An Approach to Lineament Analysis for Groundwater Exploration in Nicaragua

Jill N. Bruning, John S. Gierke, and Ann L. Maclean

Abstract 2004; Hung et al., 2005; Murphy and Burgess, 2005; Wells in bedrock aquifers tend to yield more water where Khan and Glenn, 2006; Meijerink et al., 2007). There is no they intersect fracture networks. Lineament analysis using well-accepted or proven protocol for mapping lineaments, nor satellite imagery was employed to identify surface expres- have different approaches been compared in non-ideal sions of subsurface fracturing for possible new well loca- regions. The technical aim of this work brings together tions. An imagery integration approach was developed to different data-processing tools, such as ERDAS Imagine® and evaluate satellite imagery for lineament analysis in terrain ArcMAP®, in conjunction with a variety of satellite images where the influences of human development and vegetation (QuickBird, Landsat-7 ETM, ASTER, and RADARSAT-1), field confound lineament interpretation. Four satellite sensors observations, geological and topographic maps, and a DEM in (ASTER, Landsat-7 ETM, QuickBird, RADARSAT-1) and a DEM order to compare lineament interpretations derived from were used for lineament mapping a volcanic region of multiple sources in a rural municipality location in Nicaragua. Image processing and interpretations obtained 12 Nicaragua. complementary products, which were synthesized into a raster image of lineament-zone coincidence for creating a Previous Work lineament delineation map. Nine of the 11 previously Lineament identification was first performed with aerial mapped faults were identified from the coincidence-based photography and first generation satellite imagery using map along with 26 new lineaments. The locations of ten stereo pairs, light tables, and transparencies (Gupta, 2003). new lineaments were confirmed by field observation. Optical sensors of moderate spatial resolutions have only RADARSAT-1 products were best for minimizing anthro- recently been used extensively for lineament analysis pogenic features but not able to identify all the geological (Loizzo et al., 1994; Drury and Andrews, 2002; Lee and lineaments. Moon 2002; Ricchetti, 2002; Inzana et al., 2003; Hung et al., 2005; Ricchetti and Palombella, 2005; Arellano-Baezo et al., 2006; Khan and Glenn, 2006, Meijerink et al., 2007; Sander, Introduction 2007; Mutiti et al., 2010). Certain bands of ASTER and Lineament analysis in hard-rock terrains has been performed Landsat were specifically designed to detect geological widely as a means for groundwater exploration. Using structure (Drury and Andrews, 2002). Khan and Glenn satellite imagery, lineaments are detected by surface feature (2006) mapped a remote area of northern Pakistan using patterns such as vegetation, drainage, outcrop truncations, ASTER imagery and discovered two active strike-slip faults soil moisture, and topography. Such lineaments are indica- previously unmapped. tive of secondary porosity in the form of fractures, and if A few studies have employed imagery from multiple intersected by a well have the potential to supply large and sensors for lineament detection (Akman and Tüfeçi, 2004; reliable quantities of water (Mabee et al., 1994; Kresic, 1995; Hung et al., 2005; Murphy and Burgess, 2005). Hung et al. Sander et al., 1997; Edet et al., 1998; Magowe and Carr, (2005) compared lineament interpretations from Landsat-7 1999; Mabee, 1999; Park et al., 2000). ETM and ASTER imagery and observed fewer erroneous A variety of lineament analysis techniques using results from ASTER derived lineaments compared with remotely sensed data exist and have been developed in near Landsat-7 ETM derived lineaments. They attributed the ideal settings where influences of anthropology and climatic difference to the higher spatial resolution of ASTER data. situations are minimal and vegetation, if any, is in a natural Satellite imagery with finer spatial resolution has not been state (Boeckh, 1992; Krishnamurthy, 1992; Mabee et al., 1994; commonly employed in lineament studies for groundwater Henderson et al., 1996; Mahmood, 1996; Edet et al. 1998; exploration primarily due to cost and limited spectral resolu- Mabee, 1999; Magowe and Carr, 1999; Robinson et al., 1999; tion (Sander, 2007). However, water-resource studies have Abouma-Simba, 2003; Paganelli et al., 2003; Glenn and Carr, used high spatial resolution sensors such as QuickBird and Ikonos to monitor land-use and land-cover Sawaya et al. (2003). Although shadows from tall objects affected classifica- tion results, these types imagery have great potential for water studies at local scales (Sawaya et al., 2003). This was further Jill N. Bruning is a Consultant, Chester, VT 05143. demonstrated by Loveless et al. (2005) who utilized Ikonos John S. Gierke is with the Department of Geological and Mining Engineering and Sciences, Michigan Technological University, 1400 Townsend Drive, Houghton, MI 49931 Photogrammetric Engineering & Remote Sensing ([email protected]). Vol. 77, No. 5, May 2011, pp. 509–519. Ann L. Maclean is with the School of Forestry and 0099-1112/11/7705–0509/$3.00/0 Environmental Sciences, Michigan Technological University, © 2011 American Society for Photogrammetry 1400 Townsend Drive, Houghton, MI 49931. and Remote Sensing

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 509 data to map surface expressions of geological “cracks” associ- provide structural information for geological mapping in ated with the tectonic setting in coastal Chile. These cracks northern Alberta, Canada. Lineament interpretations were range in aperture from a few centimeters to 2.5 m and were calibrated using several detailed, reputable structural studies. easily observable due to the hyper-arid climate of the region Their study showed principal components 2 and 3 preserved and the sensor’s 1 m spatial resolution (Loveless et al., 2005). topographic information, such as pattern and texture, A variety of image processing techniques have been necessary to interpret bedrock structures. RADARSAT has also utilized to enhance linear features in optical imagery been used to delineate geomorphic features using change (Krishnamurthy et al., 1992; Ricchetti and Palombella, 2005; detection between two scenes acquired at different times of Khan and Glenn, 2006). Khan and Glenn (2006) employed the year (Radarsat Geology Handbook, 1996; Glenn and Carr, decorrelation stretches and principal components analysis 2004) due to differences in surface moisture. High reflectivity (PCA) to aid in geological mapping. Offsets of rock types values in radar imagery can be caused by increased moisture were apparent using these techniques as they exploit the content in both soils and vegetation and have been shown to unique spectral signatures of each lithological formation. enhance linear topographic features (Glenn and Carr, 2003). Krishnamurthy et al. (1992) explored a variety of digital DEMs have also been shown to be useful for detecting image processing techniques for groundwater investigations lineaments because they can eliminate bias caused by using a Landsat TM image of Karnataka, India. Their research inherent east-west sun illumination (Henderson et al., 1996; resulted in 13 output products, which were assessed to Yun and Moon, 2001). Studies that detect lineaments solely delineating geologic and geomorphic features. A qualitative from DEMs rely on the assumption that the majority of assessment of the maps was made and the images were lineaments in a given study area are geomorphic rather than ranked as good, moderate, or poor. tonal and this assumption is valid for most regions as valley Active radar sensors, such as RADARSAT-1 and JERS-1 SAR, and cliff orientations are typically controlled by faulting employ longer wave lengths and complement optical sensors direction (Yun and Moon, 2001). for lineament detection. Radar sensors respond to surface topography, roughness, and dielectric properties; optical sensors respond to optical and thermal attributes Study Area (Mahmood, 1996; Radarsat Geology Handbook, 1996). Thus, This research was conducted in and around Boaco, radar images contain less anthropogenic and vegetation Nicaragua, a rural municipality in need of additional information and exhibit more topographic information drinking water wells located in the interior highlands, (Mahmood, 1996). roughly 100 km northeast of Managua (Figure 1). Nicaragua’s Radar has also proven to be advantageous for mapping population is highly dependent upon groundwater resources in vegetated regions (Paradella et al. 1998, Abouma-Simba with 95 percent of people relying on groundwater (Bund- et al., 2003). Natural variations in vegetation cover are often schuh and Alvarado, 2007). Nicaragua’s groundwater closely linked to geology and are sometimes detectable using dependency is significantly higher than the rest of the RADARSAT-1/Landsat TM integrated products (Paradella et al., world, where only about 30 percent to 50 percent of the 1998) due to the differences in incidence angles of the two global population relies on groundwater (Bundschuh and types of imagery. Paradella et al. (1998) produced a geologi- Alvarado, 2007). Groundwater resources of Nicaragua are cal map of their study area from RADARSAT-1/Landsat TM naturally high quality and are reliable throughout the year imagery and mapped five geological units and two primary (Bundschuh and Alvarado, 2007). Conversely, surface waters lineament systems. Digital image processing of radar imagery in Nicaragua are generally polluted and rivers and streams is often performed to highlight geological features. Paganelli experience low flows during the dry season (November et al. (2003) used a stack of four RADARSAT-1 scenes to through April) (Bundschuh and Alvarado, 2007).

Figure 1. Location of Study Area. The municipality of Boaco, Nicaragua, denoted in the ASTER image above, is located at 12°28N, 85°40W.

510 May 2011 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING TABLE 1. IMAGE PARAMETERS

Image Acquisition Acquisition Solar Azimuth Solar Angle Off-Nadir Angle Orbital Sensor Specifications Date Time (local) (degrees) (degrees) (degrees) Direction

QuickBird Standard 2A 01 Jun 2006 11:39:15 54.2 72.5 10.3 n/a ASTER Level 1B 24 Nov 2005 11:16:36 151.1 52.4 5.7 n/a Landsat-7 ETM Geolocated, 13 Nov 2001 10:45:11 142.1 142.1 0 n/a RADARSAT-1 Projected 24 Feb 2007 29.8 Ascend Standard Beam 09 Sep 2006 n/a n/a n/a 29.8 Ascend Mode 3 08 Nov 1997 31.7 Descend

The study area, comprising 64 km2, was chosen because scene without cloud cover over the study area was not suitable groundwater resources must be proximal to where available, meaning that assessment of the thermal bands for the water will be used. Proximity is influenced by the lineament detection was not possible. Three RADARSAT-1 availability of technological and economic resources for scenes included two ascending orbits acquired on 09 developing the water supply. A realistic area to consider for September 2006 (more than half way through the wet developing a groundwater source for this community is the season) and 24 February 2007 (near the end of the dry surrounding 4 km2. However, faulting and fracturing most season) and one descending orbit acquired on 08 November often occur on scales spanning many kilometers. In order to 1997 (transition from wet to dry season). The scenes were avoid overlooking geologically significant features that acquired using Standard Beam Mode 3. A summary of key penetrate into areas proximal to Boaco, the study area was image parameters is provided in Table 1, and the imagery, sized larger than is practical for groundwater development. plus the DEM are shown in Plate 1. Boaco is located upon the Chortis Block in the north- In addition to the satellite imagery, a DEM was produced west part of the Caribbean Plate. The Chortis Block is the from a printed topographic map (INETER, 1987) by manu- only continental constituent of the Caribbean Plate (Weyl, ally digitizing the 20 m contour lines as no available high- 1980; Bundschuh and Alvarado, 2007), and its basement is resolution DEM existed for the study area and using an composed primarily of Precambrian-Paleozoic metamorphic iterative finite difference interpolation technique to model and igneous rocks (Rogers, 2003; Bundschuh and Alvarado, sharp changes in topography while minimizing sinks using 2007). Locally, the basement materials are overlain by ArcMAP®. Quaternary alluvium, lavas, and pyroclastics and tertiary ash, basalts, andesites, dacitic ignimbrites, and plutonic intrusives of the Coyol and Matagalpa Groups (Weyl, 1980; Methods Rogers, 2003). Faulting occurs in two primary directions at Prior to image processing, the RADARSAT-1 scenes were azimuths of approximately 75° and 150°. The terrain is orthorectified with shadow and layover areas filled using characterized by widespread plateaus and fault-block ASF MapReady© Tool (2007). This required a mosaic of four mountains, which are present in much of Honduras and 90 m Shuttle Topography Radar Mission (STRM) (USGS, Nicaragua and in portions of El Salvador and Guatemala 2000) DEMs. The RADARSAT-1, ASTER, and Landsat-7 ETM (Weyl, 1980). Rivers and other drainages incise deeply into images were then georeferenced (UTM, Zone 16N, WGS84) to the topography, often following fracture patterns and create the QuickBird image, which was utilized as a base map. a rugged landscape. Dense areas of vegetation are limited to Image processing was performed with ERDAS Imagine®. remote areas (high elevations) and valleys are, impacted by Processing steps for the four images are outlined in Table 2, land-use. with extensive detail provided in Bruning (2008). Results Several characteristics make lineament analysis in this from processing the Landsat-7 ETM scene did not delineate area difficult. Few outcrops are available for structural any faulting characteristics. Hence the scene was eliminated measurements, and the tropical climate, severe anthro- from further consideration in the research. pogenic influences on land-use/land-cover, and a lack of Image differencing was used to compare the wet and hydrological and geological information limit traditional dry season RADARSAT-1 images. A change detection image lineament analysis techniques. These conditions are not was created by subtracting the wet-season image (acquired unique to Boaco and the results of this study can serve as a 09 September 2006) from the dry-season image (acquired proxy for similar fractured hard-rock terrains not only in 24 February 2007). Central America but elsewhere in the world. A substantial number of images were created by utiliz- ing the various processing techniques, and it was impracti- cal to attempt lineament interpretations on all of them. To Data Used eliminate images that lacked potential for lineament delin- Imagery from four satellite sensors, QuickBird, Landsat-7 eation, each image was visually evaluated for its ability to Enhanced Thematic Mapper Plus (ETM), ASTER, and delineate faulting in the two known fault directions RADARSAT-1, were used in this study. Acquisition dates for (azimuths 75° and 150°) by overlaying known faults digi- the data were limited by the extent of cloud cover, and it tized from the geological map. The images were ranked as was only possible to obtain one scene for each sensor. poor, moderate, or good in a similar fashion to Krishna- QuickBird imagery, available to the public through Digital- murthy et al. (1992). The results of the assessment are given Globe, Inc. was chosen primarily for its high spatial resolu- in Table 3. Images highlighted in grey were deemed accept- tion (0.6 m). Landsat-7 ETM was selected due to popularity able for further analyses. Two composite products were among earth observing scientists and the three infrared generated from the RADARSAT-1 data. The first composite bands (bands 4, 5, 7). ASTER imagery was utilized for its six included the first principal component from each of the short-wave infrared (bands 4 through 9), as well as its high three despeckle iterations. The second composite contained spatial resolution very near infrared bands. A nighttime the first principal component from the second despeckle

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 511 Plate 1. False color composites of the four satellite images: (a) RADARSAT-1, (b) ASTER, (c) DigitalGloge QuickBird, and (d) Landsat-7 EMT. The DEM generated from a topographic map with a verticle exaggeration of 2 is shown in (e). The RADARSAT-1 image is a stack of three scenes acquired on 24 February 2001, 08 November 1997, and 09 September 2006, and displayed as red, green, and blue, respectively.

iteration, the change detection image and ASTER band 1. features of any extent (regional or local) can be easily Creation of composite images followed by a re-extraction of visualized. lineaments from these composites was performed to investi- All of the digitized faults were overlain on the Quick- gate whether the combined use of two or more products Bird image. Attributes exhibited by the QuickBird image further enhance the ability to discern lineaments compared near these faults, such as tonal variations in the soil and to a single product. vegetation, were noted as possible indicators of fracturing. Each image kept for further analyses was manually Distances perpendicular to the faults where measured. In all, interpreted for lineaments and twelve interpretations were twenty-four measurements were made in areas where the produced. The data was displayed in ArcMAP® and visible exhibition of fracture-like features in the QuickBird image lineaments were digitized. An advantage of using GIS was distinct. The average width was 172 m, with a standard software, such as ArcMAP®, for lineament interpretations is deviation of 56 m (equal to approximately 93 QuickBird that it enables the user to change viewing scales so that pixels, 4.5 RADARSAT-1 Standard Beam Mode pixels, and 3.7

512 May 2011 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING TABLE 2. PROCESSING FLOW FOR SATELLITE IMAGERY were removed from the coincidence raster shapefile because these areas would allow lineaments of minor extent to pass 1) QuickBird: Statistical Assessment through the filter (Bruning, 2008). Areas of high coinci- a) Radiometric Correction: Dark Image Adjustment (Result dence, where many lineament interpretations agree, show Negative) b) Optimum Index Factor (OIF) features likely geologic in origin. A ground-truthing-focused i) Intensity Hue Saturation (IHS)Transformation field campaign to Boaco occurred in March 2008, corre- ii) Various Stretch Enhancements sponding to the latter part of the dry season, with the c) Texture Enhancement objective to evaluate lineament interpretation using visual d) Edge Enhancement inspection of possible lineaments. The validation was e) Principal Component Analysis (PCA) developed to identify lineament-like features in the field f) Normalized Difference Vegetation Index (NDVI) without guidance from existing lineament maps. The g) Tassel Cap (TC) Transformation validation was constrained by accessibility and time. h) Various Stretch Enhancements on Various Band Combinations Therefore, only regions accessible by road were inspected 2) ASTER for lineament occurrence. Field-observed lineaments were a) VNIR: Stack & Subset, Statistical Assessment located using a Garmin GPSmap76 and required attributes i) Radiometric Correction: Dark Image Adjustment (Result such as the strike of the feature were recorded. Negative) ii) Intensity Hue Saturation Transformation iii) Various Stretch Enhancements Results and Discussion iv) Principal Component Analysis The ability of the various image composites to produce b) VNIR & SWIR: Stack & Subset acceptable lineament interpretations were assessed using i) Various Stretch Enhancements ii) Principal Component Analysis two complementary approaches: determination of the iii) Optimum Index Factor percent of the lineament interpretation that identified false 1. Stack Bands 2, 3, & 4: Various Stretch Enhancements positives, called “% False Identification”; and determination 2. Stack Bands 4, 5, & 6: Various Stretch Enhancements of how much of the coincidence raster is explained in a single lineament interpretation, called “% Coincidence” 3) Landsat-7 ETM : Stack & Subset, Statistical Assessment (Table 4). These two assessment methods used area a) Radiometric Correction: Dark Image Adjustment (Result Negative) calculations of the original lineament interpretation and b) Optimum Index Factor the coincident interpreted polygons. Four of the final i) Intensity Hue Saturation Transformation interpretations (highlighted in grey in Table 4), are illus- ii) Various Stretch Enhancements trated in Plate 2a through 2d and provide a visual compari- c) Principal Component Analysis son to the coincidence raster. Overall, RADARSAT-1 based d) Tassel Cap Transformation images ranked the highest for producing acceptable linea- e) Various Stretch Enhancements on Various Band Combinations ment interpretations. 4) RADARSAT-1: Orthorectification & Geolocation, Subset, The False-Identification Assessment method (Table 4) Statistical Assessment ranked the RADARSAT-2 Despeckle Level 3 product as a) Stack detecting the fewest (16.2 percent) false lineaments. This i) Various Stretch Enhancements assessment method is biased towards products that have a ii) Principal Component Analysis lower number of lineaments. The Coincidence Assessment b) Despeckle, Stack method, listed in Table 4 and displayed in Plate 2a, ranked i) Principal Component Analysis RADARSAT-1 Despeckle Level 2 product as the best with 62.8 ii) Change Detection percent of the original lineament interpretation in common iii) Edge Enhancements with the coincidence raster. Unlike the False-Identification Assessment method, the Coincidence method is biased towards products that have the greatest number of linea- ments, as it results in more overlap, or coincidence, with ASTER VNIR pixels). The width measurements are most likely other lineaments. However, these areas of coincidence are larger than the fracture zones they represent due to exten- not necessarily made from lineaments of the same orienta- sive weathering in the study area. tion and therefore do not identify the same lineament. This The buffered lineament interpretations were overlaid phenomenon is especially evident in the QuickBird interpre- and areas of coincident lineaments noted. According to tation results displayed in Plate 2c, indicated by the many Magowe and Carr (1997), a coincident lineament is a segmented lines of coincidence (shown in blue). These lineament which has been interpreted in either two or more biases are overcome by considering results from both image types, interpretation trials, or observers (i.e., coinci- methods simultaneously. dence number 2). Unlike Magowe and Carr (1997), a Image-rank results show poor performance for products coincidence level 4 was used in this study as the thresh- derived from optical sensors, suggesting that these wave- old to detect lineaments that are coincident. This level was lengths are not appropriate for lineament delineation in this chosen because several of the lineament interpretations type of study area. One explanation for this is portrayed in used as input for creating the coincidence raster were made Table 3, which assesses the ability of each image product to from similar image products. For example, three individual exhibit faulting in the two primary directions. QuickBird interpretations produced from RADARSAT-1 products are very products exhibit NW/SE faulting better than NE/SW faulting, similar, including interpretations from the stack of all where as ASTER products exhibit NE/SW faulting better than original RADARSAT-1 scenes and the despeckled products NW/SE faulting; however neither QuickBird nor ASTER are (levels 2 and 3). Agreement with the coincidence raster at a classified as “good” at exhibiting either faulting direction. coincidence level 2 is therefore definite in these cases. This implies that solar illumination conditions (sun and To remove bias inherent to the coincidence raster, coinci- shadow) for each scene control what features can be delin- dence raster values 3 were removed and a binary coinci- eated. Features striking perpendicular to solar illumination dence raster was created to include only areas having a direction are much more observable due to enhancement of coincidence level 4. Sliver polygons of areas 60,000 m2 topography contrasts, which may be associated with

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 513 TABLE 3. ASSESSMENT OF PRODUCT IMAGES FOR EXHIBITING FAULTS; THE ABILITY OF EACH IMAGE PRODUCT TO EXHIBIT FAULTING IN THE TWO DOMINANT DIRECTIONS IS RANKED AS EITHER: P POOR, M MODERATE, OR G GOOD

Processing 150° Trending 70° Trending Processing 150° Trending 70° Trending Sensor Level Faults Faults Sensor Level Faults Faults

QuickBird Original Image M P ASTER VNIR Stretch M M Stretch Enhanced – Enhanced - Std. Std. Dev. (1.5) Dev. (1.6) band 1 MM band 1 PP band 2 MP band 2 MP band 3 PM band 3 MP VNIR PCA P M band 4 MP PC 1 PM PC 2 PM PCA PC 3 PP Composite of MP VNIR IHS P P PC 1, 2, & 3 I PP PC 1 MP H PP PC 2 MP S PP PC 3 MP PC 4 PP VNIR & SWIR Texture P P Stack Enhanced band 4 PM band 5 PP Tassel Cap band 6 PP TC 1 PP band 7 PP TC 2 PP band 8 PP TC 3 PP band 9 PP TC 4 PP OIF (2, 3, 4) M P VNIR/SWIR PCA Stretch Enhanced – Composite of PC Std.Dev. (1.8) 1, 2, & 3 PP band 2 PP PC 1 PM band 3 MP PC 2 PP band 4 MP PC 3 PP PC 4 PP IHS (2, 3, 4) P P PC 5 PP I PP PC 6 PP H PP PC 7 PP S PP PC 8 PP NDVI M M PC 9 PP RADARSAT-1 Stack of Scenes G M OIF (2, 3, 4) P P Stretch Stretch Enhanced – Std. Enhanced - Std. Dev. (2) Dev (1.8) Original Image MM band 2 PP (Sept.) band 3 PP Original Image PM band 4 MM (Nov.) OIF (4, 5, 6) M P Original Image PP Stretch (Feb.) Enhanced - Std. Despeckle - G M Dev. (1.8) 1st Iteration band 4 MM Despeckle - G M band 5 PP 2nd Iteration band 6 MP Despeckle - G M DEM Topographic lines M M 3rd Iteration Hillshade, z 2M M PCA G M PC 1 GM PC 2 MP PC 3 PP PC 1, 1st Despeckle G M PC 1, 2nd Despeckle G M PC 1, 3rd Despeckle M M Change Detection G G (Image Subset) Edge Enhanced NW Filter PM NE Filter MP SW Filter MP SE Filter MP Non-directional MP

514 May 2011 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING TABLE 4. EVALUATION OF IMAGE PRODUCTS TO IDENTIFY LINEAMENTS Vegetation is very reflective in the portion of the electromagnetic spectrum employed by ASTER band 3 and False- QuickBird band 4. The statistical assessments of these bands Identification Coincidence show significantly more variation in brightness values Sensor Product (%) (%) compared to the other bands in each sensor and strongly RADARSAT-1 Original 18.0 60.3 influence the information in principal component 1. Despeckle #2 18.7 62.8 Vegetation patterns in Boaco are manipulated by both PCA Despeckle #2 21.3 61.9 anthropogenic conditions and location of drainages. Despeckle #3 16.2 57.4 Anthropogenic conditions cause linear alignments of PCA Despeckle #3 25.0 60.8 vegetation to follow field boundaries and roads and are not Change Detection 39.9 53.9 necessarily reflective of subsurface structure. It is typically Composite #1 29.9 60.0 difficult to distinguish between these two types of vegetation ASTER Original VNIR 41.3 54.8 occurrences in satellite imagery, especially imagery of PCA VNIR 32.0 47.1 QuickBird Original 55.2 47.2 moderate spatial resolution. This further reduces the value Topographic DEM 43.3 36.1 of optical products to detect geological faults as many of the Map lineament interpretations based on vegetation are not ASTER & Composite #2 29.0 49.5 geological or geomorphic lineaments. RADARSAT-1 The image ranks show that ASTER product outperformed QuickBird products for both assessment methods. The spectral resolutions of ASTER VNIR bands and QuickBird bands 2, 3, and 4 are nearly identical, and the major differ- faulting. Conversely, the sun suppresses features striking ence between these two data sets is spatial resolution. The parallel to solar illumination direction because the down- results indicate that the 15 m spatial resolution of the ASTER strike illumination minimizes topographic contrasts. The VNIR bands provide the most appropriate level of detail; QuickBird scene was acquired with a solar azimuth of 54.2° whereas QuickBird 0.6 m spatial resolution maybe too suppressing NE/SW faulting, whereas the ASTER scene was detailed and hinders manual interpretation of the imagery. acquired with a solar azimuth of 151.1° suppressing NW/SE (Plate 2c). On the other hand, ASTER SWIR bands and Land- faulting. sat-7 EMT have a spatial resolution of 30 m, which pro- Another explanation for the poor performance of the vided too little detail to delineate even the regional faults. optical sensor products is that optical images in tropical The innate cloud cover of tropical regions and the dense climates are visually dominated by vegetation patterns. vegetation limit the amount of useful information which can

Plate 2. Level of agreement between select interpretations: (a) RADARSAT-1 PCA Despeckle Level 2, (b) ASTER VNIR Original, (c) QuickBird, and (d) DEM and the coincidence raster. Original lineament interpretations are displayed as white dashed lines surrounded by a 172 m wide buffer in red. Black areas represent the coincidence raster. Locations where the coincidence raster and original lineament buffer overlap are shown in blue.

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 515 be derived from optical imagery. Advanced image process- and anthropogenic targets. For this reason, lineament ing, including the successful processing techniques identi- interpretations made from RADARSAT-1 products are more fied by Krishnamurthy et al. (1992), did not show significant likely than optical image products to reflect subsurface improvement in the products’ ability to identify faulting in structure. this region. The regular cloud cover also hinders the choice RADARSAT-1 Despeckle Levels 2 and 3 products per- of scene acquisition dates, and the interpretations may have formed the best due to backscatter removal. The original been different if more satisfactory images were available. RADARSAT-1 product was saturated with random backscatter Moreover, spectral resolutions employed by the optical pixels inherent to this sensor and causing suppression of sensors are remarkably similar and provide little unique lineament features. When random backscatter was reduced, information. many of the lineament features became observable. Level 3 On the other hand, these results show that RADARSAT-1 exhibited less lineament features than Level 2, demonstrat- products are not subject to the problems inherent to optical ing that it is possible to over compensate for speckling and imagery, making these products better suited for lineament remove important features that aid in lineament delineation. interpretation in this region. The longer wavelength utilized Utilizing both the coincidence raster and DEM allowed for make RADARSAT-1 less influenced by atmospheric conditions a comprehensive mapping of interpreted lineaments to be and allows the sensor to collect imagery at any time in made with a high degree of confidence (Plate 3) with all but either an ascending or descending orbital path. Fusing two of the previously mapped faults discerned by interpreta- ascending and descending scenes produces a composite tion. It can be seen in Plate 3 that there are slight dislocations image that has two illumination directions, overcoming between the mapped faults and interpreted lineaments. This feature suppression in any direction. This is summarized in may be due to inconsistencies in scale or mistakes within the the product assessment listed in Table 3. The 12.5 m spatial geological map. The two mapped faults not observed using resolution of RADARSAT-1 Standard Beam Mode images fall the coincidence raster and DEM do not appear as topographic into the same level of detail as ASTER VNIR products, again expressions and still require ground confirmation. Lineaments providing an optimal level of detail for lineament interpreta- interpreted where there are no mapped faults display the tions. In addition, RADARSAT-1 products are dominated by same attributes in the coincidence raster and DEM as locations topographic features and are influenced less by vegetation where previously mapped faults are located.

Plate 3. Coincidence Raster and Final Lineament Interpretation. Level of coincidence is shown as a color gradient, locations of greatest agreement between lineament interpretations is displayed as red. The dashed lines display the interpreted faults based on the coincidence level. The solid lines are locations of previously mapped faults present on the most recent geologic map (Office of Cadasters and Natural Resources Inventory 1971).

516 May 2011 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING Plate 4. Visual evaluation of final lineament interpretation.

Ground-based observations of lineament features are property ownerships, municipality lines, etc. tend to reflect compared to lineaments interpreted from remotely sensed orientations of regional fault patterns and structure trends. imager and previously mapped faults in Plate 4, which Although performance of the QuickBird product to delineate shows that of the 42 features observed in the field, 21 were lineaments was poor, the imagery’s high spatial resolution interpreted using the coincidence raster and DEM. Disagree- proved useful in other aspects of the study. Fault-zone ment between the two is most likely due to errors contained aperture measurements were made from the QuickBird in both data sets. The study area lacks outcrops; therefore imagery which is a key aspect of the GIS analysis methodol- ground truth relied purely on observation of geomorphic ogy (Bruning, 2008). Finally, other features are easily seen in lineament features. These features include alignments of the QuickBird imagery and allow for analysis of land-cover valley, drainages, and cliffs and their locations have error. and land-use prior to deployment of a field campaign. The For example, the position of a drainage observed from a quality and detail of the measurements and observations ground vantage point does not exactly reflect the location discussed here would not have been possible using the other and orientation of a possible underlying structural feature. optical sensor types. This causes displacements and disorientations between satellite-derived and ground-observed features. Furthermore, it is possible that some of the ground-observed features are Conclusions not true lineaments. Many of the linear drainages observed Four complementary satellites were utilized for lineament from the ground were minor and may not be caused by analysis of volcanic terrain in a 64 km2 area around Boaco, subsurface fracturing whereas the remotely sensed based Nicaragua: ASTER, Landsat-7 ETM, QuickBird, and observations are much more regional. For these reasons, 21 RADARSAT-1. Lineament interpretations based on RADARSAT-1 of 42 lineaments being confirmed by ground truth is conser- products were superior to interpretations from other sensor vative, and it is probable that the satellite-interpreted products for this area that has undergone intensive weather- features correlate better with subsurface structure than these ing, is heavily vegetated, and significantly affected by ground validations results indicate. community and agricultural development. The most success- While producing high-quality lineament interpretations ful products were composites of three RADARSAT-1 scenes requires anthropogenic features to be minimized, such acquired 24 February 2007 (ascending orbit), 09 September features can provide some geological information and should 2006 (ascending orbit), and 08 November 1997 (descending not be ignored completely. In this region as well as in other orbit) after being processed to remove speckle. These results developing countries, orientations of roads, field boundaries, show that producing high-quality lineament interpretations

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 517 in this region requires anthropogenic features to be mini- Glenn, N.F., and J.R. Carr, 2004. The effects of soil moisture on mized and topographic expressions to be maximized. synthetic aperture radar delineation of geomorphic surfaces in However, utilizing RADARSAT-1 imagery alone for lineament the Great Basin, Nevada, USA, Journal of Arid Environments, interpretations may result in missed lineaments. It is 56(4):643–657. therefore recommended that products derived from both Gupta, R.P., 2003. Remote Sensing Geology, Second edition, Springer-Verlag Berlin Publishers, Heidelberg, Germany, 655 p. radar and optical sensors in conjunction with a DEM are employed to generate a coincidence raster from which a Henderson, D.B., D.A. Ferrill, and K.C. Clarke, 1996. Mapping final lineament map can be drawn. geological faults using image processing techniques applied to hill-shaded digital elevation models, Proceedings of the Southwest Symposium on Image Analysis and Interpretation, 08–09 April, pp. 240–245. Acknowledgments Hung, L.Q., O. Batelaan, and F. De Smedt, 2005. Lineament This material is based upon work supported by the National extraction and analysis, comparison of Landsat ETM and Science Foundation under Grant No. 0530109. Software, ASTER imagery. Case study: Suoimuoi tropical karst catchment, RADARSAT-1 data credits, and other data-processing resources Vietnam, Proceedings of SPIE: Remote Sensing for Environmen- were provided by the Alaska Satellite Facility. tal Monitoring, GIS Applications, and Geology, 29 October 2005, 5983:59830T, 12 p. INETER (Elaborado por el Instituto Nicaragüense de Estudios References Territoriales), 1987. 1:50,000 Topographic Map, Region V, Department of Boaco. Abouma-Simba, S., J.P. Deroin, and J.M. Regnoult, 2003. Combina- tion of SAR, SPOT, and geophysical data for geological Inzana, J., T. Kusky, G. Higos, and R. Tucker, 2003. Supervised mapping: the Nyanga Basin (SW Gabon) example, Proceedings classifications of Landsat TM band ratio images and Landsat of International Geoscience and Remote Sensing Symposium, TM band ratio image with radar for geological interpretations of 5:3314–3316. central Madagascar, Journal of African Earth Sciences, 37(1–2):59–72. Akman, A.Ü. and K. Tüfekçi, 2004. Determination and characterisa- tion of fault systems and geomorphological features by RS and Khan, S.D. and N.F. Glenn, 2006. New strike-slip faults and litho- GIS techniques in the WSW part of the Turkey, Proceedings of units mapped in Chitral (n. Pakistan) using field and ASTER XX ISPRS Congress: Geo-Imagery Bridging Continents, 12–23 data yield regionally significant results, International Journal July, Istanbul, Turkey, XXXV:899–904. of Remote Sensing, 27(18–20):4495–4512. Alaska Satellite Facility Distributed Active Archive Center Program Kresic, N., 1995. Remote sensing of tectonic fabric controlling and NASA, 2007. RADARSAT-1 scene R159031031G1S002, goundwater flow in dinaric karst, Remote Sensing of Environ- Standard Beam Mode 3, ASF, Fairbanks, Alaska, 24 February ment, 53(2):85–90. 2007. Krishnamurthy, J., P. Manavalan, and V. Saivasan, 1992. Applica- Alaska Satellite Facility Distributed Active Archive Center Program tion of digital enhancement techniques for groundwater and NASA, 2006. RADARSAT-1 scene R156630031G1S002, exploration in hard-rock terrains, International Journal of Standard Beam Mode 3, ASF, Fairbanks, Alaska, 09 Remote Sensing, 13(15):2925–2942. September 2006. Lee, T.H., and W.M. Moon, 2002. Lineament extraction from Alaska Satellite Facility Distributed Active Archive Center Program Landsat TM, JERS-1 SAR, and DEM for geological applications, and NASA, 1997. RADARSAT-1 scene R110503419G1S001, Proceedings of the International Geoscience and Remote Standard Beam Mode 3, ASF, Fairbanks, Alaska, 08 Sensing Symposium, 6:3276–3278. November 1997. Loizzo, R., G. Sylos Labini, M. Pappalepore, P. Pieri, P. Blonda, G. Arellano-Baeza, A.A., A.T. Zverev, and V.A. Malinnikov, 2006. Pasquariello, and M. Antoninette, 1994. Multitemporal Analysis Study of changes in the lineament structure, caused by of ERS-1 and Landsat TM data for Geological Studies, earthquakes in South America by applying the lineament Proceedings of the International Geoscience and Remote analysis to the ASTER (Terra) satellite data, Advances in Space Sensing Symposium, 08–14 August, 3:1478–1780. Research, 37(4):690–697. Loveless, J.P., G.D. Hoke, R.W. Allmendinger, G. González, B.L. Boeckh, E., 1992. An exploration strategy for higher-yield boreholes Isacks, and D.A. Carrizo, 2005. Pervasive cracking of the in the West African crystalline basement, Hydrogeology of northern Chilean Coastal Cordillera: New evidence for forearc Crystalline Basement Aquifers in Africa (E.P. Wright, and W.G. extension, Geology, 33(12):973–976. Burgess, editors) Geological Society Special Publication, Mabee, S.D., 1999. Factors influencing well productivity in glaciated (66):87–100. metamorphic rocks, Ground Water, 37(1):88–97. Bruning, J.N., 2008. A Digital Processing and Data Compilation Mabee, S.B., K.C. Hardcastle, and D.U. Wise, 1994. A method of Approach for Using Remotely Sensed Imagery to Identify collecting and analyzing lineaments for regional-scale fractured- Geological Lineaments in Hard-rock Terrains: An Application bedrock aquifer studies, Ground Water, 32(6):884–894. for Groundwater Explorations in Nicaragua, M.S. Thesis, Mahmood, A., 1996. Lineaments as groundwater exploration guides Department of Geological & Mining Engineering & Sciences, in hard-rock terranes of arid regions, Canadian Journal of Michigan Technological University, Houghton, Michigan, 132 Remote Sensing, 22(1):108–116. pp., Available online at: http://ils.lib.mtu.edu/vwebv/hold- Magowe, M., and J.R. Carr, 1999. Relationship between lineaments ingsInfo?bibId 661588 (last date accessed:28 February 2011). and ground water occurrence in western Botswana, Ground Bundschuh, J., and G.E. Alvarado, (editors.), 2007. Central America; Water, 37(2):282–286. Geology, Resources, and Hazards, Volumes 1 and 2, Meijerink, A.M.J., D. Bannert, O. Batelaan, M.W. Lubczynski, and T. Philadelphia: Taylor and Francis, 1436 p. Pointet, 2007. Remote sensing applications to groundwater, IHP- DigitalGlobe, 2006. QuickBird scene 005547368010_01_P001, Level VI, Series on Groundwater, United Nations Educational, Standard 2A, DigitalGlobe, Longmont, Colorado, 01 June 2006. Scientific and Cultural Organization, Paris, France, Drury, S.A. and M.E. Andrews Deller, 2002. Remote sensing and No. 16, pp. 304. locating new water sources, URL: Murphy, M.A., and W.P. Burgess, 2006. Geometry, kinematics, and http://www.unoosa.org/pdf/sap/2002/ethiopia/presentations/12s landscape characteristics of an active transtension zone, peaker01_1.pdf (last date accessed: 28 February 2011). Karakoram fault system, Southwest, Tibet, Journal of Structural Edet, A.E., C.S. Okereke, S.C. Teme, and E.O. Esu, 1998. Applica- Geology, 28(2):268–283. tion of remote-sensing data to groundwater exploration: A case Mutiti, S., J. Levy, C. Mutiti, and N.S. Gaturu, 2010. Assessing study of the Cross River State, southeastern Nigeria, ground water development potential using Landsat imagery, Hydrogeology Journal, 6:394–404. Ground Water, 48(2):295–305.

518 May 2011 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING Robinson, C., F. El-Baz, and V.H. Singhroy, 1999. Subsurface NASA Landsat Program, 2004. Landsat ETM scene imaging by RADARSAT: Comparison with Landsat TM data and L7CPF20011001_20011231_05, L1G, USGS, Sioux Falls, South implications for ground water in the Selima area, Northwest Dakota, 13 November 2001. Sudan, Canadian Journal of Remote Sensing, 25(3):268–277. Office of Cadasters and Natural Resources Inventory, 1971. Geologi- Rogers, R.D., 2003. Jurassic-Recent Tectonic and Stratigraphic cal Map. Impreso por el Instituto Geográphico Nacional History of the Chortis Block of Honduras and Nicaragua 1:50,000 Geological Map of Boaco, area 3053 II, series E751, (Northern Central America), Ph.D. dissertation, University of Managua, Nicaragua. Texas, Austin, Texas, 289 p. Paganelli, F., E.C. Grunsky, J.P. Richards, and R. Pryde, 2003. Use of Sander, P., 2007. Lineaments in groundwater exploration: a review RADARSAT-1 principal component imagery for structural of applications and limitations, Hydrogeology Journal, 15:71–74. mapping: A case study in the Buffalo Head Hills area, northern central Alberta, Canada, Canadian Journal of Remote Sensing, Sander, P., T.B. Minor, and M.M. Chesley, 1997. Ground-water 29(1):111–140. exploration based on lineament analysis and reproducibility tests, Ground Water, 35(5):888. Paradella, W.R., A.R. dos Santos, R. Dall’Agnol, R.W. Pietsch, and M.V. Sant’Anna, 1998. A geological investigation based on Sawaya, K.E., L.G. Olmanson, N.J. Heinert, P.L. Brezonik, and airborne (SAREX) and spaceborne (RADARSAT-1) SAR M.E. Bauer, 2003. Extending satellite remote sensing to local integrated products in the central Serra Carajá Granite area, scales: Land and water resource monitoring using high- Brazil, Canadian Journal of Remote Sensing, 24(4):376–392. resolution imagery, Remote Sensing of Environment, 88:144–156. Park, Y., K. Lee, and J. Kim, 2000. Effects of highly permeable geological discontinuities upon groundwater productivity and USGS and Japan ASTER Program,2007. ASTER scene well yield, Mathematical Geology, 32(5):605–615. AST_L1B_00311242005161636_200707120941 42_26205, Level 1 B, USGS, Sioux Falls, South Dakota, 11 November 2004. Radarsat International, 1996. Radarsat Geology Handbook, Vancou- ver, British Columbia, Canada, 56 p. Weyl, R., 1980, Geology of Central America, Second edition, Science Publishers, Germany, 371 p. Ricchetti, E., 2002. Structural geological study of southern Apennine (Italy) using Landsat-7 Imagery, Proceedings of the International Yun, S., and W.M. Moon, 2001. Lineament extraction from DEM Geoscience and Remote Sensing Symposium, 1:211–213. using drainage network, Proceedings of the International Geoscience and Remote Sensing Symposium, 09–13 Ricchetti, E., and M. Palombella, 2005. Application of Landsat-7 July, 5:2337–2339. ETM imagery for geological lineament analysis of Southern Italy, Proceedings of the International Geoscience and Remote (Received 20 April 2010; accepted 22 July 2010; final version 10 Sensing Symposium, 25–29 July, pp. 5200–5203. January 2011)

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING May 2011 519