Northeastern Brazil Groundwater Project

Interpretation of LANDSAT Digital Images for Groundwater Exploration in Crystalline Terrain, Ceará

Introduction
This paper presents an approach to the interpretation of Landsat images for hydrological studies and groundwater exploration in the semi-arid crystalline bedrock/fracture-controlled aquifer environment of northeastern Brazil. It examines seasonal variations in the abundance of vegetation (biomass) in an attempt to identify water-bearing bedrock structures. Comparison is made with geological/structural and airborne geophysical data.  

Characteristics of the Study Area and 
Available Information

Geology
Northern Ceará is underlain by Precambrian crystalline rocks without significant alteration cover. Groundwater is confined primarily to fracture zones of various orientations within these rocks, and in alluvial deposits along stream channels. Groundwater in the bedrock fractures is slightly saline (between 1,000 and 5,000 mg/litre TSD), due mainly to the presence of sodium chloride. A recent 1/100,000 scale bedrock geology map (Souza Filho et al, 1990) shows that the bedrock is made up of a variety of Proterozoic granitoids and high-grade metasediments. Lineaments and other geological data were extracted from this map to produce a transparent overlay to be used with the other maps and the Landsat images. 

Topography and Landscape
There were no detailed topographic maps available for this study, only a coarse terrain map taken from an atlas.   A digital elevation map is also available for the area covered by the airborne geophysical survey (see below). The landscape of northern Ceará varies from flat and gently rolling, to rugged in areas of granitic intrusive rocks. A large circular granodioritic pluton in the north-central part of the study area (herein after called the north-central massif) is a good example of this more rugged topography (see photos below).  

Typical Landscapes in Study Area 
(all three photos taken April 2003, at the end of the rainy season)
Left: west of Itapajé, the more rugged terrain of the north-central massif area
Middle: north of Juá looking north towards north-central massif 
Right: Juá reservoir at full capacity
(click thumbnails to enlarge)

Landsat Images
Two Landsat-7 digital images were acquired (ETM+ Bands 1-5, 7 (30 x 30m); 6, 9 (60 x 60m); 8 (15 x 15m), Path/Row: 217/63, Date: 7 Oct 1999 and 16 Apr 2000). Those were the two best images available, with minimal cloud cover, taken at the end of the dry and wet seasons respectively. Choosing those dates allowed us to monitor vegetation changes between consecutive wet and dry seasons. In Ceará, the dry season occurs during the months of May to November while the rainy season normally extends from December to April.  Background information on climate (precipitation), land use and vegetation at the time the images were captured was not available. This information would have been very useful to help interpret the LANDSAT images, particularly to monitor the differences in the vegetation cover between the October and April scenes.  The two figures below show the entire scenes of the two Landsat images; the study area is outlined in the northwest corner. Image processing was done using ER Mapper geographic image processing software. 

The two Landsat Scenes Used in this Study
Left: October 7, 1999 (end of dry season)
Right: April 16, 2000 (end of wet season)
Note the city of Fortaleza, capital of Ceará 
(pop approx. 2 M) in the NE corner
(click thumbnails to enlarge)

Vegetation
Vegetation types and abundance vary according to the amount of moisture in the soil.  During the wet season, thin soils and water-bearing structures support abundant vegetation throughout the region. During the dry season, the thin soils dry up, but the water-bearing geological structures continue to provide moisture to the overlying soils and to support vegetation. And since these structures support vegetation throughout the entire year, the type of vegetation that grows in their vicinity is different and generally more developed than the vegetation that grows in the thinner soils. 

Airborne Geophysical Data
Part of the study area, about 140km² centered on the community of Juá, was covered in March 2001 by a helicopter-borne magnetic-electromagnetic (frequency domain) survey done to test the technique for water exploration.  Some of the results are presented and discussed elsewhere on this site. The total magnetic, apparent conductivity and digital elevation maps are presented below.  The electromagnetic results are especially interesting as they define conductive lineaments interpreted as saline groundwater-bearing fractures in the bedrock. 

Airborne Geophysical Maps, Juá Area
Left:  Total Magnetic Field
Middle: Apparent Conductivity (4,500 Hz)
Right: Digital Elevation
(click thumbnails to enlarge)

Water Well Database
An excellent interactive water well database, know as the Groundwater Atlas of the State of Ceará, was recently assembled by CPRM and is now available on CD. This allowed for additional correlations to be made between  the satellite images and certain water well parameters such as locations and yields, providing a means of testing our interpretation of the remotely sensed data.  The following figure shows the locations of all the wells within the Landsat image grouped according to their yields.

Water Well Locations and Yields, plotted 
on Landsat Image, Northern Ceará
(click thumbnail to enlarge)

Methodology

The study is based on three factors that can reveal the presence of groundwater using remote sensing imagery, within the semi-arid context of northeast Brazil. These are : 

• the influence of bedrock geology (lithology and structure) on groundwater;

• the effects of seasonal variations on groundwater; and 

• the influence of groundwater conditions on vegetation.

Remote Sensing Analysis

Image Rectification and Enhancement
The Landsat images were georeferenced (UTM projection, WGS 84, SUTM 24) and registered to the bedrock and geophysical maps. The figures below show georeferenced, histogram enhanced, Landsat images of the study area. Bands 2,3 and 5 were used in order to enhance the images based on vegetation differences (see Biomass Index below for explanation).

Enhanced Landsat Images of Study Area; 
Histogram enhancement of TM bands 2,3, and 5 
Left: April 16, 2000; Right: October 7, 1999
(click thumbnail to enlarge)

Comparison between the October and April Enhanced Images
The first striking difference between the two enhanced Landsat images (above) is the higher discrimination of land cover, both natural and anthropogenic, and the enhancement of geological lineaments on the October image (dry season) compared to the April one (wet season).  On the April image (left above), vegetation covers most of the ground reflecting the availability of water in the soil nearly everywhere. Variations in the spectral signature of the vegetation is too subtle to allow to differentiate the various types of vegetation, thus masking any differences in the moisture retention capacity of soils and bedrock structure.

On the October image (right above), vegetation stress due to the lack of moisture at the end of the dry season enhances the differences in the density and types of vegetation between areas where water is available and areas that are dry.  Since the availability of water during the dry season is to a large extent related to the presence of groundwater in bedrock structures, these structures are enhanced on the October image. 

Biomass Index
Biomass index derived from Landsat images is based on the chlorophyll production of vegetation.  There is an inverse relationship between red radiance (TM Band 3) and chlorophyll production (green biomass) and a direct relationship between near-infrared (TM Band 5) and chlorophyll production. The method used here to enhance the biomass (Normalized Difference Vegetation Index, NDVI) uses the formula: (Band 5 - Band 3)/(Band 5 + Band 3) which gives spectral values between -1 and 1. Normalizing spectral signatures enables the comparison of different images in order to monitor vegetation changes over a period of time. In this case the NDVI will monitor vegetation changes between the October and April Landsat scenes.

Biomass Index (NDVI) of Study Area
Left: April 16, 2000; Middle: October 7, 1999
Right: Difference between April and October
(click thumbnail to enlarge)

• As observed on the enhanced images, the vegetation cover (biomass) is more uniform on the April image (Left above) than on the October one (Middle above). 

• The density of biomass on the October image seems closely related to the availability of water such as at the bottom of valleys, along rivers, on possibly along bedrock fractures. 

• On the biomass difference image (Right above), the areas in blue indicate higher biomass in April than in October whereas areas in red indicate equal biomass in October and in April.  Consequently, areas represented by reddish colours on the biomass difference image have a higher capacity to retain water either in the soil or as a result of groundwater seeping into the soil from underlying bedrock fractures. 

False Biomass Readings
due to clouds and shaded hillsides
Caxitore River Area
Left: April; Right: October
(click thumbnails to enlarge)

• Areas of clouds on the April image (Left) should be excluded from the interpretation of biomass changes as the NDVI is altered by the presence of clouds giving false higher biomass in October. 

• Shaded hill sides should also be excluded as they can also produce erratic results. 

Comparison between the Geological Map and Remote Sensing Information

Features on geological maps (geological contacts and structures) differ from features extracted from satellite images (polygons and lineaments), which in turn differ from features on airborne geophysical maps (anomalies and patterns) because the technologies involved in  producing these documents are different. Each of these technologies is affected differently by certain factors such as the presence of thick overburden masking bedrock information or the presence of saline groundwater producing geophysical anomalies that may be reflecting bedrock or overburden aquifers. 

In most cases, these documents and technologies are complementary. Comparison between them, as is presented here, is preliminary and should be followed by ground investigations (ground truthing) by geoscientists who are familiar with the geology and geobotany of the study area. 

Since aquifers in Ceará are confined to bedrock fracture zones, geological barriers and alluvial deposits, the geological map will be compared to the remote sensing images to see if the type and structure of the geological formations can be revealed by satellite imagery and/or airborne geophysical data, and to see if there is any relationship between vegetation patterns and possible water-bearing geological structures.

Geological Overlay Draped on Landsat and Biomass Images
Upper Left: October Landsat Image;  
Upper Right: October Biomass Index Image
Lower Left: October Landsat Image of Rio Caxitore Area
Lower Right: October Biomass Index Image of Rio Caxitore Area
(click thumbnails to enlarge)

The transparent overlay produced from the geological map is overlaid on the Landsat TM October image (upper and lower left above), the Biomass Index of the October image (upper and lower right above), the Biomass Difference between October and April images (below), and geophysical images of the Total Magnetic Field and the Apparent Conductivity (see below). The April images were not used because they do not reveal as much information as the October ones.

There is a conspicuous correspondence between the Landsat Image (upper left above) and the geological map, mainly in areas of hilly bedrock outcrops (purple to dark green areas).  Identification of bedrock types and fractures is more difficult in flat-lying areas but general trends can be identified as seen on the close-up portion, centered on Rio Caxitore area (lower left above). Interpretation of east-west trending lineaments and patterns on the Landsat image and their correspondence with bedrock lineaments on the geological map would have to be verified in the field, but initial observations indicate that a considerable amount of information can be extracted from the Landsat image and, possibly, some lineaments (fractures) on the geological map could be repositioned based on the regional patterns revealed on the Landsat image.

The geology overlaid on the October Biomass Index image (upper right above) tends to confirm the observation made earlier, that denser vegetation is located preferentially in river valleys and along bedrock fractures (which often coincide). A close-up of the Landsat image, centered on Rio Caxitore, confirms these observations (1 and 2 on lower right figure, above), but also shows other bands of denser vegetation (3 and 4) which are not explained by the geological map. Field verification will be necessary to explain these bands of denser vegetation, as they could correspond to either water-bearing fractures or elevated water table in unconsolidated sediments behind groundwater barriers.

Geological Overlay Draped on 
Biomass Index Difference Image
(click thumbnail to enlarge)

The Biomass Difference image (above) shows similar correspondence with the geological map as the October Biomass Index image with minor variations. Detailed study of the Biomass Difference image could possibly reveal information not apparent on the October Biomass Index image, but the climatic conditions at the end of the dry season seem to create enough vegetation stress, in northern Ceará, to provide most of the information required for the remote sensing analysis.

Geological Overlay draped on Total Magnetic (upper) 
and Apparent Conductivity (lower) Maps
(click thumbnails to enlarge)

The total magnetic field and the shaded magnetic relief maps (azimuth of 315°;elevation of 45°) (upper set above)  show regional trends that generally correspond with those on the geological map. Similarly, the apparent conductivity maps (standard and shaded relief, lower set above) show that some lineaments correspond to geological contacts. According to an interpretation presented elsewhere on this site, the majority of lineaments on the apparent conductivity map, especially those trending N and NNW, do not appear to be lithologically controlled and are probably related to groundwater filled fractures. Groundwater in much of the Northeast of Brazil is characterized by an elevated salt content and behaves as a conductor in an induced electromagnetic field.  Ground follow-up in 2002 and 2003 has confirmed the patterns and features on the airborne maps, and a few test wells have been drilled on selected targets, generally corroborating this interpretation.  

It should also be noted that many of the faults and fractures reported on the bedrock geology map are not apparent on the apparent conductivity maps. Assuming the hypothesis that all fractures in the bedrock are filled with saline groundwater, then why aren't these fractures showing on the apparent conductivity maps? Similarly, most of the north-south lineaments on the apparent conductivity maps are not represented on the geological map. Why? On the other hand, many of the east-west trending faults and fractures reported on the geological map do show on the total magnetic field map but are not apparent on the apparent conductivity map. There are, no doubt, valid explanations for these observations (e.g. older fractures being filled with quartz or other non-conductive material), but it remains that the lack of correspondence between the various maps and images is intriguing and should be investigated further in the field, and possibly by drilling.   

Shaded Apparent Conductivity Map draped on: 
Left: October Landsat TM Image; Right: October Biomass Index Image
(click thumbnails to enlarge)

There is some correspondence between structural features on the geological map and on the airborne geophysical maps, but many of the lineaments on the apparent conductivity map remain unexplained. Landsat images are now compared to the geophysical data to see if the lineaments apparent on the geophysical maps are reflected on the Landsat images.

There is a strong correlation between the October Landsat image and the apparent conductivity map (left above). The correspondence seems to be associated with river valleys (A on left figure above) and lithology, faults and fractures (B). Some north-south lineaments that are apparent on the apparent conductivity map are also apparent on what seems to be thin overburden areas (C, dark green and purple tones) on the Landsat image, but disappear in areas that seem to be thicker overburden (medium-to-pale green tones); others (D) can be traced with slight shift even in thicker overburden areas. (The correspondence is not as clear on a "fixed" image as presented here compared to when viewed interactively in a GIS by turning the draped apparent conductivity map on and off.)

Comparison of the apparent conductivity map with the biomass index of the October Landsat image (right above) does not seem to provide any additional information, from what is visible on the 3-band Landsat image (left above).

Total Magnetic Field draped on October Landsat Images
Left: Shaded TMF; Middle: Unshaded TMF; Right: TMF on Biomass Index
(click thumbnails to enlarge)

There is little correspondence between the shaded total magnetic field map and the Landsat image (left above). Correspondence is more apparent on the non-shaded version of the total magnetic field (middle above), but seems of limited use in locating groundwater aquifers in overburden or bedrock fractures. Similarly, the biomass image shows little correspondence with the total magnetic field map (right above).

Location of Existing Water Wells in Relation to the Geology

Plot of Existing Groundwater Wells:
Left: on Geology; Middle: on Landsat Image; 
Right: on October Biomass Index
(click thumbnails to enlarge)

In order to determine if there is a relationship between the location of existing water wells and their yields and bedrock geology and structure in the study area, water well locations and yields were plotted on the bedrock geology map (above left), the apparent conductivity map (below left) and the total magnetic field map (below right). Well locations and yields were also plotted on the October Landsat image (above middle) and on the October biomass index images (above right) to see if there is any rapport between well locations and yields and visible features on the Landsat images as, for example, a correlation between vegetation density and well yield. On all these maps and images, the wells are plotted with proportionate circles size according to their yield (1 to 100,000 litres per hour) but only the order of magnitude of the yields are relevant because the study is essentially qualitative. The following observations are made: 

• There is no obvious correlation between well locations or yields and bedrock lithology or structure (above left), apparent conductivity (below left), or total magnetic field (below right). This, however, should not be construed as indicating that structural mapping, either from field observations, satellite imagery/air photographs, or airborne geophysics does not have a role to play in locating high yielding wells. In fact, most of the wells in this study are unlikely to have been spotted to target specific mapped structures and, without detailed field investigations, it is not possible to determine which of the wells, if any, has intersected a mapped structure. Furthermore, as most wells were drilled in river valleys, their yields are greatly affected by the thickness of alluvium and their recharge potential, making it virtually impossible to establish a correlation between yields and the presence or absence of mapped structures. 

• The October Landsat image (above middle) confirms that wells are mostly located along river valleys, which may correspond with fracture zones in the bedrock, and to a lesser extent along roads. However, it was not possible to establish whether there is a significant correlation between spectral signatures and well yields because the sample size (i.e. number of wells) is too small.

• The biomass index image (above right) shows that all wells are located in areas of high vegetation index. This strong relationship is due to the coincidence of well locations and river valleys. Once more, it was not possible to establish whether a significant correlation exists between well yields and the biomass index because of the small sample size (i.e. insufficient number of wells).
   

Plot of Existing Groundwater Wells:
Left: on Apparent Conductivity; Right: on Total Magnetic Field
(click thumbnails to enlarge)

 last modified: 2004-10-01