<<

Quantifying Outcrops in NRCS Sol Map Units Using Landsat-5 Data

Carol A. Moore, Glenn A. Hoffmann, and Nancy F. Glenn

abstract

Rock outcrop places many limitations on land use. For this reason, identifying its presence in a map unit is as important as identifying major . Presently the most accurate way to de- termine the percentage of rock outcrop in any one soil map unit involves transecting, which is time-consuming and not practical for large areas. Remote sensing can provide more accurate data for the surveys and a tremendous time saving. Endmembers, which are spectrally unique materials selected within the scene, were chosen for rock outcrop by analyzing National Aerial Photography Program (NAPP) photography under a stereoscope. Spectral Angle Mapper (SAM) classifications were performed on Landsat data of the study area to classify vegetation and rock outcrop. Global Positioning Satellite (GPS) ground-truth data with 30-meter buffers were used to validate the rock classifications. Although it is likely the amount of rock outcrop is overestimated in the final image, additional methods could ultimately produce a successful methodology to accurately map and quantify rock outcrop.

he objective of this work was to develop a remote sensing 250,000 acres. Clark County lies at the foot of the continental divide Tmethodology to quantify basalt rock outcrop for NRCS soil mapping and includes the cold, arid, northeastern Snake River plain. The purposes in the Clark Area Soil Survey. Soil surveys provide users with mountainous area is dissected with many canyons and creeks, all assessments of vegetation, soil, and rock outcrop in given areas. A draining south. The soils are formed dominantly in and eobian common member of any soil map unit description in this survey area is sands and have varying degrees of development depending on the age exposed basalt bedrock. The amount of rock outcrop present is highly of the lava flows and where they are found. variable because of the relief of the lava flows and the amount of soil The specific area of study for this remote sensing project is that has been deposited on them. The quantity of forage available for concentrated along the mountain ranges and valley surrounding the domestic animals and wildlife, as well as the placement of routes for cities of Dubois and Spencer, ID (Fig. 1). The elevations for these water pipelines and roads, are highly dependent on the amount and cities are 5148 and 5883 feet, respectively. Pleistocene basalt lava location of rock outcrops. Due to the limitations rock outcrop places covers much of the central and eastern valley floor. Quaternary alluvial on land use, identifying its presence in a map unit is as important as deposits cover the western portion of the valley floor, moving into identifying major soils. Presently the most accurate way to determine Pliocene and Upper Miocene felsic volcanic rocks and rhyolite flows the percentage of rock outcrop in any one soil map unit involves in the mountain range. Vegetation in conducting transects where the soil scientist walks across the landform this area includes antelope bitter brush, and counts the total number of paces. The number of paces that fall on sagebrush, bushy birds beak, buckwheat, rock outcrop divided by the total gives the percentage of rock. While this blue-bunch wheatgrass, and needle-and- is an accurate estimation method for a small area, it is not a practical thread grass (Fig. 2). The average annual air means of determining the amounts of rock outcrop on 250,000 acres of temperature is approximately 43° Fahrenheit land. Remote sensing can provide a valuable tool for these purposes, and the average annual precipitation is providing more accurate data for the surveys as well as a tremendous approximately 13 inches. The average time savings. number of days with at least one inch of The Clark Area Soil Survey covers approximately 770,000 acres snow on the ground is 118. Farming and in Clark and Butte counties. Of this area, lava fields comprise about ranching are the most important industries in the county. Most irrigated lands are C.A. Moore, Idaho State University, Dep. of Geosciences, Idaho Falls, ID situated near the 5200-ft elevation level (email: [email protected]); G.A. Hoffmann, MLRA Soil Survey Project and are a hay base for the livestock. These Leader, USDA-NRCS, Idaho Falls, ID; N.F. Glenn, Research Associate Professor, Idaho State University, Dep. of Geosciences, Boise, ID.Carol lands are irrigated from the Eastern Snake Moore is currently an Idaho State University Junior pursuing a Bachelor River Plain Aquifer. Wheat and other Fig. 1. General location of of Science degree in Earth and Environmental Systems. This research study area within the Clark was designed and conducted as part of a class project and independent small grains are grown on dry farms. Area Soil Survey (Idaho study at ISU. Work was completed during the spring and fall semesters Other nonirrigated areas are mainly of 2006 while the senior author was a student employee with NRCS. Geospatial Data Clearing- Published in Soil Surv. Horiz. 48:59-62 (2007). wildlife habitat. house, 2004). FALL 2007 Endrnember Collection Spectra

Veg W2

X. Brush NE2 Brush

Outcrop South

Cliff Bands West i I - Band Number 4

Data and Methodology Landsat 30-rn resolution data acquired July 6, 2006 were used for this project, along with USDA National Agriculture Imagery Program (NAIP) 1-rn resolution imagery, a digital elevation model (DEM), and black and white National Aerial were created for rock outcrop: one representing basalt outcrop in the Photography Program (NAPP) 1998 photographs at 1:24,000 scale, a Eastern portion of the valley and one representing some cliff bands in map, and GPS ground-truth data. All imagery was processed the mountain range surrounding the Western portion of the valley. Using using ESRI ArcGIS 9.1 (Environmental Systems Research Institute, the false color Landsat image and field knowledge as a guide, eight 2005) and Environment for Visualizing Images (ENVI) 4.3 (ITT Visual additional regions of interest were created for various other components Information Solutions, 2006), a software package for analyzing all types such as irrigated agricultural land, brush, grasses, and other vegetation. of remote sensing data. These were selected from arbitrary areas in the image that were The first step in the process was to identify several areas of large believed to contain these components. Due to time constraints, no outcrop to be used as training data. Outcrops of 30 m or larger were ground-truth data were obtained to provide confirmation. These 10 desired due to the 30-rn resolution of the Landsat data. The NAIP image regions of interest were used to define the training data endmembers, or was visually inspected for this purpose, but even when draped over the spectrally unique materials selected within the scene (Fig. 3). Spectral DEM, it was a cumbersome task to differentiate rock outcrop from other profiles of healthy, green vegetation are generally characterized by features. For this reason, the NAPP photography was inspected under higher values in NIR band 4 than in SWIR band 5; however, this can vary a stereoscope. This made the relief much more apparent and seemed greatly depending on variables such as species of plant, environmental to make the task of locating the outcrops a bit easier since many of the conditions during growth, and background reflectance (mixed larger outcrops are located on pressure ridges or large mounds. Several components within the pixel). We can see in Fig. 3 the spectral profiles areas believed to be outcrop were chosen and digitized over the NAIP for Rock, Cliff Bands, and Irrigated Ag clearly fit their respective criteria. image, taking care to ensure the geographic coordinate system datum However, the spectral profiles for the five vegetation endmembers at the matched that of the Landsat datum. top of the plot resemble criteria for rock in SWIR band 5, and the Brush North spectral profile contains high values in both NIR band 4 and SWIR Landsat data often are used for geological mapping. Traditional band 5. Mixed components within the pixels are likely the reason for this, interpretation of the data couples false color composites, principal but we can only speculate due to the lack of ground-truth data. Although component analysis, and ratio images with image interpretation to visually similar, each endmember is spectrally unique and therefore determine the geology of a region (Riley et al., 2006). The traditional provides more data from which to differentiate the rock and cliff band near-infrared false color composites display healthy vegetation as red components during classification. in color, making readily apparent the areas that are not likely to be rock or bare soil. The Landsat data were displayed in false color, and Having now defined the training data, the next step was to perform the shapefile of digitized training polygons was then used as a vector a spectral classification of the image using the Spectral Angle Mapper overlay. Spectral profiles were interactively displayed for various pixels (SAM) classification method. SAM determines the spectral similarity within the digitized training polygons. The Landsat spectral profile of between two spectra by calculating the angle between them. The closer rock is characterized by higher values in the short wave infrared (SWIR) the angle, the more similarity exists. The SAM classification method was band 5 than in near-infrared (NIR) band 4. The spectral profiles for the run using three different methods: (1) using the original stacked Landsat selected pixels seem to fit the basic reflectance characteristics for rock. image, (2) creating a Principal Component Analysis (PCA) 2/1 ratio image and adding it to the Landsat stack to maximize and enhance the A spatial subset of the image was created to reduce the large spectral differences in the image, and (3) running the classification with Landsat image to a smaller size concentrated over the specific area of a spectral subset of the original Landsat stack by removing the thermal study. Using the training polygons as guides, two regions of interest infrared (TIR) band 6. TIR band 6 was removed because it has a 120- 60 SOIL SURVEY HORIZONS I

(-.

ji X, L I

- 4 .

r (y-

•Unc1sified - iC1itt aand WBruh Worth U Rock Outcrop / Irrigated Ag t,. Other veg l9reSh F I : Orazb NZ2 . + , egWl .., + 4 vog W2 + Là -

m spatial resolution that results in a different pixel size than the 30-rn within the valley. However, the cliff band (red) endmember, which was spatial resolution of the other bands in the stack. acquired from the western mountain range, seems to be including soil and/or surface cobble, since it is highly distributed throughout the classified image. The same circumstances may exist for some of the Results and Discussion vegetation endmembers that included dirt roads in the classes. Method 1 resulted in a great deal of rock (blue color) and cliff The data in Fig. 4 were analyzed to determine estimates of the band (red color) classes in the image. Based on our field knowledge of amounts of rock outcrop and cliff band present within the entire study the study area, the amount of rock and cliff band classifications in the area. image seemed excessive. In an attempt to achieve a more reasonable classification, Method 2 was applied. Unfortunately, the PCA band ratio Class Distribution Summary image did not seem to add any visibly significant improvement over the Full Scene: 1,207,752 points classified image created in Method 1. In fact, the two classified images Cliff Bands West (Red): 231,408 points (19.160%) (187,961,148.0000 m2) were very similar. Method 3 seemed to produce the most reasonable results (Fig. 4). Based on field knowledge and comparisons against Outcrop South (Blue): 160,148 points (13.2600/6) (130,080,213.0000 m2) the false color Landsat image, the classified irrigated agricultural areas A limited accuracy assessment was performed using 11 GPS points (yellow circles) and other vegetation (all other colors except blue and collected as ground-truth data. To allow for possible georectification red) appear to be fairly accurate. The highway and various areas of errors, 30-rn buffers were placed around each of the GPS points. They rangeland were left unclassified, which may be an accurate reflection, were then laid over the classified image for evaluation. The two zoomed since no training data were defined for pavement, and its likely that all images in Fig. 6 display four buffered GPS ground-truth data points laid the vegetation types were not specified in the chosen regions of interest. over the classified image. The buffer is represented by a white circle According to the geology map (Fig. 5) and the soil scientist notes, the surrounding a center GPS point. Any pixel classified as rock or cliff band western mountain range consists mainly of rhyolite flows, whereas the (red or blue pixels) positioned within the 30-rn buffer was considered valley floor consists mainly of basalt outcrops. This being the case, the accurate. Based on these criteria, 9 of the 11 ground-truth data points rock (blue) endmember, which was acquired from the eastern portion were accurate, constituting an accuracy rating of 82%. of the valley floor, appears to be quite accurate, as it occurs primarily

61 FALL 2007 .r r; ...... __ [mw

I ii Quatemary alluvial deposits Quaternary surlicial cover, including colluvium, fluvial, alluvial fan, lake, and windblown deposits. Included fluveolian cover on Snake River Plain, (Snake River Group). Quaternary windblown deposits; sand dunes and loess. The accuracy rating of 82% is encouraging, although this is Pleistocene silicic volcanic rocks; rtiyolite lava and ash-flow tuft (includes Yellowstone believed to be extremely inflated due to the limited number of ground- Pleistocene basalt lava, 2 million to 12,000 years old, flows have some vegetation and surface . truth data points used for the accuracy assessment. A greater number of Pleistocene and Pliocene basalt lava and associated basaltic tuft (deposited close to collected ground-truth data points for each of the classes would provide basaltic vent). Pliocene and Upper Miocene stream and lake deposits (Salt Lake Formation, a more effective accuracy assessment. [!] Starlight Formation, Idaho Group). Pliocene and Upper Miocene fetsic volcanic rocks, rfiyolite flows, tufts, ignimbntes. (in Owyhee [J County and Mt. Bennett Hills, this should be Tmf). Further interpretation should be done using band math ratios Eocene granite, pink granite, syenife, rhyolite dikes, and rhyolitic shallow intrusive; last phase of the Challis magmatic event (46 to 44 Ma). Forms craggy scenic mountain landscape in particularly suited to distinguish specific geological features. For central and northern Idaho. Pleistocene and Pliocene stream and lake deposits; sand, gravel and mud; Lake Idaho example, lrtzana et al. (2003) suggested that Landsat band ratio 5/1 will sediments; Glenns Ferry Formation; Idaho Group distinguish mafic igneous rocks and band ratio 5/43/4 will successfully [] Miocene intrusive rock; gabbro and diabase sills (includes part of Salt Lake Formation). Paleocene and Cretaceous and conglomerate (Beaverhead Formation). discriminate between mafic and nonmafic rocks. Since common mafic Permian and Carboniferous sedimentary rocks (Snaky Canyon Formation). rocks include basalt, this may be a viable option. Other methods may Cretaceous sedimentary rocks. include using the TIR band due to the heat absorption properties of the Jurassic sedimentary rocks. basalt rock or coupling the Landsat data with other imagery such as [II] Triassic sedimentary rocks. radar to emphasize both spectral and textural features. With many other Permian and Pennsylvanian sedimentary rocks. I] options still left to consider, it is believed that further research could [] Mississippian sedimentary rocks. result in a successful methodology to differentiate and quantify rock, Devonian, Silurian, and Ordovician sedimentary rocks. Neoproterozoic sedimentary rocks undivoled soil, and vegetation components within soil map units. Over one million acres of the lava field landform are found collectively within Jefferson, Madison, Bonneville, Bingham, Fremont, and Power counties, which are scheduled either for soil survey update or Conclusions maintenance in the near future. Since this landform is extensive on the Due to the variability of rock outcrop within lava field landforms, Snake River Plain, any soil survey effort within this plain would benefit estimated percentages of outcrop will fluctuate between transects, from successful methodologies to quantify basalt rock outcrops. sometimes even within the same map unit. Bearing this in mind, the percentages of rock outcrop derived from the class distribution summary Acknowledgments appear to support preliminary percentages obtained by transecting. We thank Bill Hiett and Carla Rebernak for their assistance with fieldwork. While these results are promising, we believe that the area of rock outcrop is being overestimated. The vegetation endmembers seem to References contain too much soil, and the rock and cliff band endmembers seem Environmental Systems Research Institute. 2005. ArcGlS 9.1. ESRI, Redlands, CA. to have too much vegetation and/or soil. We can see from the spectral Idaho Geospatial Data Clearinghouse. December 2004. Color shaded relief of Idaho profiles of the endmembers (Fig. 3) that the spectral signature for several with a horizontal grid spacing of 10 meters. Available at httpJltnsideidaho.orgl of the vegetation endmembers is very similar to that for the rock and cliff data/IGDC/archive/shdrlfcolorl Om_id_igdc.tgzldaho (accessed January 2007, verified July 2007). Geospatial Data Clearinghouse, Moscow, ID. band endmembers. A typical spectral response for healthy vegetation Inzana, J., T. Kusky, G. Higgs, and R. Tucker. 2003. Supervised classifications will show a considerably higher value in the NIR region of the spectrum of Landsat TM band ratio images and Landsat TM band ratio image with radar for geological interpretations of central Madagascar. J. Afr. Earth Sci. (band 4) and is often referred to as the "red edge." It was noted for the 37:59-72. vegetation endmembers in question that the red edge was not apparent. ITT Visual Information Solutions. 2006. ENVI 4.3. ITT Visual Information Solutions, Rather, the signatures more closely resembled those for rock and/or Boulder, CO. soil. From this it may be concluded that future methods should include Riley, D.N., M. Barton, and C. Dalton-Sorrell. 2006. Fusion of Landsat-5 thematic mapper and shuttle imaging RADAR-C data for geological mapping in Eastern steps to obtain larger, better defined or spectrally "pure" endmernbers. Maine, USA. Available at www.gis.usu.edu/docs/protecled/procs/asprsl This may prove challenging because of the wide 30-rn spatial scale of asprs2000/pdffiles/papers/222.pdf (accessed November 2006, verified July 2007). Landsat data.

62 SOIL SURVEY HORIZONS