Automatic Soil Identification from Remote Sensing Data
Total Page:16
File Type:pdf, Size:1020Kb
DR. KAM W. WONG University of-Illinois at Urbana-C hampaign Urbana, IL 61801 DR. T. H. THORNBUR Las Vegas, NV 89109 M. A. KHOURY University of Illinois at Urbana-Champaign Urbana, IL 61801 Automatic Soil Identification from Remote Sensing Data Small differences in soil types can be distinguished by the use of quantitative terrain factors which are computed from digital terrain data. INTRODUCTION pends primarily on the ability of the in HE CHARACTERISTICS of a soil and by terpreter, and (3) they are relatively time T extension its engineering properties are consuming. With the advance of electronic chiefly a function of the parent material, to computers and the corresponding develop pography, vegetation, climate, and time. ments in automatic topographic mapping Thus, having a thorough understanding of and remote sensing, the automation of soil geomorphology and the natural forces ofero classification has become an interesting as sion and weathering such as wind, ice, and well as promising possibility. water, a soil scientist experienced in air This paper describes a systems approach photo interpretation techniques can identify for the automatic identification of soils by soil characteristics from the land forms, ero- the combined application of remote sensing ABSTRACT: A reliable method ofautomatic soil identification can be developed by the combined application ofremote sensing and digital terrain data. Research results have demonstrated that small differ ences in soil types can be distinguished by the use of quantitative terrain factors which are computed from digital terrain data. Con tinuing research effort is dil-ected towards the improvement of the terrain factors, the development ofstatistical prediction techniques, and testing the effectiveness ofthese factors in the identification of soils. sion patterns, vegetative cover, land use, and and digital terrain data. The major problem the tonal distribution in the air photos. The in the development of such a system is the accuracy of the interpretation will depend successful development of a set of quantita chiefly on the technical skill of the interpre tive factors which can adequately describe ter and his unique and instinctive ability to the geometric characteristics of the topog detect, correlate, and deduce from the many raphy. A research program directed towards minute hints present in air photos. the development of such 11 set of factors for Conventional photo interpretation the purpose ofsoil identification is presently techniques have three major shortcomings: in progress at the University of Illinois. Pre (1) they require highly trained personnel, liminary results are very encouraging and (2) the accuracy of soil classification de- will also be reported in this paper. 73 PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING, Vol. 43, No.1, January 1977, pp. 73-80. 74 PHOTOGRAMMETRIC E GINEERING & REMOTE SENSING 1977 M ULTISPECTRAL ANALYSIS ic and reliable soil identification technique cannot be based entirely on multispectral Several investigators have reported in the analysis. The accuracy of soil identification literature attempts to identify soil types di using either spectral analysis or microwave rectly from their multispectral signatures. Anuta et al. (1971) reported an attempt to radiation techniques has been degraded classify two types of fine sands, four loamy greatly by the data noise caused by vegeta tive cover, atmospheric condition, instability soils, and one clayey soil in the area of El Centro in the Imperial Valley, California. Al of the sensors, and sun angles. Even differ ences in tillage practices on agricultural though many bare soil fields were correctly identified, the computer classification was fields alter greatly the spectral radiance of the surface soil and confound identification found to be incorrect for areas covered by by spectral analysis (Kristof and Zachary, vegetation. Cihlar and Protz (1972) achieved 1971). Another reason for the failure of an identification accuracy of39.7 percent for spectral analysis techniques is that the sur ideal training sets and reported that the ac face topography, which is a major soil form curacy decreased gradually if these sets were modified to suit practical require ing factor, is not used as a characteristic fac ments. Kristof and Zachary (1974) reported tor for soils identification. similarly limited accuracy in an attempt to classify soils in Indiana. In a slightly differ A SYSTEMS ApPROACH ent approach, Piech and Walker (1974) used Figure 1 illustrates the basic concept of the red-to-blue reflectance ratio to identify using the systems approach for the automatic the effects of moisture and texture on the identification of soils. The approach makes tone of the soil. Although the tonal differ use ofmodern remote sensing and automatic ence between a dark and a light color soil mapping techniques to identify all the soil can be attributed to either moisture or tex forming factors (such as vegetation, ground tural difference from the ratio test, the rela temperature, moisture, drainage, topog tionship between the ratio values and the raphy) as well as ground signatures of soil degree of moisture and textural change can types (such as land use, texture, and spectral not be presently quantified. characteristics). These data can then be The principle ofthe multispectral analysis supplemented by environmental informa approach differs significantly from that ofthe tion such as climate and geological history of traditional method of photo interpretation. the area to predict the soil types. While a photo interpreter attempts to iden Digital models of the relief of the terrain tify from the air photos the factors which can now be obtained as direct output from cause the formation of a particular type of photogrammetric mapping instruments such soil, the multispectral approach attempts to as the Automated Analytical Stereoplotter identify directly the surface characteristics (AS-H), the Universal Automatic Map Com of a particular type of soil. pilation Equipment (UNAMACE), and It is generally recognized that an automat- stereoplotters. The first two types of instru- Data Source Data Vegetation Landuse Remote Sens i n9 Mo; sture Color Correlation Ground temperature Classification ==:;;> SOILS Ora inage Identification Automatic Cartography Digital terra; n C1 imate I nfonna ti on Time Data Bank Geological History FIG. 1. A systems approach for soil identification. AUTOMATIC SOIL IDENTIFICATION FROM REMOTE SENSING DATA 75 ments are being used for routine automatic direction changes) to successfully delineate map compilation in the military agencies; 25 distinct regions in Central Europe be and the conventional stereoplotters, because tween 48° and 52° OIth latitude and 16° East of their lower cost, are widely used in civi longitude. These regions agreed quite well lian applications. In addition, automatic re with a qualitatively formed physiographic mote sensing techniques based on the prin map of the area. James (1961) used a rough ciple of spectral analysis and density slicing ness index to distinguish various Wisconsin have been developed for the automatic clas substages in Indiana. sification of vegetation and identification of Salisbury (1962), using a random grid drainage ways, bodies of water, and snow sampling technique, found high correlation coverage. Thus, instrumentation and coefficients between the landslopes of techniques are presently available for pro selected points and the type ofglaciated ter ducing the digital data necessary for the au rain in which the points were located. tomatic classification of soils by this ap Several of the more recent efforts in the proach. field have been conducted, or sponsored, by The development of such automatic soil the U. S. Army Waterways Experiment Sta classification technique will involve (1) the tion, with primary emphasis on evaluating development of a set of quantitative factors the terrain for vehicular ground mobility which either individually or in combination (Dornbusch, 1967; Garrett and Shamburger, can be used to describe the geometry of the 1967; Wright and Burns, 1968). surface topography; (2) the development of The work by Vadnais (1965) and Philips suitable computation techniques to compute (1970) at the University of Illinois were the these terrain factors from digital terrain data; first attempts directed toward the applica (3) the determination of the quantitative ter tion of quantitative terrain factors to soil rain factors for different soil types, and classification. Vadnais conducte,d his study (4) the development of statistical methods with test areas located within the Wisconsin for the identification and classification of glacial deposits in Wisconsin, Illinois, and soils based on their quantitative terrain fac Indiana. A total of 221 test cells of I-mile tors and on data collected from remote sens radius each were used. The soil associations ing. found in these test areas included sandy Recent research at the University of Il loam till, loam till under prairie vegetation, linois has shown that quantitative terrain fac loam till under forest vegetation, loam till tors can be computed from digital terrain with 42 to 60 inch loess cover under prairie data to describe the terrain and that direct vegetation, silty clay loam till under prairie correlation exists between these terrain fac vegetation, silty clay loam till under forest tors and soil types. vegetation, and medium and moderately