow

81

79

86

54

Water

152

80

77

58

Snow

25

96

92

92

.

Overall performance

a1146

90

80

79

a Total number of training pixels

 

 

....not used. These two urban classes alone would account for most of the decrease in overall performance from 90 to 80 percent.

The importance of the thermal-infrared wavelength band in discriminating the urban land use classes from the other cover types is shown in figure 2-14. In these wintertime S192 data, the commercial and industrial classes and, to a somewhat lesser extent, the residential class have equivalent black-body temperatures significantly higher than those of the other land use classes. The effect of these parameters on determining area percentage of land use classes with S192 bands is shown in table 2-XI. Comparing the area estimates of the various land use classes obtained by the computer classification over the entire test site to the area estimates developed by conventional methods used by county and city planning agencies offers a procedure to establish the quality of the results. Such a method was pursued in this study and was found to be especially useful for showing the value of the thermal-infrared band in this analysis involving an urban area. Comparison of the computer-derived areal estimates with those provided by the county and city agencies for the five major categories in the data shows that the estimates obtained by using the thermal-infrared data for band 13 approximate the estimates provided by the city and the county. The differences in the urban class can be explained, for the most part, by the differences in the cover types that were included in the urban area. For example, cover...

 


FIGURE 2-14.-Equivalent black-body temperatures computed from the S192 thermal-infrared band for training classes.

FIGURE 2-14.-Equivalent black-body temperatures computed from the S192 thermal-infrared band for training classes. Values include the mean plus or minus one standard deviation. Multiple classes are shown for bare land and water.

 

[39] ...types such as golf courses and parks were classified, according to the data provided by the county and city agencies, as urban area; but these cover types were classified as grassland in the computer classification and therefore grouped with the winter wheat as agricultural land. The area classified as snow was largely agricultural land. The underestimate of commercial and industrial area was largely due to classification errors such as the railroad right-of-way through Fort Wayne; the right-of-way was considered as commercial in the estimates provided by the city but was classified by the computer as bare soil. Also, the Allen County Planning Commission estimates for the industrial and commercial category included the areas owned by businesses or industries (i.e., buildings, parking lots, and landscaped property). In the computer classification, the buildings themselves were largely delineated as commercial or industrial and the crushed stone lots as hard surface, or the asphalt parking lots were often identified as being residential. These results provide some insight into the comments by several of the investigators that the land use classes designated by user groups or USGS Circular 671 (ref. 2-1) often may not coincide with the cover types that can be spectrally discriminated by using remotely sensed data.

As in table 2-X, the value of the thermal- and middle-infrared wavelengths for accurate identification of land use classes, particularly the urban classes, is shown in table 2-XI. Absence of the thermal-infrared band causes a very significant overestimation of the area in the urban category. With the thermal- and middle-infrared bands present, the areal estimates obtained by computer classification compare favorably with those obtained by conventional techniques.

Comparison of Skylab S192 and Landsat multispectral scanner analyses.-Several investigators compared classification results obtained by using Skylab S192 and Landsat data. In essentially every case, both Level I and Level II land use cover types could be mapped with approximately the same overall degree of accuracy by using either the Skylab or the Landsat data. In general, the investigators found that the wavelength bands above 1.1 µm, which were available on the Skylab data but not on the Landsat data, were frequently selected by the various computer analysis routines as being very important and valuable bands to use in the classification. The data quality generally was much better for the Landsat data being used than for the Skylab data. The results did indicate that the improved spectral resolution available with the Skylab S192 data enabled improvement in classification accuracy.

One of the more precise comparisons between classification results obtained by using the Skylab S192 data and the Landsat data was accomplished by Hoffer...

 

 

Table 2-XI.- Land Use Area Percentage Estimates for Major Portion of Allen County, Indiana.

 

Land use category

Area estimate, percent (b)

.

Local Allen County reference data a

Skylab 4 S192 data

.

Optimum four bands overall (4, 8, 11, 13)

Optimum four bands exclusive of thermal infrared (4,7,8,11)

Optimum four bands exclusive of middle infrared (4,8,9,13)

.

Urban

12.6

9.9

27.7

7.3

Residential

(10.1)

(9.1)

(26.7)

(6.8)

Commercial-industrial-hard surface

(2.5)

(.8)

(1.0)

(.5)

Agricultural/forest

85.6

85.6

68.1

87.3

Water

-

1.4

1.4

1.6

Other land

1.8

-

-

-

Snow

-

3.1

2.8

3.8

a Data from the Allen County Planning Commission and the Fort Wayne Department of Community Development and Planning.
b Values in paratheses are subtotals.

 

 

[40] ....(ref. 2-5). For this effort, a frame of Landsat data was used that had been digitally registered to both the S192 data and to a USGS 7.5' quadrangle map base with a reasonably high degree of precision (±1 pixel). The Skylab S192 and Landsat data were obtained on the same day (within 2.5 hours) and under completely cloud-free conditions. Statistically defined grids of the test area (each, four by four pixels in size) were used to evaluate the classification results so that they could be quantitatively compared for exactly the same resolution elements on the ground with use of the different data sets.

Three classifications were conducted. First. the major cover types present in the area were classified by using the four optimum wavelength bands of Skylab S192 data (bands 2, 7, 9, and 11, as determined by the divergence algorithm). Second, the four S192 wavelength bands that most closely corresponded to the four Landsat bands (bands 3, 5, 6, and 8) were used to classify the data. Third, the Landsat data were classified.

Silva (ref. 2-14) followed a very similar procedure for comparing Skylab S192 and Landsat data obtained over a test site in central Indiana. In his study, the Landsat data were obtained a day before the Skylab data were obtained, and the two data sets were not digitally registered. The test blocks for Silva's study involved hand-selected test areas rather than a statistically defined grid. The basic approach was the same, although the optimum four Skylab bands used by the authors were different.

The overall results of these two land use studies are shown in table 2-XII. The four optimum combinations of wavelength bands from Skylab data produced classification results that were almost identical in both investigations. In both studies, the four wavelength bands of Skylab data that most closely corresponded to the Landsat bands produced less accurate results than those obtained using the actual Landsat data. Considering that the cover types were rather different in the two test locations, it is significant that the results of both comparisons are approximately the same. The Colorado....

 

 

TABLE 2-XII.-Comparison of Classification Performances Using Skylab and Landsat Multispectral Scanner Data.

 

Data

Wavelength bands used

Overall classification performance, percent

.

Hoffer's results a (ref. 2-5)

Silva's results b (ref. 2-14)

.

Optimum four wavelength bands of Skylab data

0.46 to 0.51, 0.78 to 0.88, 1.09 to 1.19, and 1.55 to 1.75 µmc (bands 2, 7, 9, and 11)

85.0

87

Skylab data using wavelength bands that correspond to Landsat

0.52 to 0.56, 0.62 to 0.67, 0.68 to 0.76, and 0.98 to 1.08 µm (bands 3, 5, 6, and 8)

82.5

80

Landsat data

0.5 to O.6, 0.6 to 0.7, 0.7 to O.8, and 0.8 to 1.1µm (bands 4, 5, 6, and 7)

85.7

88

a Cover types were coniferous forest, deciduous forest, grassland, water, and snow.
b Cover types were residential, commercial-industrial, extractive, soil, grass, deciduous forest, coniferous forest, river, and lake.
c For Silva's data, the optimum combination of four-wavelength bands included the 0.52 µm- to 0.56 µm band (3) rather than the 0.46- to 0.51 µm band (2), and the 0.98- to 1.08 µm band (8) rather than the 1.09- to 1.19-µm band (9). In both cases, these bands are adjacent to those used by hoffer. The other two bands were identical in both studies

 

 

[41] ....site (Hoffer, ref. 2-5) involved complex mountainous terrain consisting primarily of forest and rangeland cover types, whereas the Indiana test site (Silva, ref. 2-14) involved urban and agricultural cover types.

Both investigators concluded that the better spectral resolution and the extended spectral range of the Skylab S192 scanner contributed to an improvement in classification performance. This improvement was indicated by the difference in the classification results obtained by using the best four wavelengths that corresponded to the Landsat scanner systems. The fact that the Skylab data did not yield better classification results than the Landsat data was attributed to the noisy characteristics of the Skylab data. The investigators believed that the results obtained indicated the importance of good-quality data, the value of using the optimum combination of fairly narrow, properly located spectral bands for cover-type mapping, and the value of using computer-aided analysis techniques.

Polcyn et al. (ref. 2-17) also conducted a comparison of Skylab and Landsat classification results, for a test site in Ontario, Canada. The percentage of the total area classified into the various cover types by using four S192 bands (bands 4, 5, 7, and 13; 0.56 to 0.61, 0.62 to 0.67,0.78 to 0.88, and 10.2 to 12.5 µm, respectively) was compared with the Landsat data classification . The results indicated that a reasonably similar classification was obtained with both data sets. A detailed comparison with aerial photographs indicated that both Skylab and Landsat data had enabled achievement of a reasonable classification , with some variation among categories within both data sets. The most severe misclassification occurred with the Landsat data in the manmade category of cover types. Overrecognition occurred, and, in general, the bare soil category caused the most confusion. It was concluded that the Skylab S192 and Landsat data appeared to be reasonably equivalent in terms of information content and distinction of the various cover types in the area involved. Most of the difference in the percentage of the area recognized as particular classes can be accounted for by differences in the training-set signatures used rather than by any fundamental difference in the information content in the spectral data from the scanner system.

Cover-type maps of the Green Swamp area in Florida were generated with the use of both Landsat and Skylab S192 data (Higer et al., ref.2-16). The resulting classifications were representative of the cover types present, and the vegetation maps produced from S192 and Landsat categorized data were in accord with county land use maps by 82.8 and 87.2 percent, respectively. Hannah et al. (ref. 2-2) stated that computer classification of S192 data over an urban area (Orlando, Florida) resulted in cover-type maps that were of generally comparable quality to those previously obtained from Landsat.

To summarize, the comparison of Skylab and Landsat classification results showed that the increased number of spectral bands of the Skylab scanner system enabled definition of a better combination of four wavelength bands for computer analysis.

Wavelength band evaluation.-Detailed investigations were conducted to determine the combinations of wavelength bands that are optimum for land use mapping. This subsection is a brief summary of some of these results. The reports by Hoffer, Silva, and Simonett (refs. 2-5, 2-14, and 2-10, respectively) are particularly detailed on this subject.

In the initial phases of Hoffer's investigation of the use of S192 data in mountainous terrain, different approaches and techniques were applied to define the data quality of the different bands. These analyses resulted in a numerical data-quality index for each wavelength band (table 2-XIII). To interpret the numerical values, a quantitative evaluation designation was also defined. Comparison of the data-quality indexes with the imagery of the individual wavelength bands showed that, in several cases, the visual appearance of the imagery or the qualitative evaluation (table 2-XIII) was not a reliable indication of the spectral information content of the data. The investigators stated that the quality of multispectral scanner data can be effectively evaluated only by quantitative evaluation techniques (rather than qualitative techniques) if the data are to be analyzed by computer.

Another phase of Hoffer's investigation (ref. 2-5) was directed to determining the number of wavelength bands required for effective classification with use of the S192 data. Previous work has indicated that, as the number of wavelength bands increases, the classification performance initially increases rapidly when four to six wavelength bands are used but increases at a slower rate above this number. The amount of computer time required to classify the data increases significantly for more than four to six bands, as shown in figure 2-15. The effect of increasing the number of....

 

 

[42] TABLE 2-XIII.-Data-Quality Evaluation Results.

 

Spectral region

Band no.

Wavelength band, µm

Qualitative evaluation designation

Quantitative index

Quantitative evaluation designation

.

Visible

1

0.41 to 0.46

Very poor

7.1

Fair

2

0.46 to 0.51

Poor

2.0

Good

3

0.52 to 0.56

Very good

1. 8

Very good

4

0.56 to 0.61

Poor

14 8

Poor

5

0.62 to 0.67

Fair

12.1

Poor

Near infrared

6

0.68 to 0.76

Fair

4.1

Fair

7

0.78 to 0.88

Very good

2.2

Good

8

0.98 to 1.08

Very good

5.6

Fair

9

1.09 to 1.19

Very good

1.8

Very good

10

1.20 to 1.30

Good

6.2

Fair

Middle infrared

11

1.55 to 1.75

Very good

1.6

Very good

12

2.10 to 2.35

Good

2.9

Good

Thermal infrared

13

10.2 to 12.5

Poor

11.8

Poor

 

.....wavelength bands on performance results was tested on the S192 data. The divergence algorithm was used to determine the optimum combination of I to 13 wavelength bands. The data were then classified by using the maximum-likelihood algorithm. The results of this analysis indicated that classification performance was not significantly improved when more than four wavelength bands were used (fig. 2-16). These results were based on test-area classification performance for both major and forest cover types. The overall classification accuracy for the major cover types as a function of the number of wavelength bands is shown in table 2-XIV. For this data set, the 1.09- to 1.19-µm wavelength band in the near infrared was the single most valuable wavelength band. The best combination of four wavelength bands consists of one in the visible region (the 0.46- to 0.51-µm band, which, in table 2-XIII, was indicated to be visually of qualitatively poor data quality), two in the near-infrared region (the 0.78 to 0.88-µm and 1.09- to 1.19-µm bands), and one in the middle-infrared region (1.55 to 1.75 µm). The best combination of six wavelength bands consists of two in the visible, two in the near infrared, one in the middle infrared, and one in the thermal infrared. Furthermore, detailed studies indicated that various combinations of four wavelength bands were required to achieve optimal classification performance for different individual cover types. The near-infrared portion of the spectrum (especially the 1.09- to 1.19-µm wavelength band) was shown to be of particular value for effective vegetation mapping. The relative importance of the different spectral regions and the individual wavelength bands varied significantly as a function of the cover types to be mapped.

Simonett (ref. 2-10), using a series of statistical procedures, also found that the 1.09- to 1.19-µm wavelength band was the most valuable single band for discriminating among land use categories. For overall land use mapping, in order of ranking, the most useful six spectral bands were band 9 (1.09 to 1.19 µm), band 3 (0.52 to 0.56 µm), band 6 (0.68 to 0.76 µm), band 1 (0.41 to 0.46 µm), band 11 (1.55 to 1.75 µm), and band....

 


[
43]

FIGURE 2-15. Overall cover-type-classification accuracy and computer-processing time required compared to number of S192 bands used.

FIGURE 2-15. Overall cover-type-classification accuracy and computer-processing time required compared to number of S192 bands used.

 

....13 (10.2 to 12.5 µm). The results indicated that the optimal spectral bands identified for discriminating among general land use categories were significantly different from the spectral bands identified for discriminating within specific land use categories.

Different sets of spectral bands were selected when analyzing different groups of Level II land use categories (all belonging to the same Level I land use catego-.....

 


FIGURE 2-16. Overall classification accuracy as a function or the number of S192 wavelength bands used for classifying two levels of detail of land use information.

FIGURE 2-16. Overall classification accuracy as a function or the number of S192 wavelength bands used for classifying two levels of detail of land use information.

 

-....ry). The following sets of spectral bands were selected as providing the best discrimination.

 

Category

Bands

.

.

Urban

9,3,11,6,13

Agricultural

9,7,1,11,6

Forest

9,3,5,4,11

Water

6,1,3,8,9

Wetlands

10,3,1,8,6

 

Silva's wavelength-band study (ref. 2-14) also included an evaluation of the various combinations of wavelength bands and a comparison of interim and filtered S192 data sets. These results are summarized in table 2-XV. Silva's results emphasize the value of the middle-infrared wavelength band (1.55 to 1.75 µm), the near-infrared bands (0.78 to 0.88 µm and 0.98 to 1.08 µm), and the visible band (0.46 to 0.51 µm) for identifying land use classes on the Skylab 2 data obtained on June 10, 1973. The combination of the thermal-infrared band (10.2 to 12.5 µm), the middle-infrared band (1.55 to 1.75 µm), and the visible band (0.56 to 0.61 µm) was particularly useful in analysis of the Skylab 4 data obtained on January 25, 1974. As these results indicate, there is no single set of bands that is best under all circumstances.

 

[44] TABLE 2-XIV.-Optimal Wavelength Bands for Major-Cover- Type Classification Using Skylab S192 Data Obtained on June 5, 1973.

 

Spectral region

Band no.

Wavelength band, µm

Data-quality index

Optimum wavelength-band combinations a

1

2

3

4

5

6

7

8

10

13

. .

Visible

1

0.41 to 0.46

7.1

.

.

.

.

.

.

.

x

x

x

2

0.46 to 0.51

2.0

.

x

x

x

x

x

x

x

x

x

3

0.52 to 0.56

1.8

.

.

.

.

.

.

.

.

x

x

4

0.56 to 0.61

14.8

.

.

.

.

x

x

x

x

x

x

5

0.62 to O.67

12.1

.

.

.

.

.

.

.

.

.

x

Near-infrared

6

0.68 to 0.76

4.1

.

.

.

.

.

.

.

.

.

x

7

0.78 to 0.88

2.2

.

.

x

x

x

x

x

x

x

x

8

0.98 to 1.08

5.6

.

.

.

.

.

.

.

.

.

x

9

1.09 to 1.19

1.8

x

x

x

x

x

x

x

x

x

x

10

1.20 to 1.30

6.2

.

.

.

.

.

.

.

.

x

x

Middle infrared

11

1.55 to 1.75

1.6

.

.

.

x

x

x

x

x

x

x

12

2.10 to 2.35

2.9

.

.

.

.

.

.

x

x

x

x

Thermal infrared

13

10.20 to 12.50

11.8

.

.

.

.

.

x

x

x

x

x

Overall classification performance, percent

75.7

76.8

81.9

85.0

84.1

83.7

85.3

84.1

85.2

86.0

a Band 9 was selected as the single best wavelength band, bands 9 and 2 were selected as the two best wavelength bands, etc.

 

 

TABLE 2-XV.-Optimum Combinations of Wavelength Bands for Mapping Land Use in Indiana [Skylab 2 data].

 

No. of bands

Bands of optimum value a

Interim data set

Filtered data set

.

1

11

111

3

1,2,11

1,2,11

4

3,7,9,11

2,7,8,11

5

2,3,7,9,11

2,7,8,9,11

6

2,3,7,9,10,11

2,7,8,9,10,11

a Band 13 was not available in the interim data set, and bands 4, 5, 6, and 12 were not available in the filtered data set.

 

 

The results reported by Silva (ref. 2-14) (and partly summarized in table 2-XI) demonstrate the value of the thermal- and middle-infrared wavelengths for obtaining accurate areal estimates of the urban land use class (particularly the residential). The S192 data had been obtained during the winter when some snow was on the ground. In this area, the optimum combination of four wavelength bands (4, 8,11, and 13) resulted in an areal estimate for the urban residential area of 15 000 hm2, compared to 16 900 hm2 estimated by the Fort Wayne Department of Development and Planning. When the thermal-infrared wavelength band was not used, the classification resulted in a residential areal estimate of 43 850 hm2. Without the middle-infrared data (when using one visible, two middle-infrared, and the thermal-infrared bands), the areal estimate based on the computer classification was 11 200 hm2, compared to 15 000 hm2 when all four major spectral regions were represented.

[45] In the Higer et al. study (ref. 2-16) of the Green Swamp (which involved mostly vegetation, wetlands, and water categories), the five wavelength bands that provided the largest contribution to the categorization of the cover types present were, in order of preference, bands 11, 8, 2, 10, and 6. Band 11 was particularly important for identifying vegetation; specifically, Higer indicated that the thermal-infrared band would also be very useful but that, in this case, it was excessively noisy. This investigator noted that bands 8, 9, 10, and 11, which, except band 8, are beyond the range of the Landsat scanner system, were "useful in detection and categorization of cover types in the Green Swamp."

Individual wavelength bands of S192 imagery obtained over several test sites in North Carolina were examined by Welby and Lammi (ref. 2-13), using a density-slicing technique. Their results showed that various vegetative cover types could be separated best by using the near-infrared wavelength bands, which were also fairly effective for separating the cropland areas from forest cover and in defining the boundaries between vegetation and water features.

A color-additive viewer was used by Welby and Lammi to combine selected wavelength bands of the S192 imagery. Their work showed that different combinations of wavelength bands produced variable results in terms of spectral discrimination of cover types or land use categories. They concluded that working with combinations of wavelength bands through use of the color-additive viewer was more effective than analysis of individual wavelength bands and that many of the cover types present could be effectively separated by using this technique. A particularly important conclusion was that "the breaking of the near-infrared portion of the spectrum into a series of relatively narrow bands appears to be a very useful approach to acquisition of earth resource information."

Welby and Lammi identified some of the complexities encountered in attempting to manually interpret and analyze many individual wavelength bands of multispectral scanner imagery. There are many wavelength bands to be considered, and distinct differences in reflectance levels are often found in different wavelength bands in the same spectral region (near infrared, in this case) for the various cover types of interest. These results also tended to emphasize the value of several discrete wavelength bands in the near-infrared portion of the spectrum. If differences in infrared reflectance among the cover types of interest were not present in one wavelength band, another band would enable effective discrimination.

To summarize, the wavelength-band evaluations showed that the optimum wavelength bands for effective classification of various land use cover types are the 1.09- to 1.19-µm band in the near-infrared region, the 1.55- to 1.75-µm band in the middle-infrared region, and the 10.2- to 12.5-µm thermal-infrared wavelength band. Also, each of the four major spectral regions (visible, near infrared, middle infrared, and thermal infrared) is significant with respect to accurate classification, and the importance of each region varies as a function of the cover type and scene characteristics.

 

Environmental Studies

 

The discussion on selected environmental studies includes strip mining, wetland mapping and ecology, and migratory waterfowl habitat evaluation.

Strip mining.-increased energy demands have resulted in accelerated strip-mining activities, the environmental effects of which are of increasing concern. Estimates in 1970 of accumulated coal production (3 992 Tg) from strip mining were slightly more than 3 percent of the total estimate of strippable coal reserves (116 119 Tg) in the United States. The potentially disturbed land areas (those to be strip mined) are estimated to exceed an area larger than the combined size of Pennsylvania and West Virginia. The photograph in figure 2-17 is an example of strip-mining endeavors covering an extensive area in Alabama. On the basis of the growing pressure to increase coal production, it is apparent that a more effective and efficient methodology will be required to map disturbed mining areas and to monitor mining and reclamation activities.

Skylab EREP investigations have provided substantial and positive evidence that remote-sensing data, specifically spacecraft-acquired photographs, can be of value in the overall planning, detection, and monitoring of surface mining activities. The Skylab strip-mining studies are of two major types: (1) the detection and mapping of disturbed areas and (2) the identification....

 


[
46]

FIGURE 2-17.

FIGURE 2-17.-A portion of an S190B color-infrared photograph illustrating the extent of strip-mining activities in an area west of Birmingham, Alabama. The well-defined boundaries formed by the strip-mining operations and the natural vegetation should be noted. Two distinctly different stripping patterns are easily discerned (SL4-93-152). [For a larger picture, click here]

 

[47] ....and interpretation of key physical features that are necessary to monitor the components of ongoing strip-mining operations and reclamation activities.

The simplified photointerpretation technique used by Brooks and Parra (ref. 2-19) requires the use of S19OB 2x color positive transparencies, an overhead projector, and a white poster board as the projection screen. It enables mining officials who do not have extensive photointerpretation backgrounds or expensive scanning equipment to outline, on the basis of color tones and density patterns, salient strip-mining features (fig. 2-18).

Weir et al. (ref. 2-20) have shown that a 1:100 000 scale black-and-white S19OB enlargement (made from color film) is satisfactory for accurately mapping past and current surface mines and for detecting several classes of reclamation assessment. Cultural details at this scale were adequate for preparation of new base maps or updating existing topographic maps that were enlarged from a 1:24 000 scale.

Disturbed strip-mining areas could be detected and discriminated on both S19OA and S19OB infrared photographs. For monitoring surface mining activities, S190B photographs, having high spatial resolution and spectral discrimination, provided the more detailed interpretation results for useful monitoring practices.

The Coshocton County, Ohio, S19OB color-infrared photographs were used by Baldridge et al. (ref. 2-12) to define four major strip-mining-land categories: (1) active areas, (2) orphaned or abandoned land, (3) areas undergoing reclamation or restoration, and (4) natural or planned reclaimed lands. The regrading of stripping operations was the most apparent feature observed on the color-infrared images. Areas covered by varying degrees of vegetation, high walls, and water impoundments were also identified. Baldridge et al. stated that land that had been thickly revegetated had the appearance of being completely reclaimed and was difficult to identify as having been stripped. In some areas, high walls and pond water remained to indicate past stripping operations.

Table 2-XVI is a summary of key features identified by Baldridge, Brooks, and Weir (refs. 2-12, 2-19, and 2-20, respectively). Their results illustrate the positive attributes of using S190B-type photographs. The three investigators used S190B enlargements to discriminate various categories of mining activity and reclamation (vegetation stages, refuse areas and slurry ponds, water bodies, high walls, haulage roads, unmined areas, orphaned areas, and possible acid drainage effects).

The most effective imagery scales for strip-mining and reclamation activities would encompass three ranges. The first range, 1:125 000 to 1:250 000, provides a synoptic overview of the general terrain and topography being mined. Over a period of time, this range would present a vivid pictorial history of extensive and evolving patterns. The second range extends from 1:62 500 to 1:100 000. Two of the Skylab experimenters did most of their investigative work within this range; specifically, at the 1:80 000 scale. For more subtle details, scales ranging from 1:24 000 to 1:50 000 are required. The S190B 2x positive transparencies were optically enlarged to achieve a usable 1:24 000-scale image.

Skylab photographs were also used by Baldridge et al. (ref. 2-12) in conjunction with frames of high-altitude aircraft photographs to identify topographic features that are not measurable in the Skylab photographs alone because of inadequate stereographic parallax. This procedure involved simultaneous viewing of the aircraft photographs (acquired at an earlier date) and Skylab photographs of the same area and at the same scale. The resultant stereoscopic effect provided the investigator with a means to define the slope of land features and to evaluate the temporal impact on the landscape.

The potential of using S190B photographs for the detection of nonfuel surface mining (e.g., clay, sand, gravel, and phosphate mining) was also assessed by Hannah, Welby, and Weir (refs. 2-2, 2-13, and 2-20, respectively). Because clay mines range in surface area from 0.4 to 2 hm2, they are difficult to detect in space acquired imagery. However, the larger sand and gravel mines, with their lobate geometry and associated water bodies, provided a distinctive pattern and were easily detected and identified. The very distinct signature created by phosphate mining in central Florida (highly reflective surface produced by bare sandpiles) indicates that this type of excavating and associated reclamation activities can easily be identified and mapped by using space photographs (ref. 2-2).

 


[
48]

FIGURE 2-18.

FIGURE 2-18.-An S190B photograph showing strip mines in the Madisonville, Kentucky, area (SL4-90-032). (a) This synoptic view allows identification of surface coal strip-mining operations in an agricultural region. (b) An enlargement of a small segment of figure 2-18(a). The features associated with an active strip-mining operation are identified. This scene shows the amount of detail that can be seen in an enlarged portion of an S190B photograph. [For a larger picture, click here]


[
49]

FIGURE 2-18. Concluded (b).

FIGURE 2-18. Concluded (b). [For a larger picture, click here]

 

[50] TABLE 2-XVI.-Summary of Qualitative Feasibility of Using Skylab S190B Data to Identify Surface Coal-Mining/Reclamation Features

 

Feature

Skylab Principal Investigator

Baldridge (ref. 2-12)

Brooks (ref. 2-19) results

Weir (ref.2-20) results

Results

Minimum area discernible using color-infrared film, hm2

.

Active strip mine land

.

Lack of vegetation

(a)

0.5

(a)

(a)

High walls

(b)

(c)

(a)

(a)

Slope (recognized but not measurable)

(b)

(d)

(e)

(e)

Coal seam

(b)

(c)

(c)

(e)

Spoil banks

(b)

(c)

(b)

(b)

Access roads

(a)

(c)

(a)

(a)

Equipment

(e)

(d)

(g)

(g)

.

Orphaned strip mine land

.

No vegetation or sparse vegetation

(a)

0.5 to 1

(a)

(b)

High walls

(b)

(c)

(b)

(b)

Spoil banks

(b)

(c)

(b)

(b)

Impoundments

(a)

0.5 to 1

(a)

(a)

Impoundment quality

(e)

(d)

(b)

(g)

Access roads

(b)

(c)

(b)

(b)

.

Ongoing reclamation and reclaimed strip mine areas

.

Equipment

(e)

(d)

(g)

(g)

Smooth slopes

(b)

(d)

(f)

(f)

Vegetation. 0- to 40-percent cover

(b)

1 to 5

(b)

(b)

Vegetation: 40- to 80-percent cover

(b)

1 to 5

(b)

(b)

Vegetation: 80- to 100-percent cover

(b)

1 to 5

(b)

(b)

Impoundments

(a)

0.5 to 1

(a)

(a)

Impoundment quality

(e)

(d)

(b)

(g)

Access roads

(b)

(c)

(b)

(b)

a Usually determined with ease from S190B.
b Determined often enough to make data useful.
c No Information presented.
d Not applicable.
e Information desired but unobtainable from S190B
f Rarely obtained from S190B
g Information desired but not subjected to investigation with S190B.

 

 

[51] The overall conclusion is that S19OB-type photographs can be used for detection and mapping of the small surface mines and that this knowledge can be used by Federal and State agencies and local groups that are concerned with resources, reclamation, and land use management. Although none of the Skylab investigators specifically evaluated the S190B photographs for other types of surface mining activities (copper, uranium, and limestone), it is obvious that Skylab-type photographs could be useful for mapping and monitoring disturbed areas associated with such mining operations (fig. 2-19).

Welby and Lammi (ref. 2-13) indicated that S190A color-infrared film can be used effectively as an aid in detecting sediment discharge from active or abandoned quarry operations, providing the receiving-stream waters are approximately 60 m wide. Normally, water with high sediment content will have a more highly reflective surface than will other stream water.

Wetland mapping and ecology.-The fragile ecological zone that forms the boundary between land and water mass, termed the "coastal wetlands," has become an increasingly critical area requiring the establishment of effective management practices. Laws regulating the types of activity in wetlands have been enacted by most of the affected States. Anderson et al. and Klemas et al. (refs.2-21 and 2-9, respectively) examined the potential of using spacecraft-acquired data to monitor and map these areas in a practical and inexpensive manner.

Anderson's findings (ref. 2-21), in particular, indicate that orbital photographs (primarily the S190A color-infrared and S190B color films) are the best data base being used for rapid, relatively low cost wetland mapping and monitoring on a regional basis. Figures 2-20(a) and 2-20(b) are two S190A color-infrared prints acquired over Anderson's test site at the mouth of the Nanticoke River in Dorchester County, Maryland; figure 2-20(c) is the compiled wetlands map. In mapping the marsh categories, the tonal contrast of the color-infrared film and the texture patterns were found to be the most important recognition elements in the photointerpretation analysis. Attempts with the use of....

 


FIGURE 2-19.

FIGURE 2-19.-An S190B color photograph showing the large openpit copper mine in Bingham Canyon, Utah. The actively mined area and the mine tailings should be noted (SL3-83-300). [For a larger picture, click here]

 

....other enhancement techniques (color-additive viewing and density slicing) were unsuccessful in producing distinctive signatures in the areas of interest and effectively separating the various categories and individual species. The S190A color film provided slightly better resolution, but tonal contrast (greens and browns) was not as distinguishable as the shades of red and blue in the color-infrared film. The S190A black-and-white infrared film also provided some information in the....

 


[
52]

FIGURE 2-20.- (a), (b).

FIGURE 2-20.- (a), (b). [For a larger picture, click here]



[
53]

FIGURE 2-20. (c)

FIGURE 2-20.-S19OA color-infrared photographs and a compiled wetlands map for the marshes of Dorchester county, Maryland. This type of temporal photographic coverage was useful for the delineation of wetland boundaries. (a) S19OA color-infrared photograph obtained in June 1973 (SL2-15-174). (b) S19OA color-infrared photograph obtained in September 1973 (SL3-39-123). (c) Compiled wetlands map. [For a larger picture, click here]

 

....marsh areas. The final product prepared from both scenes (June and September) indicated the value of temporal data in accurately determining boundary placement during the mapping of marsh categories and in identifying individual species. The June data were used to distinguish the marsh areas from the uplands and the brackish river marsh from the fresher water; they also enabled the delineation of certain individual species by their characteristic color. The September photographs were superior for delineating the upper marsh boundary (i.e., wooded swamps are drier in the fall and provide better contrast) and marsh/water interfaces because of reduced vegetation cover.

Anderson points out that, from analysis of the S19OA data, it was possible to develop a wetland classification system that included freshwater tidal wetlands and also to detect individual species when they occurred in relatively large stands. Several different species (not shown in fig. 2-20) were identified and mapped, and several subcategories of marshes were also identified.

The S190B color photographs of Great Egg Harbor, New Jersey, were examined by Anderson, who indicated that these photographs were similar to high-altitude aerial data with respect to the amount and sharpness of detail. Within this area, the upper marsh boundary, the marsh/water interface, and the perimeters of marsh areas were easy to delineate with the S19OB photographs, and less subjective judgment was involved than with the S19OA photographs. Figure 2-21(b) is a map of the same area compiled from analysis of figure 2-21(a), an S19OB photograph of the....

 


[
54]

FIGURE 2-21.

FIGURE 2-21.-S19OB color photograph (fig. 2-21(a)) of Great Egg Harbor, New Jersey, and a coastal wetland map (fig. 2-21(b)) compiled from an analysis of the Skylab photograph. The increased spatial resolution of the S19OB color film enhanced boundary delineation and placement and also enabled identification of more botanical categories than was possible with the S19OA photograph. (a) S190B color photograph (SL3-86-303). (b) Coastal wetland map. [For a larger picture, click here]

 

....New Jersey site with an approximate scale of 1:110 000. The land/water interface, drainage patterns, ditching activity, and vegetational distribution are well displayed on the photograph. As a result of the excellent tonal contrast on the S19OB color photographs, it was possible to distinguish the boundary between saline and brackish wetland and to delineate the transition zone (between wetland and upland) with the brackish wetland. The transition zone was easier to delineate on the Skylab photograph than on low-altitude-aircraft photographs. As a result of these studies, Anderson (ref. 2-21) modified the classification system to separate saline wetlands that have been affected by human activity from unaffected saline wetlands (a naturally occurring state).

Migratory waterfowl habitat evaluation.-The major objective of the migratory waterfowl habitat investigation (conducted in eastern North Dakota) was to monitor changes in the breeding habitat of migratory waterfowl between May (the peak nesting season for several....


[
55]

FIGURE 2-21.- (b). Concluded.

FIGURE 2-21.- (b). Concluded. [For a larger picture, click here]

 

....species of ducks) and July or early August (when most duck broods have hatched). Proposed indicators of habitat quality were surface water, general degree of terrain wetness, plant phenology, and land use patterns. Primary emphasis was placed on the observation of surface-water features (ponds and lakes) to obtain statistical data on the number of surface-water features and their areal extent, distribution, and frequency. Such information is used in models for predicting annual waterfowl production. The EREP data were not obtained over the test site during the May 1973 breeding period but were obtained on June 12,1973, between the May and July dates desired. Results of the analysis of the EREP data were therefore compared to those obtained by Landsat on May 14 and July 7,1973.

In this Skylab study, the S192 scanner system data were analyzed by digital computer analysis techniques. The use of computer data-processing techniques is particularly well suited to this type of analysis because of the wide expanse of prime waterfowl-breeding area involved and because of the need to quickly assimilate and collate information on habitat conditions.

 


[
56]

FlGURE 2-22.

FlGURE 2-22.-A segment of a computer-generated surface-water map produced by Processing band 11 (1.55 to 1.75 µm) data from the Skylab S192 multispectral scanner observation of an area north of Jamestown, North Dakota. The open surface water is shown in black. [For a larger picture, click here]

 

To delineate surface water boundaries, a level-slicing technique was used to analyze the data from a single near-infrared band. This reasonably reliable and fairly simple method was effective because of the very high absorption and therefore low reflectance of water in these near-infrared wavelengths. Preliminary study of the data indicated that any one of the five wavelength bands from 0.78 to 1.75 µm would be potentially useful for discriminating open surface water by using such a thresholding technique. Thus, the EREP data offered an opportunity to appraise the relative usefulness and reliability of several different near-infrared wavelength bands. The results of this evaluation indicated that bare soil was the terrain feature most likely to be mistaken for open surface water. Although the problem was not severe, there did seem to be a tendency for water and bare soil to have some overlap in level of spectral response. The preliminary work with the S192 data indicated that this overlap decreased with increasing wavelengths. Thus, the 1.55- to 1.75-µm wavelength band was the most useful single band for effective water discrimination. Using the thresholding technique and the 1.55- to 1.75-µm wavelength band, a computer-generated thematic map identifying open surface water was obtained for a 3618-km2 area (fig. 2-22).

In a comparison of the Skylab results with those from analysis of Landsat data collected in May and July 1973, Gilmer and Work (ref. 2-15) found that, on a synoptic basis, the two sensor systems appeared to provide consistent answers in that both data sets indicated a decline in area and number of surface water features. A more detailed comparison of the size of 21 individual lakes that appeared on all data sets indicated that the Skylab multispectral scanner was not capable of achieving as consistent a measure of area as the Landsat scanner. The investigators attributed this result to the conical scan configuration used by the Skylab S192 scanner. Gilmer and Work believed that both the scanning format and the associated techniques for data processing appeared to have the net effect of slightly but [57] systematically altering the measurements and the geometric fidelity of small ponds.

Another phase of this study involved a more limited testing of a different technique for improving the apparent spatial resolution of the Skylab S192 data. This technique was described as a proportion estimation technique and involved the use of a computational algorithm for estimating the fractions of pure materials present within the resolution cell of a multispectral scanner. The details of this technique are described in section 6. Results of this analysis indicated that the minimum discernible size of a water body was fourtenths of the minimum size that could be detected by using the single-band thresholding algorithm. Therefore, this technique seemed to offer considerable promise for mapping and tabulating much smaller water bodies than could be achieved by using the thresholding technique. Gilmer and Work also found that, by using the proportion estimation technique, some lakes that had been only partly defined with Landsat data could be fully defined with the S192 data. Such lakes were shallow and alkaline, with a high level of suspended solids and/or precipitated alkali sediment. The improved capability for delineating such lakes was attributed to the middle-infrared spectral bands associated with the S192 data that extended to 2.35 µm, whereas the Landsat scanner system had only two near-infrared wavelength bands with a maximum wavelength of 1.1 µm. The 1.55- to 1.75-µm wavelength band is particularly effective in the delineation of water and hygric-scene features in general.

 

User Evaluations

 

In addition to yielding the specific benefits derived by the individual investigations, the Skylab Program promoted a much broader concept of technology transfer to a diverse user audience. Through the efforts of the Principal Investigators and their direct contact with different user-agency groups, including local, county, regional, State, and Federal user communities, the use and evaluation of the Skylab EREP data have received a significant amount of exposure.

From the initial selection of the Skylab Principal Investigators and extending through the data analysis period, considerable emphasis was directed toward actual or potential user involvement with the data products generated from this unique remote-sensing system.

 

 

TABLE 2-XVll.- Non-Principal-lnvestigator User Agencies.

 

Skylab Principal Investigator

Geographic area

User agencies

.

P.E. Baldridge

Ohio

Department of Economics (community development), Department of Natural Resources (city, county, and regional agencies)

R.L. Brooks

Kentucky

State coal strip mine inspectors, other State officials

J.W. Hannah

Florida

City, county (regional planning)

E.E. Hardy

New York

City, county, regional (State planning and environmental research)

R.M. Hoffer

Colorado

U. S. Forest Service, National Park Service

I.J. Sattinger

Michigan

Bureau of Outdoor Recreation, Oakland County Planning commission, Michigan Department of Natural Resources

C.W. Welby

North Carolina

Department of Natural Resources economic resource planning, evaluation)

 

In some cases, the Principal Investigators themselves were representing various user agencies. In many other instances, the Principal Investigators contacted various agencies having responsibility within their test-site area and worked with the personnel of those agencies to produce specific Skylab products for use and evaluation by those agencies. Table 2-XVII is indicative of the number and variety of non-Principal-lnvestigator user agencies that were involved indirectly in Skylab land use investigations. The following paragraphs cite some examples of the specific types of activities that were involved.

The major user of remote-sensing data for surface mining applications in Ohio (the Ohio Department of Natural Resources) indicated that the potential usefulness of EREP-type data for current mining and [58] reclamation activities has been satisfactorily demonstrated and that both satellite and aircraft data are now being used in State of Ohio mining and reclamation programs. Another Skylab investigator worked directly with State of Kentucky coal-mining officials who indicated that if Skylab quality imagery were available on a regularly recurring basis, the overall strip-mining program could be upgraded through its use.

The usefulness of space photographs was cited by the members of the Ohio Department of Economic and Community Development in their study on urban growth encroachment on agricultural land. Participating regional planners concluded that the capability to measure urban growth and its impact on the region could be of immediate use to planners in evaluating the effectiveness of both regional and local policies.

The S192 digital data and S190 imagery were used to derive computer-generated classification maps and photointerpreted land use maps, respectively, for different levels of planning agencies in Florida. A majority of the user group surveyed expressed a strong preference for the largest scale map possible, regardless of the data source used.

In surveying the user evaluation of EREP data, a two-phase approach was initiated-an introductory phase during which the potential user was familiarized with the concept of multispectral analysis and with the use of EREP data products to delineate certain land uses and natural characteristics and an in-depth follow-up interview structured around a detailed questionnaire. These surveys showed distinct differences in response between different categories of users. The regional planners were more interested in long-term interrelationships and were enthusiastic about the EREP data because the synoptic view obtainable from space offers an unparalleled method for accurately showing various land use patterns on a regional scale. Local agencies, involved in day-to-day decisions, generally indicated a requirement for more detailed information than could be obtained from the EREP products. In most instances, their needs could best be met by using data collected from aircraft. The important point to be made is that certain user groups do have a need for such synoptic data, whereas other user groups require more detailed data over smaller areas.

 

SUMMARY

A significant number of Skylab investigators developed and demonstrated the usefulness of the photographic sensor systems and their application to resource inventories and analysis for large geographical areas, as well as to regional and local uses. Both conventional photointerpretation and computer-assisted procedures were effective in the analysis of EREP data. Many special applications (including surface mining operations, wetland area mapping, change detection of urban patterns, and general land use mapping) involving the use of multiband and color-infrared photographs were successfully demonstrated. With few exceptions, repetitive data of the type obtained by Skylab (primarily with the S190B) can meet most inventorying and mapping requirements. Participating planners concluded that the capability to monitor urban growth and its impact on the region has immediate value to land use planners in evaluating the effectiveness of both regional and local policies related to growth.

The Skylab S190B photographs enlarged to scales of 1:63 360 and 1:24 000 provided significant detail for easy and efficient mapping. The S190B photographs were cited by most investigators as being superior to the S190A photographs because of the greater amount of detailed information that could be derived from these photographs. Studies involving urban areas were particularly amenable to the use of the S190B color film because this film provided adequate spatial resolution and therefore the most detailed information. Also, many investigators stated that the color-infrared photographs were invaluable for many applications requiring identification of vegetative cover. Therefore, an optimum system for many investigations would have combined the high-resolution color film and the high-resolution color-infrared film (SO-131) into a dual S190B camera system. A specific recommendation was that future orbiting space stations include a multispectral camera array composed of four S190B type cameras. This design would incorporate the added flexibility that is essential to expand and enhance the type and quality of information for general land use programs.

A group of Skylab investigators provided much in [59] sight into the use of computer-processing techniques for analyzing S192 multispectral scanner data for land use applications. Overall accuracies of 75 to 90 percent for classification of Level III and use maps were achieved by using such analysis techniques. Areal estimates based on computer-aided analysis were highly correlated with those obtained through standard photointerpretation techniques applied to aircraft photographs.

The increased spectral range, from the visible through the thermal-infrared wavelength, offered by the S192 scanner system provided investigators with the first opportunity to analyze this wide spectrum of data from satellite altitudes. The consensus of the investigators was that at least one wavelength band from each of the four major portions of the electromagnetic spectrum (visible, near infrared, middle infrared, and thermal infrared) was necessary to achieve optimal computer classification of land use categories. The near-infrared portion of the spectrum was found to be particularly important for accurate discrimination among various vegetative cover types. In other studies, a combination of six wavelength bands was cited as being optimal for land use mapping with the use of computer-aided analysis techniques.

In general, the Skylab S192 land use investigators concluded that the improved spectral resolution and the increased spectral range available in the S192 scanner systems (as compared to the Landsat-l system) enabled significant improvement in classification performance for land use mapping. A few investigators indicated that the improved spectral resolution obtained in the Skylab scanner data was more important for mapping many cover-type features than was the spatial resolution obtained through use of the S190 photographic sensor systems.

In one investigation, camera data from Skylab and Landsat were geometrically corrected to a topographic map (scale, 1:24 000) of the Durango, Colorado, area for the purpose of quantitative and qualitative comparisons and analyses. The use of various analysis techniques with this data set provided some insight into the value of working with topographic data in conjunction with multispectral scanner data for land use and major cover-type mapping in a topographically and vegetationally complex mountainous region.

In nearly every study in which S192 data were used in conjunction with computer-aided analysis techniques, the investigators concluded that traditional definitions of land use categories often will not produce spectrally separable informational classes of data output. To obtain maximum benefit from multispectral scanner data, it will be necessary, in many cases, to establish land use category definitions that are based on spectrally discriminable classes of cover type.

The Skylab EREP experiment demonstrated the value of photographic and multispectral scanner data obtained from satellite altitudes for many land use mapping activities.

 

CARTOGRAPHY

The process of producing and maintaining quality maps and other precision cartographic products is complex, time consuming, and costly. Despite the obvious need for more and better maps and the use of highly sophisticated equipment and techniques in their production, it is estimated that only about 30 percent of the Earth's landmass is now adequately mapped. Furthermore, in rapidly developing locales, maps are often obsolete by the time they are constructed and published.

Before the advent of practical aerial photography, maps were made in the field by teams of cartographers who painstakingly measured their way over the landscape. During the 1920's and 1930's, significant advances were made in the development of aircraft, photography, and optics. These advances made photogrammetry-the art and science of deriving reliable measurements from photographs-increasingly important in the production of maps. Through the use of specialized cameras, customized photographic flight equipment, and complex monoscopic and stereoscopic plotting equipment, a major portion of the mapmaking process was shifted from the field to the office. This use of aerial photographs made it possible to accelerate map production and to produce maps having increased geometric accuracy and detail. In recent years, cartographers have turned to large-scale, high-speed electronic computers and improved mathematical techniques to increase the speed and accuracy of map pro-[60] duction. Today, photogrammetrists and cartographers consider photographs and digital imagery from orbiting spacecraft as the next major step in the preparation and revision of many types of maps.

The Skylab S190A and S190B camera systems produced satellite photographs of relatively high metric and resolution qualities and provided cartographers with a viable means of testing and evaluating the cartographic potential of space photographs. In the United States, the primary problem in mapping is the revision and updating of existing maps. In many other regions of the world, particularly in developing countries and/or remote areas, the production of new maps is of paramount concern. Most of the investigations using Skylab S190A and S190B photographs for original mapping were conducted by agencies or organizations outside the United States. The purpose of this subsection is to summarize the results of the Skylab EREP cartographic investigations and to indicate possible improvements in cartographic instrumentation and techniques for future space missions.

 

Sensor Technology

In considering the results of experiments performed by the cartographic investigators, it is important to note that neither the S190B terrain camera nor the cameras comprising the S190A multispectral array were designed for cartographic applications. However, the S190A and the S190B did represent significant metric and resolution advances in camera systems for Earth photographic observations from space, and several cartographers attempted to exploit fully some of the photographs.

 

Applications

Aside from charts, which are special-purpose maps used for air or water navigation, the most widely used cartographic products are planimetric and topographic maps and controlled photomosaics. Planimetric maps reveal only the horizontal locations of surface features, whereas topographic maps show additionally the vertical positions of features by displaying relief in some measurable form. On maps, relief is depicted by a contour line, which is an imaginary line on the ground that connects all points that are at the same elevation above a specific datum surface (usually mean sea level). A photomosaic is a continuous photographic representation of a portion of the Earth's surface, prepared by assembling individual photographs that have been rendered "tilt free" by a process termed rectification.

Planimetric and topographic maps and photomosaics are produced in a wide range of scales; however, most of these items range from 1:24 000 (1 cm equals 0.24 km) to 1:500 000 (1 cm equals 5 km). in the United States, scales of the standard national cartographic products are 1:24 000, 1:62 500, and 1:250 000. In metric-system-oriented parts of the world, the more commonly used scales are 1:25 000, 1:50 000, 1:100 000, 1:250 000, and 1:500 000. Maps having scales between 1:75 000 and 1:600 000 are generally classed as medium-scale maps, whereas maps having scales greater than 1:75 000 are considered large-scale maps. In most usages or applications, one generally seeks the smallest scale map capable of depicting the degree of detail required to support the particular application.

Photogrammetric mapping requires that the position and the orientation of the camera taking the photograph be determined at the instant of exposure. This information is generally obtained by means of a network of photoidentifiable "control points," for which the horizontal and/or vertical locations have been established by ground survey. After a basic network of ground control has been established, photogrammetric triangulation methods are usually used to extend the basic control network. For each photograph being used in photogrammetric mapping or in the preparation of controlled photomosaics, six to nine control points, well distributed over the format, are required. The establishment of basic ground control and densification of the control network are usually the most costly and time-consuming portion of the overall photogrammetric mapping process, especially in remote regions.

If camera parameters such as lens focal length and film format remain constant, the higher the altitude from which a photograph is taken, the greater the ground area that appears on the photograph. An increase in altitude reduces the number of photographs required to cover a given area and, most importantly, increases the distance between the required surveyed ground control points and thus reduces the overall number of mandatory control points. The formats of the S190A and S190B are significantly smaller than conventional mapping cameras (5.7 cm and 11.4 cm, [61] respectively, compared to 22.8 cm), and the 45.72-cm focal length of the S190B is three times longer than the lens used in many conventional mapping cameras. Nevertheless, as shown in figure 2-23, the increase in area of ground coverage by the S190A and S190B systems is striking by comparison to conventional aircraft photographs.

Most of the Skylab EREP investigative efforts in cartography were directed toward (1) the revision and updating of existing maps, (2) the establishment of photogrammetric ground control, and (3) the construction of new planimetric and topographic maps and photomosaics. In almost all investigations, the emphasis was directed toward determining the largest scale mapping task or product that the S190A and S190B Skylab photographs were capable of supporting.

Many innovations of procedures and techniques have been developed over the past few years to improve the technology of revising and updating maps. The use of remote-sensing data from space platforms for this purpose is not new; limited experimental revision products have been published, based on Gemini, Apollo, and Landsat data. Because Skylab photographs have improved spatial and spectral resolutions, several map revision projects have been accomplished with the use of Skylab data.

The most extensive cartographic investigation was conducted by 17 Latin American cartographic agencies through the Inter-American Geodetic Survey at Fort Clayton in the Panama Canal Zone (Staples et al., ref. 2-22). A few representative examples are cited in the following paragraphs to provide some insight into the usefulness of and the economic benefits derived from Skylab-quality photographs.

A map revision project performed by Fernandez (ref. 2-22) for a 1:50 000-scale map of Santa Cruz, Bolivia, reveals the economy of such revision procedures. Stereopairs of black-and-white S190A (0.6 to 0.7 µm) photographs at a 1:1 500000 scale (enlarged 2 x from the original scale) were used in a stereoplotter to compile planimetric features. In this study, the image quality of the S190A photographs limited the detection of changes to linear cultural features (roads) and the extent of new major urban growth patterns. Major changes to the river channels were also compiled. Significantly, the map revision was accomplished within a 24-hour period, with the use of existing photogrammetric equipment.

 


FIGURE 2-23.

FIGURE 2-23.-Relative areal coverage of the S190A, S190B, and conventional aircraft camera systems. The coverage is roughly square; therefore, the side dimensions given in kilometers and statute miles are typical. [For a larger picture, click here]


[
62]

FIGURE 2-24.

FIGURE 2-24.-Example of map revision using Skylab S190A station 5 photographs. Original map scale is 1:50 000. Changes to original map are overprinted in pink. [For a larger picture, click here]

 

[63] Investigators in several other Latin American countries updated existing maps with S19OA photographs. In Chile, Puccio (ref. 2-22) also used S19OA (0.6 to 0.7 µm) film to revise a 1:50 000-scale map (fig. 2-24). The black-and-white negatives were photographically enlarged and rectified, then registered to the topographic map. Puccio indicated that the technique was inexpensive and that significant planimetric features such as urban density patterns and major new roads were extracted easily.

Although the S19OA photographs were used in some map revision activities, most investigators preferred the S19OB photographs because of the spatial resolution of as much as 15 m. The type of scene contrast and image quality determines the extent to which S19OB photographs will enable identification of individual features that can be compiled onto 1:50000-scale maps. With optical enlarging viewers and/or stereoplotters, many small nonlinear surface features are discernible and can be plotted. However, the capability to define some required features depends on the availability of ground-truth data. The S19OB data are mainly color photographs, but selected data passes were also taken with black-and-white and color-infrared film. The S19OB field of view of 11881 km2 provides coverage of most major metropolitan areas with excellent image quality for map scales of 1:250 000 and larger. A 9.5 x enlargement will provide a scale of approximately 1:100 000, and enlargements of twice this scale to a 1:50 000 scale still provide sufficient image quality to enable extraction of map data directly from the photographic print or from optical enlargers and/or stereographic plotters.

In Argentina, Micro (ref. 2-22) prepared a partial revision of the Chascomus 1:250 000-scale planimetric map, using a black-and-white print from the S19OB color film. He indicated that the data enabled the mapping of new roads, rivers, lakes, small urban areas, and shorelines in this topographically subdued region along the Atlantic coastline of eastern Argentina. Railroads could be detected when they were parallel to roads, and farm buildings were visible when grouped. Other investigators (Romero, Venezuela (ref. 2-22); Morrell, Dominican Republic (ref. 2-22); Stewart, Canada (ref. 2-23); and Mott et al., England (ref. 2-24)) indicated that the S19OB color film provided better interpretative details for updating 1:100 000- and 1:50 000-scale products. This detail is primarily due to the scene contrasts in the particular frames used rather than the resolution characteristics of the color film.

Excellent examples of large-scale planimetric map revision are two 1:24 000-scale maps of Lubbock, Texas, prepared by the Defense Mapping Agency Aerospace Center in St. Louis, Missouri. The S19OB color film for this task was used in January 1974 when the vegetation was dormant, and this condition afforded excellent scene contrast. The revision mapping was accomplished at the original 1:24 000 scale, and polyester matte positives of two topographic map sheets were used for the compilation base. A color photographic print (1:48 000 scale) and a color transparency (1:200000 scale) were enlarged from second-generation materials. The area of interest on this Skylab frame covered approximately 3 percent of the total photographic area. Planimetric features were transferred to the map base compiled by use of a zoom transfer scope. This instrument provides a capability for viewing the photographic image and the map base simultaneously and thus enables the operator to revise or add detail in relationship to existing map features. A portion of an enlarged S19OB photograph used for this revision is shown in figure 2-25(a). Figures 2-25(b) and 2-25(c), respectively, depict only a small part of a 1:62 500-scale map produced in 1957 and the changes and revisions derived from the Skylab imagery. The photointerpretation and compilation for this map revision effort of the two 1:24 000-scale Lubbock topographic sheets required 72 man-hours. In this case, reliable interpretation of ground features was accomplished on 40x enlargements (from the original scale).

In other examples, both Stewart (ref. 2-23) and Colvocoresses (ref. 2-25) indicated that, because of errors in photointerpretation, revisions to 1:50 000-scale maps with the use of S19OB data were not completely reliable. They do, however, indicate that the S19OB data can be used for partial map revision activities if the quality of the original film data is maintained.

 

 


[
64 - 65]

FIGURE 2-25.- (a), (b).

FIGURE 2-25.- (a), (b). [For a larger picture, click here]

 

FIGURE 2-25. (c)

(c)

FIGURE 2-25.-Planimetric map revision of Lubbock, Texas, and vicinity. (a) Enlargement of a portion of an S190B color photograph (SL4-94-111). (b) Planimetric map produced in 1957 by using conventionaI photogrammetric methods. Scale of originaI map is 1:62 500. (e) Revision to planimetric map shown in figure 2-25(b) as derived from S190B photograph. Changes are shown in color. [For a larger picture, click here]

 

Photogrammetric Establishment Of Ground Control

The establishment of ground control is one of the most expensive and time-consuming tasks in cartography. Three quantitative investigations by three separate agencies in Canada, each using different techniques of aerial triangulation adjustment (Stewart, ref. 2-23), and an investigation in England (Mott et al., ref. 2-24) demonstrated that horizontal photogrammetric control capable of supporting planimetric mapping at scales of 1:250 000 and smaller can be derived from Skylab S19OA photographs. These same investigators, and Keller (ref. 2-26) of the National Oceanic and Atmospheric Administration, found that photographs from the S19OB camera system, with its longer focal length and higher resolution, were capable of providing horizontal control sufficient to sustain the compilation of planimetric maps at scales of 1:50 000 and smaller.

In his triangulation investigation, Keller used a strip of 12 S190B photographs that covered a 648-km-long swath from Charlotte, North Carolina, to the Rappahannock River in Virginia. Within this area were 29 highly identifiable ground control points (road intersections, aeronautical aids, and airport runway ends) for which position could be determined from standard 1:24 000 scale USGS quadrangle maps and National Ocean Survey (NOS) airport surveys. A standard NOS computational program to perform numerous simultaneous adjustments of the 12-photograph strip was used in the analysis. In these adjustments, different subset combinations of the ground control and different empirical weightings of the ground control and photographic measurements were tried. The best results were [66] obtained using a 14-point network of ground control. The remaining 15 points of known location were used for accuracy evaluation. This "best" solution yielded a root-mean-square (rms) horizontal position error of 15 m, with 25 m being the maximum error observed.

The ability to derive vertical control or contour information from aerial or orbital photographs is largely a function of a characteristic of overlapping stereoscopic pairs of photographs, termed base/height (B/H) ratio. The B/H ratio of such a pair of photographs is the ratio of the distance between the camera exposure stations to the distance, or altitude, of the cameras above the ground. In conventional aerial photography (15.25-cm focal length, 23- by 23-cm format) with 60-percent forward overlap, this ratio is approximately 0.6. Because of the higher altitude of the EREP and smaller film format (and longer focal length for the S190B), with 60 percent forward overlap, the B/H ratio is 0.15 for the S19OA and 0.10 for the S19OB. Despite the small B/H ratio, Mott et al. (ref. 2-24) performed a vertical adjustment of a strip of seven S19OB photographs taken over the rugged terrain of Nepal and achieved an rms height error of 117 m.

In an effort to overcome the small B/H ratio of vertical EREP photographs, triangulation was performed on a two-strip block of obliquely convergent photographs taken over Paraguay (ref. 2-22). Horizontal accuracies of approximately 15 m and vertical accuracies of approximately 25 m were achieved with an rms error for photographic image measurements of 8 µm. Only 10 original ground control points were minimally required to accomplish triangulation over a 50 000-km2 area; however, because of availability, 40 ground control points were used. This block had a B/H ratio of approximately 0.9 and consisted of a vertical strip of S19OA photographs from one EREP orbital pass and a strip of S19OB photographs from a solar inertial pass from an adjacent orbit. During the S19OB pass, the spacecraft.....

 


FIGURE 2-26.

FIGURE 2-26. Conceptual geometry of vertical S190A and obliquely convergent S190B coverage over Paraguay with a 0.9 base/height ratio. The designator "PP" represents the principal point. [For a larger picture, click here]

 

....was oriented (pitched and rolled) such that the S19OB camera photographed almost the identical ground area photographed during the S19OA pass, as conceptually demonstrated in figure 2-26. This triangulation is capable of supporting planimetric mapping at scales of 1:50 000 and smaller and topographic mapping at scales of 1:100 000 and smaller.

 

[67] Original Mapping Activities

The use of Skylab photography to revise reconnaissance maps of regions in Central America and South America was reported by Staples et al. (ref. 2-22). in Guatemala, a 1:50 000-scale planimetric map (shown in reduced form in fig. 2-27(b)) was produced from a pair of S190B black-and-white photographs enlarged to 23 by 23 cm (9 by 9 in.), by using a stereoplotter. Outlines of urban areas, highways, railroads, and natural features were mapped. A 2486-km2 region in and around Concepcion, Paraguay, was planimetrically mapped at a scale of 1:100 000 by using approximately 25 percent of the stereomodel of two S19OB 2 x enlargements. A total of 36 man-hours was expended on compiling this map. If aircraft photographs had been used, 50 stereomodels and approximately 250 man-hours would have been required to accomplish the same task.

A planimetric map (scale, 1:100000) of Fundacion, Colombia, was prepared from 29.2- by 29.2-cm S19OB color diapositives and a universal stereoplotter (Fletcher, ref.2-22). A portion of this map (fig.2-28(b)) shows the location of the transportation networks, other cultural features, hydrographic features, and forests, as compared to a similar map of this area published in 1954 (fig. 2-28(a)). Topography on the Skylab map was transferred from the 1954 map sheet. A similar map (not shown) was constructed from 1:50 000-scale aircraft photographs to serve as a basis of comparison; the precision of the map prepared from Skylab photographs was found to be within acceptable limits for the 1:100000 scale. With use of the Skylab photographs, the mapping of the 2400-km2 area was accomplished in 72 man-hours, a time element approximately 8 times less than that required for mapping with use of the 1:50 000-scale aircraft photographs.

Despite the poor B/H ratios, vertical Skylab imagery and the previously mentioned obliquely convergent photographs were used to produce topographic maps of portions of the Himalaya Mountains, Arizona, and Paraguay.

Mott et al. (ref.2-24), using 2.5 x glass diapositives of vertical S19OA photographs and a first-order photogrammetric stereoplotter, produced 1:500 000- and 1:62 500-scale maps, both with 250-m contour intervals, of portions of the rugged Himalaya Mountains. In this analysis, only techniques and instrumentation commonly used in commercial mapping companies were used. Topographic mapping was performed by Goetz et al. (ref.2-27) with the use of partial frames of S19OB color photographs of central Arizona (fig. 2-29). An analytical plotter, contact diapositives, and ground control from 1:24 000- and 1:62 500-scale USGS topographic maps were used to produce a 1:100000 scale topographic map with 150-m contours covering a 57- by 66-km area of the Verde Valley. A portion of this map, at one-half scale, is shown in figure 2-29(b). The highly convergent configuration created by a vertical strip of S19OA photographs coupled with am oblique-looking strip of S19OB photographs was used to map portions of Paraguay for which reconnaissance maps were available. An analytical plotter was used to produce 16 full and 13 partial topographic map sheets at a scale of 1:100 000 with 200-m contour intervals. An example of one of these map sheets, reduced to page size, is shown in figure 2-30 and illustrates the value for this type of mapping activity.

An experimental 1:250 000-scale photomap covering the same area as the standard 1° by 2° topographic sheet of Hartford, Connecticut, was produced by the USGS. The mosaic (fig. 2-31) was assembled from portions of four S19OA black-and-white (0.6 to 0.7 µm) frames by using a photomechanical film mosaic process. Another photomap of the Hartford area was prepared at a 1:100000 scale with use of the S19OB color-infrared (0.5 to 0.88 µm) film. Both products meet national map accuracy standards for positional accuracy.

 


[
68 - 69]

FIGURE 2-27. (a)

(a)

FIGURE 2-27.-Maps of Escuintla, Guatemala, and vicinity. (a) Conventional topographic map published in 1973 at a scale of 1:50 000. (b) Planimetric map compiled from S190B frames SL4-89-290 and SL4-89-291 obtained in February 1974. [For a larger picture, click here]

FIGURE 2-27.- (b)

FIGURE 2-27.- (b) [For a larger picture, click here]


 


[
70 - 71]

FIGURE 2-28. (a)

(a) [For a larger picture, click here]

FIGURE 2-28.-Maps of Fundacion, Colombia, area. (a) Conventional topographic map published in 1954 at a scale of 1:100 000. (b) Planimetric map prepared in 1975 from S190B color photograph. Topographic details were transferred from the map shown in figure 2-28(a).

FIGURE 2-28.- (b)

FIGURE 2-28.- (b) [For a larger picture, click here]


 


[
72 - 73]

FIGURE 2-29. (a)

(a)

FIGURE 2-29.-Verde Valley, central Arizona. The broad, flat valley dominates the right half of the area shown. (a) S190B photograph (SL4-90-305). (b) Topographic map prepared using photograph shown in figure 2-29(a). (Original scale, 1:100 000.) [For a larger picture, click here]

FIGURE 2-29.- (b).

FIGURE 2-29.- (b). [For a larger picture, click here]

 


[
74]

FIGURE 2-30.

FIGURE 2-30.-A topographic map of a portion of Paraguay produced from a vertical S190A frame coupled with an obliquely convergent S190B frame. The map is a reduction of the 1:100 000-scale compilation. [For a larger picture, click here]

 


[
75]

FIGURE 2-31.

FIGURE 2-31.-A photomap of Hartford, Connecticut, and vicinity, produced from portions of four S190A black-and-white photographs. The format of the photomap corresponds to the existing USGS 1° by 2° series map sheet. The map is a reduction of the 1:250 000-scale compilation. [For a larger picture, click here]

 

[76] In Canada, Stewart (ref. 2-23) prepared a sample photomap that covered approximately one-third of a standard 1:250 000-scale map sheet. The S19OB black-and-white data were rectified directly, in one stage, and provided an excellent image (3.8x enlargement) base and planimetric accuracy. Photomapping the same area from standard aerial photographs (1:36 000 scale) required 240 prints. With an increase in the number of photographic prints required to cover an area, density and contrast-matching problems, as well as defects due to banding and tone control, occur. Space photographs such as those obtained from the S19OB system eliminate the majority of these defects, and large areas can be mapped by using relatively few space photographs.

A very practical feature of vertical S19OA- and S19OB-type data is the capability for use as uncontrolled photomaps. Individual frames can be directly enlarged to the specific scale and format that permits direct comparison between the enlarged photograph frame and existing line maps. The enlarged frame is relatively free of distortions, and positioning to map control is more than adequate for visual comparisons of changes in map detail. Photographs from the S19OB system have been enlarged to the 1:24 000 scale of standard topographic maps (approximately 40 x enlargement) to permit visual comparisons of the cultural changes that have occurred since the map was published.

 

SUMMARY

 

The accomplishments and results of the various Skylab EREP cartographic experiments have demonstrated the potential and the practicality of using quality photographs from orbiting spacecraft as a means of preparing and updating certain types of maps and other cartographic products. It was clearly shown that, with suitable spectral resolution and geometric fidelity, photographs from space can serve as an adequate source for a variety of cartographic products at scales of 1:100 000 and smaller. With improved resolution, this scale limit might be improved to 1:50 000 or even 1:24 000. It was concluded that Skylab EREP space photographs cannot completely supplant aircraft photographs and ground-truth information for cartographic applications at scales of 1:100 000 and larger. The need for convergent photographs to enable detailed topographic mapping from space images was demonstrated.

Little effort was expended on investigating the spectral aspects of the Skylab imagery in cartographic applications. This area should be specifically marked for future study.

To accomplish most cartographic objectives, cloud-free photographic coverage at repeated intervals is required over most regions. Because of problems created by weather, a satellite totally dedicated to cartography seems to offer the highest potential for satisfying this need. Such a satellite, with one or more high-resolution, large-format, long-focal-length cameras designed to photogrammetric standards, would provide a practicable means of solving many terrestrial mapping problems. Such a system would prove especially valuable for the establishment and densification of ground control, the construction of new maps and photomosaics in the more remote regions of the Earth, and the revision and updating of existing cartographic products.

 


REFERENCES

 

2-1. Anderson, J. R; Hardy, E E.; and Roach, J. T. A Land-Use Classification System for use With Remote sensor Data. U.S. Geol. Survey Circ. 671,1972.

2-2. Hannah, John W.; Thomas, Oarland L.; and Esparza, Fernando: Planning Applications to East Central Florida. NASA CR-145415, 1975.

2-3. Stoeckeler, E G; Woodman, Raymond G; and Farrell., Robert S. Multidisciplinary Analysis of Skylab Photography for Highway Engineering Purposes. NASA CR-141942, 1975.

2-4. Colwell, Robert N.; Bowden, Leonard W.; et al: Use of Skylab Imagery to Assess and Monitor Changes in the Southern California Environment. NASA CR-147561, 1974.

[77] 2-5. Hoffer, Roger M.: Computer-Aided Analysis of Skylab Multispectral Scanner Data in Mountainous Terrain for Land Use, Forestry, Water Resource, and Geologic Applications. NASA CR-147473, 1975.

2 6. Poulton, Charles E.; and Welch, Robin 1.: Plan for the Uniform Mapping of Earth Resources and Environmental Complexes From Skylab imagery. NASA CR-144484, 1975.

2-7. Hardy, E. E.; Skaley, J. E.; et al.: Enhancement and Evaluation of Skylab Photography for Potential Land Use Inventories. NASA CR-144473, 1975.

2-8. Cooper, Saal; Anderson, Duwayne; et al.: Skylab imagery: Application to Reservoir Management in New England. NASA CR-144514, 1975.

2-9. Klemas, Vytautas; Bartlett, David S.; et al.: Skylab/EREP Application to Ecological, Geological, and Oceanographic Investigations of Delaware Bay. NASA CR-144910, 1976.

2-10. Simonett, David S.: Application of Skylab EREP Data for Land Use Management. NASA CR-147457, 1976.

2-11. Alexander, Robert H.; and Lins, H. F., Jr: Selected Applications of Skylab High-Resolution Photography to Urban Area Land Use Analysis. NASA CR-139997, 1974.

2-12. Baldridge, Paul E.; Goesling, P. H.; et al.: Utilizing Skylab Data in On-Going Resources Management Programs in the State of Ohio. NASA CR-134938, 1975.

2-13. Welby, Charles W.; and Lammi, J. O.: Utilization of EREP Data in Geological Evaluation, Regional Planning, Forest Management, and Water Management in North Carolina. NASA CR-144104, 1975.

2-14. Silva, Leroy F.: A Study of the Utilization of EREP Data From the Wabash River Basin. NASA CR-147412, 1976.

2-15. Gilmer, Davis S.; and Work, Edgar A. Jr.: Utilization of Skylab (EREP) System for Appraising Changes in Continental Migratory Bird Habitat. NASA CR-147542, 1975.

2-16. Higer, A. L.; Coker, A. E.; Schmidt, N. F.; and Reed, L. E.: An Analysis and Comparison of Landsat-1 Skylab (S192) and Aircraft Data for Delineation of Land-Water Cover Types of the Green Swamp, Florida. NASA CR-144855.1976.

2-17. Polcyn, Fabian C.; Rebel, Diana L.; and Colwell, John E.: Analysis of Hydrological Features of Portions of the Lake Ontario Basin Using Skylab and Aircraft Data NASA CR-147456, 1976.

2-18. Sattinger, I. J.; Sadowski, F. G.; and Roller, N. E. G.: Analysis of Recreational Land Using Skylab Data NASA CR-144471, 1976.

2-19 Brooks, Ronald L; and Parra, Carlos G.: Applicability of Satellite Remote Sensing for Detection and Monitoring of Coal Strip Mining Activities. NASA CR-144474, 1975.

2-20. Weir, Charles E.; Powell, Richard L.; et al.: Application of EREP imagery to Fracture-Related Mine Safety Hazards in Coal Mining and Mining-Envirommental Problems in Indiana. NASA CR-144495, 1975.

2-21. Anderson, R. R.; Carter, V. P.; and Alsid, L. J.: Skylab EREP Investigations of Wetlands Ecology. NASA CR-139224, 1974.

2-22. Staples, Jack E.; Eoldan, J. J. M.; et al.: Overall Evaluation of Skylab imagery for Mapping of Latin America. NASA CR-144476, 1975.

2-23. Stewart, R. A.: Investigation of Selected imagery From Skylab/EREP S190 System for Medium and Small Scale Mapping. NASA CR-144542, 1975.

2-24. Mott, P. G.; Fullard, H.; et al.: Cartographic Research in EREP Program for Small Scale Mapping. NASA CR-144478, 1975.

2-25. Colvocoresses, Alden P: Overall Evaluation of Skylab (EREP) images for Cartographic Application. NASA CR-147423, 1975.

2-26. Keller, Morton: Analytic Aerotriangulation Utilizing Skylab Earth Terrain Camera (S190B) Photography. NASA CR-144387, 1975.

2-27. Goetz, A. F. H.; Abrams, M. J.; et al.: Comparison of Skylab and Landsat images for Geologic Mapping in Northern Arizona. NASA CR-147503, 1976.


previousindexnext