---

a University of California at Berkeley.

^b U.S Department of Agriculture Forest service.

^c NASA Lyndon B. Johnson Space Center.

^* Principal investigator.

---

[79] AGRICULTURISTS, RANGE MANAGERS, and foresters are primarily concerned with the management of resources such as crops, forage, livestock, timber, soil, and water. If wisely managed, these resources will continue to provide mankind with food, fiber, and shelter; if they are not wisely managed, man's survival will become threatened. During the past 100 years, the extent of resource management and production has accelerated rapidly. Resource managers have identified the need for timely and accurate information on what is being produced; where it is being produced; the general state of health, or condition, of the resource; and the amounts expected to be produced.

In the past decade, the data acquired by spacecraft have provided a new tool to assist in the decision-making processes of the resource managers. The purpose of this section is to show, with specific examples, how the data acquired by the Earth Resources Experiment Package (EREP) can be used to satisfy some of the information needs of resource managers. Numerous investigators addressing different information requirements submitted findings, and the most pertinent examples are cited in this report. The EREP investigators have demonstrated the capability to inventory many different types of agricultural, range, and forestry resources. The results show that the findings obtained at their test sites are applicable to analogous areas throughout the world.

The wise management of Earth resources requires implementing a three-step process: inventory, analysis, and operations. In the inventory step, an area-by-area determination is made of the amount and quality of the existing resources. In the analysis step, management decisions are made with respect to the ultimate use of the resources after consideration of the type and condition of the resources and the cost benefits. In the operations step, the resource manager implements each decision made in the analysis phase; e.g., the decision to apply the appropriate fertilizer in certain mineral-deficient parts of an agricultural area; the decision to practice deferred, rotational grazing in certain parts of a range land area; or the decision to cut only the overmature trees in a certain part of the forest area. The resources are highly dynamic rather than static; to effectively manage them, a new inventory must be obtained periodically (a process known as monitoring).

DATA REQUIREMENTS AND APPLICATIONS

The types of information necessary for monitoring vegetation resources are provided in table 3-I. Federal, State, and county agencies, as well as industrial firms (identified in table 3-II), use this information, which currently is largely obtained by means of ground surveys. Although relatively small amounts of the resource data are obtained by remote-sensing techniques, the use of such remotely sensed data has greatly increased during the last decade. Satisfying the requirements for all these users is complicated because the users want different information about vegetation groupings in....

Level	Agricultural crops	Timber stands	Rangeland forage	Brushland vegetation
.
Federal	Agricultural Stabilization and Conservation Service, Agricultural Conservation Program, Commodity Credit Corp., Agricultural Marketing Service, Statistical Reporting Service, Economic Research Service, Soil Conservation Service, Federal Crop Insurance Corp., Farmers Home Administration, Rural Community Development Service, Foreign Agricultural Service Famine Relief Program. Foreign Economic Assistance Program, Department of Commerce Agricultural Census Program	U.S. Forest Service, Bureau of Land Management, and many Federal agencies listed previously	U S. Forest Service, Bureau of Land Management, and Land many Federal agencies listed previously	Primarily U. S. Forest Service and Bureau of Land Management
State and county	Agricultural Extension Service, State tax authority	Division of Forestry, Forest Extension Service: State tax authority	Livestock Reporting Service, Range Extension Service, State tax authority	Division of Forestry,Division of Beaches and Parks, Water Resource Agency, State tax authority
Private	Fertilizer and Pesticide producers, crop-harvesting industry, food-processing and packing industry, transportation industry, food and fiber advertising and marketing industry wood products	Fertilizer and pesticide producers, logging industry, wood- processing industry, transportation industry, wood and advertising and marketing industry	Fertilizer and pesticide producers, meatpacking industry, tanning industry, transportation industry	Hunting and fishing commissions public utilities commissions local irrigation districts

[81] ....different places, at different times, and with different levels of accuracy. In addition, they have different requirements as to the speed with which vegetation information must be processed after the raw data have been collected, and the frequency with which the information must be updated (table 3-III).

The EREP investigators considered the remote-sensing capabilities that would be necessary to satisfy some of the resource information requirements in table 3-I, and, in this context, they analyzed the EREP data to determine the potential for providing needed information. Although the primary emphasis was given to vegetation resources, it was recognized that agriculturists, range managers, and foresters are also interested in animal resources such as livestock and wildlife. Moreover, they are interested in the entire complex of Earth resources (including soils, water, minerals, and atmosphere) in the areas for which they have management responsibilities. In the following subsections, the results of the EREP investigations are presented for most types of data required to monitor vegetation resources.

Agriculture

Recently, agriculturists were asked to list the specific applications of remote sensing that might prove profitable in terms of cost/benefit ratios and the total savings that might be achieved for each crop. In addition to the types of data shown in table 3-I, they selected the most important candidate problem in U.S. agriculture-the extent of damage done each year to specific crops by specific insects or pathogens.

Range

Range managers are concerned with land management and with animal management on lands that produce mature forage for animal (wild or domestic) consumption. One of the major objectives of a range manager is maximization of the production forage concurrent with conservation of the land resources. In the United States, there are two basic types of range areas, for which the same resource information is required.

Time interval	Agricultural crops	Timber stands	Rangeland forage	Other vegetation (mainly shrubs)
.
10 to 20 minutes	Observe advancing waterline in croplands during disastrous floods,observe the start of locust flights in agricultural areas	Detect the start of forest fires during periods when there is a high "fire danger rating.	Detect the start of range land fires during periods when there is a high fire-danger rating	Detect the start of brushfield fires during periods when there is a high fire danger rating
10 to 20 hours	Map perimeter of ongoing floods and locust flights: monitor the wheat belt for outbreaks of black-stem rust caused by spore showers	Map perimeter of ongoing forest fires	Map perimeter of ongoing rangeland fires	Map perimeter of ongoing brushfield fires.
10 to 20 days	Map progress of crops as an aid to crop identification (using "crop calendars ") and estimate date to begin harvesting operations	Detect start of insect outbreaks in timber stands	Update information on range readiness for grazing	Update information on fires of flowering and pollen production in relation to the bee industry and to hay fever problems
10 to 20 months	Facilitate annual inspection of crop rotation and compliance with Federal requirements for benefit payments	Facilitate annual inspection of firebreaks	Facilitate annual inspection of firebreaks	Facilitate inspection of firebreaks
10 to 20 years	Observe growth end mortality rates in orchards	Observe growth and mortality rates in timber stands	Observe signs of range deterioration and study the spread of noxious weeds	Observe changes in "edge effect" of brushfields that affect suitability as a wildlife habitat
20 to 100 years	Observe shifting cultivation patterns	Observe plant succession trends in timber stands	Observe plant succession trends on rangelands	Observe plant succession trends in brushfields

[82] These areas are (1) Federal and publicly owned lands and (2) State and privately owned grasslands. The Federal and publicly owned lands are located primarily in the 17 Western States and are managed by the U.S. Government. The State and privately owned lands are generally east of the Rocky Mountains and are managed by State governments and/or private owners. The managers for both types of areas have the same objectives, and their management efforts are governed by the information requirements given in table 3-I. Satisfying these objectives requires continual monitoring of all vegetation to avoid forage waste, overgrazing, and damage to the range resources. The private owner acquires this information by frequent in situ observations, but managers of the larger, publicly owned lands can only inspect representative portions and extrapolate the information obtained from sample areas to other nearby regions. This extrapolation is difficult and unreliable because of the many unknown factors, including rainfall patterns, vegetation differences, and habitats of livestock and wildlife.

Forestry

Foresters need essentially the same information (table 3-I) as that required by agriculturists and range managers. Using a somewhat different approach, foresters indicated that remote sensing can be especially helpful by providing information on which to base multiple-use decisions relative to each part of the forest. The multiple-use concept is complex and more applicable to forestry than to agriculture. Some foresters prefer to have all parts of an area managed with respect to maximization of timber production; others want to preserve the forests as primeval museums to be enjoyed in perpetuity. Between these two extremes are those who condone each of these uses for specific parts of a forest if such use does not interfere with the use of the forest primarily as a source of water for domestic use and minerals for industrial use and the preservation of esthetic qualities for recreational use. To support intelligent decision-making processes regarding the best use of each part of the forest, two major types of information are needed: (1) a resource map that accurately delineates the forest vegetation and all associated resources (soils, water, minerals, etc.) and (2) adequate sociological, economic, and technological data to ensure that the forests can produce "the greatest good for the greatest number." Forestry applications addressed only the first of these two complex and interrelated requirements, a resource map of forested areas. The authors of this section believe that management techniques can be enhanced by the use of space-acquired information, which will assist the large-area manager in decision-making processes.

INVENTORY BY PHOTOINTERPRETATION

A primary objective of this section is to discuss the extent to which EREP data will satisfy the informational needs of various users in the fields of agriculture, range management, and forestry. The EREP investigators identified the type of data and analysis techniques that would provide the required information for resource management. The usefulness of the information was evaluated in terms of type of data, frequency of observations, need for ancillary data, and method of using the data for resource management decisions. The EREP investigators, both domestic and foreign, have demonstrated the capability to inventory many different resources within the disciplines of agriculture, range management, and forestry.

Crop and Acreage inventory

Agricultural land use and crop-producing areas are discernible in photographs taken with the Multispectral Photographic Camera (S190A) and the Earth Terrain Camera (S19OB) when the photographs are analyzed by standard photo-interpretative techniques. The crops identified were citrus, coffee, sugarcane, and wheat.

The S19OB color and color-infrared photographs of the Rio Grande Valley, enlarged to a scale of 1:63 000, were projected onto a standard viewing screen, resulting in a scale of 1:10000 (Hart et al., ref. 3-1), and interpreted by an agricultural analyst to identify vegetative patterns and to discriminate citrus, sugarcane, and crops grown in larger fields. This data format (figs. 3-l(a) and 3-l(b)), when supported by transect low-altitude-aircraft data (fig. 3-l(c)) and ground truth (fig. 3-l(d)), provided the information required for a field-by-field interpretation. When color-infrared photographs were analyzed, only annual crops and fallow land were identified with 100 percent accuracy. Citrus....

[84] ...was identified with an accuracy of 93 percent when color-infrared film was used and with an accuracy of 80 percent when conventional color film was used. However, when both film types were analyzed concurrently, citrus was identified with 100 percent accuracy. This analysis identified the areal extent of frost damage to sugarcane growing in the test area.

The areal extent of frost damage in July 1973 (de Mendonca et al., ref. 3-2) to the coffee of Maringa, Parana, Brazil, was assessed using a multilevel surveying system in which Landsat, Skylab, low-altitude aircraft, and fled visits were the information sources. An S19OB high-resolution color photograph (fig. 3-2(a)) acquired on August 8, 1973, was used to delineate the coffee- and wheat-growing areas (fig. 3-2(b)) and to serve as a training area for analysis of Landsat computer-compatible tapes using an interactive multispectral image analysis system. The descriptor, which enabled identification of the two crops, was the cultural pattern shown in figure 3-2(a). The analysis, in which classification results based on multistage sampling were used, indicated that 852 884 hm² (2 107 522 acres) of frost-affected coffee, 33 864 hm² (83 680 acres) of normal coffee, and 302 342 hm² (747 103 acres) of wheat were present in the study area. This multilevel survey of a major agricultural commodity provided a rapid assessment of the economic losses sustained from an act of nature.

When a statistical approach was used to inventory an agricultural area for crop types and acreage, color and color-infrared photographs in a 23.0-by 23.0-cm format were used to identify the differing cultural patterns or strata and were outlined directly on the working data. The photographic texture, tone, color, pattern, size, shape, shadow, location, and associated features were used to identify a stratum (a large homogeneous area) (Colwell et al., ref. 3-3). The textures (coarse, medium, and fine) (fig. 3-3) were indicative of the field sizes in the areas of interest. From experience, the analyst knows that a coarse texture (65- to 32-hm² (160 to 80 acre) field size) indicates a predominance of field crops, a medium texture (12- to 32-hm² (30 to 80 acre) field size) indicates a mixture of field and vegetable crops, and a fine texture (4 to 12-hm² (10 to 30 acre) field size) indicates vineyards and pastures. Color and color-infrared photographs from the S19OA system, enlarged to a scale of 1:805 000 (20.3 by 20.3 cm), provided a scale that was considered optimum for this task.

Color within each stratum indicated the crop type and degree of maturity. Based on photographic texture and color, areas were identified in which to obtain ground-based transects for the development of agricultural statistics (acreage, crop type, crop maturity, irrigated fields, etc.). The image analyst used these statistics to perform stratum-by-stratum inventories for crop acreage and crop class. (Classes were mature or immature for wheat, barley, and safflower.) When this procedure was used, sampling efficiency was increased, classification errors were reduced, and an agricultural stratum inventory was achieved with an accuracy of 10 to 20 percent at the 95-percent-confidence level, with a cost reduction of 15 to 1 (ref. 3-3).

Acreage measurements of three land use categories were made by using stereoscopic study and image magnification of the S19OA and S19OB photographs. When the S19OA photographs were enlarged (20x), forest, bare soil, and wetlands were readily identified.¹ Generally, the S19OA photographs were satisfactory for interpreting gross characteristics only at the regional level. Major boundaries, such as roads and section lines, could not be identified. When forests and generalized crop categories were delineated by using late-summer photographs or the most cloud-free photographs, accuracies of 93 and 94 percent, respectively, were achieved. The quality of the S19OA photographs acquired over a Michigan test site was highly variable between passes and film/filter combinations and between duplicates of the filmstrips. The resolution was marginal for purposes of crop acreage estimation. Analysis of S19OB black-and-white photographs acquired from this same Michigan test site in late summer indicated that section lines, roads, and field boundaries were readily visible because of the improved resolution of the sensing system. The acreage measurements were performed on magnifications of 15x, with an ocular grid having a 0.25-mm line spacing.

When stereoscopic and image magnification procedures were used to estimate the acreage of 170 fields, there was a tendency to underestimate the acreage of fields larger than 4 hm² (10 acres) and a tendency to overestimate on fields smaller than 4 hm² (10 acres).

Even with these biases, an aggregated percentage error of 6 percent below the actual acreage in the test site was achieved. The resolution obtained by the S190B system was judged to be adequate for crop acreage assessment for major field crops, but further improvement in resolution will be useful for crops, such as vegetables, that are often grown on small plots.

Analysis of Digitized Photographs

Automatic data-processing techniques were used in the analysis of EREP photographs converted to a digital format to determine the extent to which agricultural information could be extracted (Colwell et al., ref. 3-3). The four bands of S190A black-and-white photographs acquired over the Salinas Valley of California during the Skylab 2 and 3 missions were digitized with a microdensitometer and stored on magnetic tape. The scale of the photographs was approximately 1:2 850 000, and the scan interval along the X- and Y-axes was 0.0254 cm with an aperture setting of 0.00254 cm. The resulting resolution cell represented a 0.45-hm² (1.12 acre) area on the ground. Actual scanning time was 0.5 hr/band, with 0.5 hour for setup, for a total time of 2.5 hours to digitize four bands of data.

The magnetic tapes were reformatted to be compatible with a computer program that used a five-step procedure to identify the predominant crops present in the valley (i.e., tomatoes, carrots, lettuce, asparagus, and cauliflower). Because tomatoes and lettuce were in distinct stages of development during the September....

.....Skylab 3 overpass, they were further divided into two subclasses: immature and mature.

The initial analysis program stratified the test area and produced gray-scale maps of discrete features to assist in the selection of training areas for the computer. With use of the training area, statistics for classifying each stratum were derived, identifying the mean density, the standard deviation, the spectrograms, and the histograms for each field and class of each band; Correlation and covariance matrices were developed for each field and class for all bands. An optimization of the combination of data bands was performed to give [88] the interclass divergences of each combination of crops for each combination of bands. When each stratum was inventoried, the statistics that had been developed and the optimum bands for classification were used to address each data point (0.45 hm² (1.12 acres)) of the test area, assigning a character to each point that corresponded to one of the training classes. For validation, the stratum was reclassified by using a nearest neighbor algorithm, which also produced an accuracy statement.

When the six bands of the S190A photography were analyzed for crop discrimination, the four bands (i.e., an optimum number of bands for classification is four or less bands of data) that consistently provided the highest accuracies for both the June 2, 1973, and September 13,1973, Skylab data takes were (1) green (0.5 to 0.6 µm), (2) red (0.6 to 0.7µm), (3) infrared I (0.7 to 0.8 µm), and (4) infrared 2 (0.8 to 0.9 µm). Of all the combinations, the green band contributed more information, particularly when used in conjunction with the other bands, because the vegetative cover of the fields was particularly dense and, as a consequence, the reflectance in the green band was high.

When this analysis procedure was applied to the digitized June 2, 1973, S190A photographs, the classification accuracy was low, with an overall 49.5 percent correct and an average of 58.3 percent correct by crop class. However, the training classes of two vegetables (carrots and asparagus) that were in an advanced state of maturity were classified with a high degree of accuracy (89.2 and 98.1 percent, respectively).

When the digitized September 13, 1973, S190A photography was classified, an overall accuracy of 85.1 percent was achieved. Most of these crops were in an advanced state of maturity. For all crops except beans, the correct classifications exceeded 80 percent. The low performance (60.3 percent correct) for beans was attributable to the harvesting of beans throughout the area. These results indicate that when vegetable crops are inventoried by means of remotely sensed data, the proper timing of data collection is more critical than it would be for field crops. Frequently, in a given area, individual vegetable crops will be in different stages of development, whereas individual field crops will be in a common stage of maturity. Therefore, if single-date data are not acquired at precisely the correct time, multidate data will be required for vegetable crop inventories.

Analysis of Multispectral Scanner Data

The Multispectral Scanner (S192) provided vast quantities of narrow-band spectral data that were used for developing techniques to perform automatic classification of agricultural crops and to assist in large area crop inventories. However, even with the use of a computer, the large volume of data presented a major analysis problem.

Based on knowledge derived from analysis of aircraft scanner and multidate Landsat data, the consensus is that when no more than six or eight bands of data are available for analysis, classification accuracies increase as the number of bands of analyzed data increases. The cost of analysis, in both time and money, also increases with an increase in the number of bands of analyzed data. Therefore, the prime task is to select the optimum bands for future classification and to minimize the number of bands required to achieve a specified level of accuracy.

Crop class and subclass signatures were developed from S192 data¹ by using 12 bands. Because of large anomalies in the data, band 2 was deleted. Supervised field clustering using center picture elements (pixels) produced distinct signatures of each class that were not contaminated by border-area pixels. To reduce the cost of classification, the number of signatures was reduced by a band-ranking criterion, which used an average pairwise probability of misclassification based on this selection, and the combination of bands most probable for identification was chosen. This process indicated that bands 5, 8, and 9 provided little aid for discrimination between major ground-cover types. Recognition signatures on three different training sets were identified, and the selected bands are shown in table 3-lV. A test site, approximately 233 km² (90 sections), in eastern Ingham County, Michigan, was chosen for reclassification. The signatures developed from the training sets in the northern half of the test site were used to reclassify 104 km² (40 sections) in the northern half of the county and then to classify 124 km² (48 sections) in the southern half of the test site. The accuracies for....

...reclassification and signature extension for the different signatures are listed in table 3-V. Initial observations indicate an exception to the rule that the more training sets used for signature development, the higher the accuracy of classification. As given in table 3-V, signatures developed from the 20-section training set achieved a better performance accuracy. This accuracy is a result of the procedures followed when developing the signatures from the 40-section training set, in which more signatures were combined or dropped in the initial signature-analysis procedures.

Man/machine interactive analysis of the S192 data was performed by Colwell et al. (ref. 3-3), by using all the 22 data outputs from the 13 channels. The analysis processes were the same as those described in the discussion of the analysis of digitized photographs, except that there was a larger array of data channels (table 3-Vl). An optimization program identified the bands of data most appropriate for classifications of the scene.

The selection process was one of either selecting different bands or discriminating between two data outputs within a discrete data band. The process was one of data source selection in which the 22 data outputs (table 3-VI) were displayed on a television monitor to screen out data outputs that were unusable because of saturation or high systematic noise. Three data outputs from the remaining data of each area of interest were displayed on a color monitor, and the training area within each stratum of interest was identified. A grid coordinate that was displayed simultaneously on the monitor assisted in locating the training sets.

When a complex agricultural scene west of Fresno, California (fig. 3-4), was analyzed, an overall classification accuracy of 81.8 percent and an average performance, by class, of 78.9 percent were achieved. During the analysis of a less complex agricultural stratum (i.e., wheat, barley, and safflower in an annual grass region of the Pacific coast range), a significant increase in accuracies was achieved. For an area in the western San Joaquin Valley, classification accuracies of 100,100, and 98.6 percent were achieved for wheat, barley, and safflower, respectively; and class designations in each crop type were identified.

When analysis for the selection of the optimum bands for crop classification was performed on the 22....

[91] ....features of the S192 data, 4 spectral features were selected. They were (1) the high-density yellow data output, (2) one of the high-density reflectance infrared data outputs (7 or 19), (3) one of the three low-density reflectance-infrared data outputs (8, 9, or 10), and (4) one of the high-density reflectance-infrared data outputs (11 or 20). Studios indicate that when discriminant analysis techniques are applied to more than four bands of data, the cost becomes prohibitive (i.e., the cost of analysis increases approximately by the square of the number of bands used) without significantly improving the classification accuracy. The S192 narrow spectral bands appear to provide more useful information for discriminating among crop typos but require precise timing for highly dynamic crop types such as vegetables.

SOILS

Soil is the basic medium for sustaining the plant, animal, insect, mineral, and hydrological ecosystem on which man depends for his existence. The basic soil units were mapped by using the standard EREP photographic products, and the soil salinity characteristics were discriminated by analysis of the spectral data recorded by the S192 instrument. The quantity of free water in the soil and of crystalline moisture in snow was measured by the use of microwave data,

Mapping by Photointerpretation

The S190A and S190B color and color-infrared photographs were analyzed visually to delineate regional vegetative patterns indicative of the agriculture, the hydrology, the soil resources, and the insect habitats of a citrus-producing area. When the S190B photographs were enlarged to a 1:63 000 scale and projected for a subsequent enlargement to 1:10000, agriculture analysts (Hart et al., ref. 3-1) identified host plants and plant distributions, which are the avenues of citrus-insect migration between Mexico and the Texas citrus belt. Supportive information identifying physical features of the area such as drainage patterns, watercourses, and some soil characteristics was derived from the S190A photographs.

Soil units and drainage patterns were mapped by de Mendonca et al. (ref. 3-2) on Skylab photographic data of Campo Grande, Mato Grosso, Brazil. The S190A color-infrared film (fig. 3-5(a)) was used to separate and map eight types of soil (fig. 3-5(b)). This map derived from the S190A data coincided with the current soil map of the area (fig. 3-5(c)), which was prepared by the use of ground transects. The map derived from the S190A photograph identified the vegetative patterns, the extent of good-quality soil, and the extent of human exploitation of the region. Using the same type of data (fig. 3-6(a)), Bannert (ref. 3-4) mapped pedological units in the Province of Corrientes, Argentina (fig. 3-6(b)), that compared favorably with soil delineations on the soil map of the world (1:5 000 000 scale).

Salinity Discrimination by Analysis of S192 Data

Vegetated areas that exhibited differences in quantity and quality of reflectance as compared with the reflectance from bare soil were analyzed through digital processing (Wiegand, ref. 3-5), and the results were correlated with field measurement of the electrical conductivity of soils of varying salinity. Data were acquired by the 13-band S192 scanner, and analysis was made by linear correlation in each discrete band. As the reflectance ratio between vegetated areas and bare soil increased (i.e., as more reflectance was received from the vegetation), the salinity level of the study area was observed to be lower. The correlations for the six continuous bands (6 to 11) in the wavelength region of 0.68 to 1.75 µm were -0.739, - 0.946, - 0.862, - 0.876, -0.963, and-0.722, respectively.

Vegetative patterns, as extracted from the S19OB color and color-infrared photographs, were used to identify subterranean freshwater levels in the pampa region of Argentina (Bamnert, ref. 3-4). in the area of investigation, ground water occurs at the surface and to a depth of 20 m. On the ground, surveys revealed relationships among the morphology, the depth, and the salinity of ground water. Those features on the Earth's surface can be observed in EREP color-infrared photographs (fig. 4-26(a) in sec. 4). Those areas are delineated on a ground-water-depth map (fig. 4-26(b) in sec. 4); the regions with a ground water table at a depth of less than approximately 5 to 7 m appear light blue and are characterized by high evaporation and increased soil and ground water salinity, which load to unfavorable conditions for certain plant species and to sparse vegetative cover. These regions are favored predominantly by halophytes.

[95] Moisture

Area and regional managers responsible for the allocation and control of surface water have attempted to forecast the available moisture either in the soil or as free crystalline water in the snowpacks, using the aridity and antecedent precipitation index models. The implications are great in terms of use for agricultural endeavors that depend on soil/water resources and flood forecasting for large watersheds. The quantity of runoff produced by a storm depends on the moisture deficiency of the basin at the onset of rain and on characteristics such as rainfall amount, intensity, and duration. The characteristics of moisture-delivering phenomena can be determined from an adequate network of meteorological gages; but the direct determination of moisture conditions throughout a storm is not feasible, and the antecedent precipitation index models can only be used to estimate relative values for soil moisture conditions.

Microwave instruments of the type onboard Skylab are potentially useful for determining moisture conditions of an area. Those on Skylab operated in active and passive modes and responded to an average value of surface or near-surface moisture content. Those instruments were not dependent on visible reflectance. Microwave backscatter (active) and emission (passive) are strongly dependent on the dielectric constant and the moisture content of soil, vegetation, and snow being sensed. The dielectric constant is a measure of the electric charge on a surface within an electric field. For water, at microwave frequencies, the dielectric constant is quite large (as much as 80), whereas that of dry soil is typically less than 5 and that of snow is less than 2. Therefore, the water content of soil or snow can greatly affect the dielectric constant.

Analysis of the response of the microwave instruments (Eagleman et al., ref. 3-6) indicates that passive radiometers, particularly the L-Band Radiometer (S194), were most sensitive to the percentage of moisture in either the surface layers of the soil or snow cover. The first step of the analysis consisted of obtaining, by direct measurement, detailed ground soil moisture information at the time the EREP instruments were collecting data. The resulting soil moisture maps showed the percentage of moisture available in each 2.5-cm increment to a total depth of 15 cm. The correlations of the radiation received by the radiometer with the moisture content of the various layers beneath the surface were computed for each of the 2.5-cm layers to evaluate the effective depth from which the L-band signal originated. In one case, the antenna temperatures correlated with the moisture content in the top 5 cm; but for the four other cases, the moisture within the top 2.5 cm provided the best correlation. This result agrees with the theoretical calculations. When data for the five different passes were combined, the correlation between the S194 radiometric temperature and the soil moisture content remained high, with a value of-0.96.

The results of the radiometer component of the Microwave Radiometer/Scatterometer and Altimeter (S193) were less definitive than for the S194 instrument. For the same test site and for the same area, when several of the S193 footprints were averaged to obtain the same area covered by the S194, the correlation of antenna temperatures with moisture content was-0.988 for the S193 as compared with-0.996 for the S194. The particular pass was across Texas on June 5, 1973. When the S193 was used as a scatterometer, the response to soil moisture, which is not as good as that for the radiometers, resulted in a correlation of 0.75.

The analysis of S190A photographs and of imagery generated from the S192 has shown that, for some areas, the infrared bands of these sensors can be used to identify significant differences in soil moisture. However, direct measurements of subsurface soil moisture content by optical and multispectral scanner data are difficult because the presence of a vegetative cover tends to shield soil moisture from detection by these methods. (See methodology by J. Colwell, sec. 6.) Although optical or multispectral scanners cannot gather quantitative soil moisture information on a practical basis, they can be used very effectively with microwave data to increase the accuracy of soil moisture measurements by providing information about vegetation type and density. Analysis of multispectral scanner thermal-band data can aid in collecting and assembling surface-temperature information.

When the microwave antenna temperatures were analyzed for surface emission, which is a function of the percent of soil moisture by weight, it was concluded (Eagleman et al., ref. 3-6) that there is a very high correlation between antenna temperature and soil moisture....

[98] ....for values of O to 35 percent. As the moisture content of the soil increased above approximately 35 percent, larger dispersions in the data occurred. It should be noted that this correlation is inverse (i.e., the microwave temperature increases as the soil moisture percentage decreases). Analysis of the microwave temperatures from S193 and S194 instruments recorded over snow-covered terrain in the Great Plains has shown that the relationship between brightness temperatures and the water content of the surface of the snow did not resemble that observed in soils. Over the snow-covered terrain, the temperatures were proportional to the water content of the surface of the snow, and there was a high degree of correlation only when the free-water content was between O and 2.5 percent. The results of this study, based on limited microwave data, indicate that it is possible to measure moisture content with a high degree of precision over relatively large areas having low quantities of moisture. Better correlations were obtained with the S194 instrument than with the S193.

RANGE

In most of the nonagricultural test sites that were studied by EREP investigators, rangelands were intermixed with forest lands. In fact, this same intermixing occurs quite commonly throughout the wild-land areas of the world. For this reason, timber resources are referred to in this subsection and, conversely, range resources are referred to in the subsection on timber resources.

Range and Wild-Land Classification Systems

A significant contribution of the rangeland investigations to the EREP program was the adaptation of existing vegetation classification systems for use with Skylab data. Classification systems are probably the most important components in the transfer of remote-sensing information, such as that provided by Skylab, to resource managers, planners, and government agencies. Among the various hierarchical classification systems, ECOCLASS ² is frequently used for ecosystem classification and for improved multiple-use planning and management of forest and range resources. The ECOCLASS system links vegetation, land, and aquatic systems with the description and classification of relatively permanent ecosystems. The system defines five categories, proceeding from the most general to the most specific, as follows.

V. Formation-The most general class of vegetation, characterized by general appearance: grassland, coniferous forest, deciduous forest, etc. The basis for this category is continental in scope (i.e., all the United States) and is controlled by continental climatic differences.

 

IV. Region-Subdivisions of the formation, associated regionally and therefore determined by subclimates within continental climates: montane grassland, temperate mesophytic coniferous forest, alpine grassland, etc.

 

III. Series-A group of vegetation systems in the region category, with a common dominant climax species: ponderosa pine forest, fescue grassland, herbaceous meadow, etc.

 

II. Habitat type-Units in a series, each with relatively pure internal biotic and abiotic structure: ponderosa pine/Arizona fescue habitat type, Arizona fescue/mountain muhly habitat type, etc. These are the elemental units of the classification scheme on which primary management is based. These units are frequently related to climax vegetation or to vegetation held in a relatively stable state of high succession by proper management.

 

I. Community type-A system that appears relatively stable under management and may be equivalent to the habitat type. The biotic components usually are dissimilar, but abiotic components are analogous to habitat type.

Of the five ecological levels of classification most useful for evaluating remote-sensing data, habitat type represents the level of information required by vegetation and land managers for making resource management decisions.

Several Skylab investigators evaluated EREP data for range resources inventory and analysis applications (Poulton and Welch, ref. 3-7; Hoffer, ref. 3-8; and Aldrich et al., ref. 3-9). They found that S19OA color-infrared photographs were consistently most useful for interpreting a wide range of natural vegetation types in Nevada and Colorado/New Mexico test sites. The [99] S19OA color-infrared film was significantly better for identifying vegetation complexes than any other film tested, although S19OB color film was nearly as good for other aspects.

A rigorous classification of range plant communities performed by the U.S. Forest Service range scientists was based on the ECOCLASS system. identification at two levels, three regional and eight series, was attempted (Aldrich et al., ref. 3-9). Photographs from the S19OA multiband camera and the S19OB terrain mapping camera (June and August 1973), photographs from high-altitude aircraft (June and August 1973), and large-scale photographs from aircraft were used in the tests. Both visual and microdensitometer techniques were tested.

Training and test sample cells were selected for interpretation on a restricted random basis. To be selected, a specific plant community had to occupy an area at least 500 m square. A 10-percent sample was selected at random from each plant community class for field validation. Overlays of sample cell locations and plant community keys were used to aid interpretation. Procedures were also developed to map cultural features from the EREP photographs.

Range ecologists, interpreting S190A and S190B photographs, classified grassland and conifer region classes with a mean accuracy of 98 percent or greater on both Skylab and support-aircraft photographs, regardless of date or film type. However, tree series classification was inconsistent. Aspen was classified with 80 percent accuracy on August color-infrared 1:10 000-scale aircraft photographs, but this accuracy was not obtained on EREP photographs. Accuracies for coniferous classification at the series level were dependent on date, film type, and scale. For instance, the Douglas fir class was accurately classified on June color-infrared EREP photographs but not on aircraft photographs. Lodgepole pine and ponderosa pine classes were interpreted accurately on EREP color photographs for June but not on aircraft photographs. Aircraft color and color-infrared medium-scale photographs made in June were best for interpreting the spruce/fir class. The greater accuracies at smaller scales were probably due to the mixing of tree species into homogeneous units with a dominant species signature and a lower resolution.

In the grassland series, shortgrass was classified with 95 percent or greater accuracy on both Skylab and aircraft photographs, regardless of date or film type. Wet meadows were classified with greater than 90 percent accuracy on both June and August aircraft photographs, regardless of film type or scale. The classification of wet meadows was also acceptable on both color and color-infrared EREP photographs taken in August. Mountain bunchgrass was not accurately classified on S190A and S190B photographs; but on the August aircraft photographs, the classification was acceptable, regardless of film type or scale. Topographic slope and aspect, mountain shadows, ecotones (interfaces between two ecosystems), season, and class mixing affected the classification of plant communities.

Experimental results in the EREP rangeland studies (Aldrich et al., ref. 3-9) were much affected by seasonal timing of S190A and S190B photograph acquisition. Although quantitative results were data dependent, it was emphatically stated that the best time for imaging natural vegetation with color-infrared film is when the vegetation types are approaching the mature growth period. The EREP investigations indicate that the peak growing season (high phenological activity) is the poorest time of year for photographing natural vegetation from space. Multidate photographs, therefore, provided the only means for consistent identification of some vegetation complexes.

Microdensitometric point sampling of region-level conifer, deciduous (aspen), and grassland classes showed significant differences in mean optical densities at the 95-percent-probability level. However, the deciduous class could be separated from the other classes with significant differences only on color film Ponderosa pine was the only series-level conifer that showed a significant difference in mean optical density from the other three conifers, regardless of date or film type. Spruce/fir and lodgepole pine were not separable at any date or on any scale or film type. The mean optical density for aspen was significantly different from that for the conifer classes, but the differences were dependent on date, scale, and film type. Douglas fir was separable from the other three conifers on both the June color-infrared and the August color S190A EREP photographs. Grassland classifications at the series level varied in acceptability. However, shortgrass, mountain bunchgrass, and wet meadows did have mean optical density differences that were significant on August S190A color photographs. The optical density was more dependent on community mixing than on the growth stage of the plants at the time (season).

Both EREP and aircraft photographs were useful for mapping the areal extent of conifer and grassland.

[100] These two classes were usually mapped with greater than 90 percent accuracy. The deciduous class could not be mapped with acceptable accuracy at either the region or the series level. Series-level conifer and grassland could be mapped with acceptable accuracy only if class complexes were formed. Class complexes were ponderosa pine and Douglas fir, lodgepole pine and spruce fir, shortgrass and mountain bunchgrass, and wet meadow.

Many disturbances to natural rangeland vegetation communities were recorded. For example, paved and gravel roads, utility corridors constructed within the last 10 years, larger mining excavations, and clusters of buildings could be mapped on S190A and S190B photographic enlargements. However, 1:100000-scale aircraft photographs were needed to map dirt roads, minor soil excavations, utility corridors more than 10 years old, and individual buildings. Foliar cover and plant litter measured on large-scale color-infrared photographs of nondiverse grasslands were related to ground measurements with a correlation coefficient of 0.75. This coefficient is considered acceptable for range surveys. The relationship for foliar cover of shrubs was acceptable only on diverse grasslands.

The use of EREP photographs for vegetation delineation was illustrated by a legend system (Poulton and Welch, ref. 3-7) in which a multidigital fraction was used to depict the vegetative associations (numerator) and landforms (denominator). This system is especially suited to multistage remote-sensing applications and is in decimal form for computer compatibility. The numerator is a three-digit number with decimal components identifying the vegetation analog or land use conditions. The denominator uses a three-component decimal system for landscape characterization. The components are macrorelief landform, and microrelief. Macrorelief refers to the largest category of classification of major relief change within the landscape system being studied, with the landform feature addressing the geomorphological categories as fluvials or deserts and the microrelief characters defining the local contours.

An arid region of the Southwestern United States (fig. 3-7(a)) was classified, and a photograph was annotated (fig. 3-7(b)) using this hierarchical classification system. The numerical classifiers used for this illustration are listed in appendix A of reference 3-7.

Evaluation of Film for Classification

Although the legend system is difficult for resource managers to use, it can be applied to landscape boundary determination on a Skylab-quality satellite photo graph, as shown in figure 3-8(a) (an S190A color photo graph) and in figure 3-8(b) (an S190A color-infrared image of the Uncompahgre area). The photographs in figure 3-9 show separation between landscape types as viewed from low-altitude aircraft. High-definition S190B color film was preferred for mapping vegetation boundaries because it has better spatial resolution (fig 3-8(c)). Ranking beneath S190B color for vegetation boundary delineation were, in descending order, S190A color (fig. 3-8(a)), S190A color-infrared (fig. 3-8(b)) and S192 (fig. 3-8(d)) imagery. When cost was considered, high-definition S190B color film (fig. 3-8(c)) was considered best for delineating vegetation boundaries. In most cases, Skylab stereoscopic data provided the best identification of vegetation complexes and delineation of vegetation boundaries, particularly in areas where changes in relief were related to changes in vegetation types (a common occurrence in wild-land vegetation communities).

FORESTS

Skylab provided the first opportunity for foresters to test the concept of a manned space laboratory for resource surveys. Thus, remote-sensing investigations in which widely separated forest and wild-land sites were studied were conducted to investigate the applicability of EREP data to forest resources inventory and analysis problems.

Interpretation techniques and instruments used by the investigation teams to analyze EREP data varied from one application to another. When the EREP photographic products were studied, a zoom transfer scope was used for mapping and dual-image correlations. In forest-stress impact analysis, a stereomicroscope was used to test, monocularly and stereoscopically, a wide range of image magnifications. Other forms of manual photointerpretation included the use of rear projection viewers that provided image magnifications as large as 29.5x. On one forest inventory [101] task, a scanning stereoscope and a lamp magnifier were used for conventional photointerpretation.

Two types of digital methods were used to analyze EREP photoproducts. Microdensitometers were used to scan and digitize S190A and S190B photographs, and television scanning systems were used to analyze digitized photographs that were displayed at different gray-scale levels. With digital tapes from the S192 Multispectral Scanner, investigators used several types of computer hardware systems and analysis algorithms.

Various forest resource applications problems were addressed with the aid of EREP products. Forest classification, a specialized area of land use classification and a function of forest inventory, was the cornerstone activity of all forestry investigations. The specific tasks related to the data needs of forest resource managers involved determining forest timber volume, stand vigor (stress), and ownership boundaries.

Classification

The EREP investigators indicated that, with S190B color photographs, Level I forest and nonforest land areas can be classified with 90 to 95 percent accuracy. (See table 2-1 for Level I and Level II classification definitions.) The accuracy of classifying Level II forest and nonforest classes varied from fair to poor. Hardwood and pine (fig. 3-10) could be separated with a confidence level of 95 percent and an accuracy of 90 and 70 percent, respectively (ref. 3-9).

The computer map shown in figure 3-ll(a) provides a visual comparison with the S190B photograph (fig. 3-ll(b)). A point-by-point comparison was made between points located on a ground-truth map and the same points on the computer map. investigators found that 93 percent of the forest points were correctly classified as printed on the computer map. The pine points were classified with an accuracy of 83 percent. Points falling in hardwood were classified with 74 percent accuracy, and nonforest points were classified with 85 percent accuracy. The results of the Aldrich et al. research (ref. 3-9) revealed that forest area could be classified and, therefore, stratified on high-resolution S190B photographs with an accuracy of approximately 96 percent. Thus, with S190B photographs, forest area can be mapped and an estimate of the land use areas can be made within limits that can be accomplished using aerial photographs.

The results of plant community classification tests indicate that both visual and microdensitometric techniques can be used to separate deciduous, coniferous, and grassland classes to the region level in the ECOCLASS hierarchical classification system. By visual analysis techniques, the classification accuracy was more than 90 percent on S190B photographs. However, the classification of deciduous forest was dependent on the date, the film type, and the scale of the photographs. By means of microdensitometry analysis of S190A photographs, an average accuracy of more than 80 percent was achieved, although this accuracy is subject to change depending on the film type and the season of data acquisition. There was no consistency in classifying tree categories at the series level by visual photointerpretation. Conifers were classified most accurately (80 percent) on S190A photographs, whereas, under certain conditions, grassland plant communities were classified at accuracies greater than 80 percent. The results of microdensitometric techniques were variable and highly dependent on the photograph date, film type, and scale.

The analysis of S192 data (fig. 3-12) involved classification of major cover types (corresponding to the Level II land use classes in table 2-1) and forest cover types. The results (Hoffer, ref. 3-8) indicated that the major cover types could be mapped with approximately 85 percent accuracy and that the forest cover types could be mapped with approximately 71 percent accuracy When areas of major cover types obtained from photointerpretation of S190A or S190B imagery were compared to the area summary based on two computer-aided analysis techniques, ECHO and per-point classification, using S192 data, a correlation coefficient of 0.929 resulted.

Comparison of the results obtained with S192 data and Landsat multispectral scanner data indicated that the improved spectral resolution of the S192 narrow-band data enabled a higher classification accuracy for forest cover types, although the classification performance for major cover types was not significantly different. The investigators believe that, had the S192 performance been optimal throughout the mission,....

....the classification results would have improved significantly. Also, the results indicated that the increased spectral range of the S192 system enables selection of a better combination of four wavelength bands for computer analysis than is present in the Landsat data. Specific results of the wavelength band evaluation study indicated that classification performance was not significantly improved when more than four wavelength bands were used in the computer analysis. Analysis of S192 data indicated that the near-infrared part of the spectrum, especially the 1.09- to 1.19-µm wavelength band, was of particular value for vegetation mapping (Hoffer, ref. 3-8), with additional wavelength bands in the visible (0.52 to 0.56 µm), near infrared (0.78 to 0.88 µm), and middle infrared (1.55 to 1.75 µm) also shown to be of significant importance. However, the combination of four wavelength bands that were of most value for classifying various cover types varied considerably.

The S192 data collected on August 5,1973, were processed to produce a classification map of part of the Gratiot-Saginaw State game area in south-central Michigan (fig. 3-13). A preliminary 10-category classification map was prepared for an area consisting of diverse vegetation cover types, including hardwood....

.....and conifer forests, wetlands, brushland, and herbaceous vegetation. Field checks indicated an overall pixel recognition accuracy of 54 percent, although accuracy of the more poorly classified categories ranged from 25 to 52 percent. After the categories were limited to five, the overall classification accuracy increased from 54 to 72 percent. When the output statistics were reduced for a square-mile grid, the accuracy increased to 82 percent, as the result of compensating for errors of omission and commission (Sattinger et al., ref. 3-10). The signal-to-noise ratio of each S192 band presented many problems. Sattinger et al. (ref. 3-10) identified the remaining usable bands in order of preference for scene classification: (1) 0.78 to 0.88 µm, (2) 1.55 to 1.75 µm, (3) 0.98 to 1.08 µm, (4) 0.68 to 0.76 µm, (5) 0.52 to 0.56 µm, and (6) 0.62 to 0.67 µm.

Type Determination

Density-slicing analyses of S19OA and S19OB color and color-infrared photographs, converted to gray-scale levels, have demonstrated the feasibility of using such data for differentiating major timber classes (including pines, hardwoods, mixed, cut, and brushland), provided such analyses were made at scales of 1:24 000 or larger (Baldridge et al., ref. 3-11). Detailed machine analyses, reinforced with data from knowledgeable field personnel, indicate that sufficient spectral differences exist to make automatic (computerized) machine separation of pine and hardwood stands possible. Further differentiation of mixed hardwood and softwood types in western Ohio study sites, which have small, extensively mixed forest stands, was not possible with the use of EREP....

....photographs. Sample results indicate that density slicing of EREP color and color-infrared film may be used to classify forest-stand maturity in Ohio into the following categories of commercial interest: mature timber, intermediate and pole timber, seedling/sapling stands, brushland, and clear cut areas.

Forest investigators in Australia were able to correlate the occurrence of different forest types with variations in color on the S19OA color-infrared photographs (Lambert et al., ref. 3-12). They also separated native forest areas into tree/crown density (crown closure) classes and delineated the major forest species, even though the vegetation boundaries were less sharp than on midsummer Landsat-l imagery. The black-and-white photographs were of little value for vegetation classification of forest lands. The investigators recommend the use of midsummer S190B color-infrared photographs for forest mapping.

Inventory

The EREP S190A and S190B photographs were successfully used to inventory and map a large woodland area in Ohio, an accomplishment demonstrating the minimal requirement for supportive ground and aircraft data (ref. 3-11). Both conventional photointerpretation and machine-assisted procedures were used effectively. It was learned that, for the determination of forest cover by counties, Skylab photographs were more accurate and economical than conventional surveys involving the use of aerial photoplot techniques.

However, this assessment is not meant to imply that cost effectiveness was maintained in providing all the inventory and mapping information generally required for forest surveys.

Volume Determination

Scientists at the University of California at Berkeley (Colwell and Benson, ref. 3-13) used S190A photographs as the first stage in a multistage sampling design to determine the abilities of human photointerpreters to....

....distinguish timber-volume classes. Gross timber volume was the variable estimated in the sampling design, based on random sampling at each of three stages. Sampling units were randomly selected at each stage because it was not known how well human interpreters could differentiate between timber and nontimber classes on S190A photographs in the first stage; therefore, the classes could be established quantitatively, as desired. After all combinations of sampling procedures had been tested, it was concluded that S19OA color-infrared photographs did not have sufficient resolution to provide timber-volume information for inventory purposes. When the same analysis procedures were applied to high-flight-aircraft data, the investigator was able to estimate timber volume in the test area.

Stress Detection

A comprehensive evaluation of EREP data showed that mountain pine beetle infestations in the Black Hills National Forest of western South Dakota could not be detected on color-combined multiband, black-and-white, normal color, or color-infrared photographs from the S190A multiband camera system (Aldrich et al., ref. 3-9). All positive identifications were made using S190B color photographs. The infestations detected were always more than 26 m in the longest dimension. On one site, only infestations of more than 50 m could be detected. It was concluded that poor detection was due to timing of the imagery (June) and low Sun angle. Optimum viewing was achieved with a microscopic viewer on a good-quality light table at a 1:75 000 scale. Stereoscopic viewing resulted in fewer errors of commission. Because of poor quality and misregistration between bands, infestations could not be detected by computer processing of the S192 data. Foresters in Ohio reported that S190A photographs, regardless of film emulsion, were inadequate to detect tree stress and damage conditions in northeastern Ohio (Baldridge et al., ref. 3-11).

[114] Image Annotation and Overlays

A precision image annotation system was developed as a critical part of a forest investigation in northern California (Langley and Van Roessel, ref. 3-14). The technique extended the existing capability of annotating the corners of primary sample units on aerial photographs, Landsat imagery, and the Landsat multispectral scanner data to S190A and S19OB photographs and to gray-scale maps made from S192 data. image overlays were produced by computer methods to provide visual identification of the sample-unit corners on each of the image types. Concurrently, the coordinates of the points in the computer tape system were determined. Hence, specific sample units were addressable over spectral bands and time for interpretation purposes. The annotation system developed is capable of correcting for image distortions caused by Earth curvature, terrain relief and distortions inherent in the imaging system.

One of the significant results achieved with the annotation system was the determination that the root-mean-square error of point location of S190A imagery was 100 m and 90 m in the X and Y directions, respectively. It was also learned that the potential gains in sampling precision attributable to space-derived imagery ranged from 4.9 to 43.3 percent depending on the image type, interpretation method, time of year, and sampling method applied. These results can be compared with the 55.1-percent gain achieved by human interpretation methods applied to high-altitude-aircraft photographs.

A significant "first" by Hoffer (ref. 3-8) was to analyze Skylab, Landsat, and topographic data all in a common format. The Skylab data were corrected geometrically, and the Landsat data were corrected to a corresponding scale for analysis. Topographic map data, which included aspect, slope, and elevations, were digitized and corrected to the appropriate scale for analysis with the Skylab and Landsat data. This process required the development of new techniques, including the production of a digital data tape containing 13 S192 wavelength bands, 4 Landsat bands, and 3 bands containing topographic data. These different data sets were all geometrically corrected and registered to a 1:24 000 scale data base. Digital display images in either a gray tone or color format and line-printer outputs were generated from topographic data; by this means, elevation, slope, or aspect could be indicated with different gray tones or colors on the digital display imagery or with different symbols on the line-printer output.

Area Determination

Random and systematic sampling designs were tested for measuring forest area proportions by using a digitized ground-truth map for one county (Aldrich et al., ref.3-9). The variance in forest area proportions was always less with the use of systematic sampling, which stratified the area into forest and nonforest and sampled the forested area before analysis. Systematic sampling, with the use of digitized S190B optical densities and linear discriminant functions for postsampling stratification, reduced variance in forest area proportions at the lower sampling rates. (At sampling fractions of more than 0.0004, the advantage decreased rapidly. This sampling fraction represents the percentage of the stratified forest that was sampled on a grid coordinate basis.) In addition, Hoffer (ref. 3-8) substantiated that reliable areal estimates can be obtained using computer-aided analyses of satellite data even in areas of rugged, mountainous terrain.

Recreational Potential

Sattinger et al. (ref. 3-10) provided a qualitative evaluation of both S190A and S190B photographs and concluded that S190A has limited application for recreational land use analysis, but that S190B, with a resolution approaching that obtainable from high-altitude aircraft, is useful for many land analysis applications. The investigators have stated that S190B photographs contain sufficient detail to map Level I and Level 11 categories of land use and land cover. Applications included mapping existing recreational facilities, identifying open spaces that might be suitable as recreational land, and site planning of geographically extensive areas, such as river basins.

Temporal variations

The evaluation of temporal S190A and S190B photographs showed the importance of season in relation to the analysis of multispectral scanner data (Hoffer, ref. 3-8). The photointerpretation results indicated that, [115] because of differences in vegetative condition, the Skylab 2 data obtained in June over Colorado were better for vegetation mapping than those obtained in August.

Langley and Vam Roessel's results (ref. 3-14) confirmed that seasonal variations, as recorded on film, were significant for forest interpretations. In their study conducted in California, the S19OA photographs obtained during Skylab 3 (September) yielded higher interpretation accuracy than those obtained in June; however, S19OA color-infrared composites from both time periods yielded the highest results of all S19OA products analyzed. No S19OB photographs were available for this study; therefore, no S19OB temporal combinations were possible.

Although mapping of all the forest and rangeland (6070 x 10⁹ m² (1.5 X 10⁹ acres)) in the United States at frequent intervals is desirable, limitations in computers and computer storage make detailed and repeated inventories unfeasible at the present time. instead, it is much more reasonable to think of sampling applications. For example, it was demonstrated (Aldrich et al., ref. 3-9) that a systematic sample grid can be overlaid on digitized land use map data by computer to estimate forest and nonforest land in an entire county. (The variance was always lower than that resulting from simple random sampling.) Using Skylab color film to classify forest and nonforest land in an entire county resulted in an accuracy of 80 percent for forest land, with a 30-percent commission error. Types of forest and other land covers may be estimated by sampling digitized data from future satellite coverage if color-infrared film is made available and if a classification system based on the existing land cover rather than on intended use is designed.

The primary advantage of Skylab S19OA and S19OB photographs in forest resource surveys is the broad area coverage within a single frame. In the 4-county experiment, 183 aerial photographs (1:20 000) were required to cover an estimated 80 percent of the total area. This was single photographic coverage without the advantage of stereoscopic overlap. A single S19OB photograph, however, will cover these four counties and from two to four additional counties as well. Complete county coverage offers better distribution of photograph samples and reduces data handling and photo acquisition costs, on the assumption that only printing and processing costs are involved.

If all other survey costs are considered equal, the costs of Level I and Level II land use and forest stratification would be 49 percent lower using S19OB photographs than on conventional 1:20 000 scale aerial black-and-white photographs. The major difference between the two methods is the cost of the photographs. Because of the small scale and the use of normal color film, more time was required to make interpretative decisions on S19OB photographs. However, if high-resolution color-infrared photographs were available on a regular, recurring basis, the advantages of current information would far outweigh the disadvantage of any additional interpretation time.

SUMMARY

Several scientific investigations in the disciplines of agriculture, range management, and forestry analyzed the EREP data to contrive techniques and procedures for extracting the required resource information. The imagery was acquired in a multispectral format, analyzed by visual interpretation, and digitized for computer analysis. The narrow-band spectral data as acquired by the S192 instrument were analyzed by computer processing to identify the optimum spectral bands and combinations of these bands that would provide maximum amounts of resource information at a minimum cost. Microwave data recorded in an electronic format provided an insight into the mapping and monitoring of available moisture for natural resource consumption. Evaluations of the utility and recommendation of data formats and required observation frequency for discrete resource monitoring or measurements using EREP-quality data were also made.

For identifying and measuring field crops (i.e., crops grown in larger tracts (32 to 65 hm² (80 to 160 acres)) and in the categories of cereal or large-area crops, as differentiated from vegetables), it was found that the multispectral photographs were appropriate for inventorying within an accuracy range of 82 to 98 percent. To identify large-area crops using single-date data, it was determined that the data would have to be acquired precisely when the crops were maturing; but when temporal or multidate data were used, the acquisition time was not as critical and data could be acquired throughout the development phase and before maturity. When data acquired from these same types of [116] crops were analyzed, the results were equally as good with either digitized photographs or electronically recorded S192 data. Large-area crops generally follow a uniform planting and harvesting calendar for a given geographical region and meteorological environment and are compatible with the identifiable types of required data and the appropriate times for acquisition of the data.

When intensively cultivated crops (vegetables) were studied, it was concluded that the timing of data acquisition was extremely critical and that data should be acquired at discrete stages of maturity. Acquisition of these temporal data constituted an additional problem because vegetable production areas frequently have one type of crop in two adjacent fields at different stages of maturity and thus require more frequent observation intervals.

Generally, field sizes smaller than 2 hm² (5 acres) were difficult to discriminate with the resolution of the EREP sensors. As the field sizes increased, the identification and areal measurement accuracies also increased.

With use of the full range of the multispectral photographic system, four bands (green, red, and two infrared) were the most useful for inventorying and monitoring vegetative resources, although only two bands were analyzed for vegetative resources. The natural color and color-infrared bands were the most frequently used. Analysis of the narrow 13-band electronic data indicated that a yellow band and 3 infrared bands provided the greatest amount of information.

The photographic system, the 13-band S192, and the microwave systems were used to map various soil parameters. The areal extent of soil units was mapped using the color and color-infrared photographs. When the soil was visible, color was the indicator of soil units; and when the soil supported vegetation, the vegetation boundaries were considered synonymous with soil units and used for mapping.

Soil salinity was mapped over a limited test site by analyzing S192 data. The indicator was the quality of vegetation as correlated with the electrical conductivity of the soil. This technique is promising for saline soil mapping but will require additional study for refinement.

The available soil moisture in the top 15 cm for plant consumption was mapped from S193 and S194 microwave data. Correlations of soil moisture between O and 35 percent by weight were very good; but for larger percentages of moisture, the instrument responses were saturated. Similar results were obtained when the water equivalent of snow covering the Great Plains area was mapped.

When mapping wild-land resources, which include rangeland and forest environments, accuracies for Level I and Level II were 90 to 95 percent and 70 to 90 percent, respectively. Conventional photointerpretation techniques were used, with the preference of film types being high-definition color from the S190B because of its better spatial resolution, followed by S190A color and S190A color infrared, respectively. Level II land use classes, as identified in table 2-1, were identified by analysis of S192 data at an accuracy of 85 percent.

[117] Investigators achieved limited success in detecting insect damage by analysis of EREP data. They indicated that the residual effect, or change in vegetative state, was sufficiently limited to prevent detection with the resolution-cell capability of the sensors and that the data were not acquired at the optimum biological periods of insect activities. In the case of the pine bark beetle damage to conifer forests and mealybug infestation on citrus, the period of maximum biological activity produces the maximum visual change in the character of the vegetation. The general conclusion of most of the investigators was that the EREP-quality data are acceptable for Level I and II monitoring but require acquisition at optimum times within the developmental stages of the vegetation and should be analyzed by personnel thoroughly familiar with the resource in question.

REFERENCES

3-1. Hart, W. G.; Ingle, S. J.; and Davis, M. R: A Study of the Early Detection of insect infestations and Density Distribution of Host Plants. NASA CR-144483, 1975.

3-2. De Mendonca, F.; Machado, J. B.; et al.: Collection of Relevant Results Obtained With the Skylab images. NASA CR-147502, 1975.

3-3. Colwell, Robert N.; Benson, Andrew S. et al.: Agriculture Interpretation Technique Development. NASA CR-144486, 1975.

3-4. Bannert, D.: Hydrological investigations in the Pampa of Argentina. NASA CR-144488, 1975.

3-5. Wiegand, Craig L.: Soil Salinity Detection. NASA CR-144403, 1975.

3-6. Eagleman, J. R.; Lin, W.; et al.: Detection of Soil Moisture and Snow Characteristics From Skylab. NASA CR-144485, 1975.

3-7. Poulton, C. E.; and Welch, R. 1.: Plan for the Uniform Mapping of Earth Resources and Environmental Complexes From Skylab imagery. NASA CR-144484, 1975.

3-8. 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

3-9. Aldrich, Robert C.; Dana, Robert W.; et al.: Evaluation of Skylab (EREP) Data for Forest and Rangeland Surveys. NASA CR-147440, 1975.

3-10. Sattinger, 1. J.; Sadowski, F. G.; and Roller, N. E. G.: Analysis of Recreational Land Using Skylab Data. NASA CR-144471, 1973.

3-11. 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.

3-12. Lambert, B. P.; Benson, C. J.; et al.: A Study of the Usefulness of Skylab EREP Data for Earth Resources Studies in Australia. NASA CR-144493, 1975.

3-13. Colwell, Robert N.; and Benson, Andrew S.: Skylab Data as an Aid to Resource Management in Northern California. NASA CR-144487, 1975.

3-14. Langley, Philip G.; and Van Roessel, Jan: The Usefulness of Skylab/EREP S190 and S192 imagery in Multistage Forest Surveys. NASA CR-147439, 1976.

² Robert D Pfister and John c Corliss, "ECOCLASS-A Method for Classifying Ecosystems (report on file at Forestry Sciences Laboratory, Intermountain Forest and Range Experiment Station, Missoula, Mont.), 1973.

---