Remote Sensing of Environment 106 (2007) 350 – 359
www.elsevier.com/locate/rse
Detection of geothermal anomalies using Advanced Spaceborne Thermal
Emission and Reflection Radiometer (ASTER) thermal infrared
images at Bradys Hot Springs, Nevada, USA
M.F. Coolbaugh ⁎, C. Kratt, A. Fallacaro, W.M. Calvin, J.V. Taranik
Great Basin Center for Geothermal Energy and the Arthur Brant Laboratory for Exploration Geophysics, University of Nevada, Reno, 89557 USA
Received 2 February 2006; received in revised form 30 August 2006; accepted 3 September 2006
Abstract
Surface temperature anomalies associated with geothermal activity at Bradys Hot Springs, Churchill County, Nevada were mapped using
Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) thermal infrared (TIR) image data. In order to highlight subsurface
contributions of geothermal heat, the ASTER images were processed to minimize temperature variations caused by the diurnal heating effects of
the sun. Surface temperature variations caused by changes in albedo were corrected with visible and near-infrared ASTER bands, and a 10-metersmoothed Digital Elevation Model (DEM) was used to correct for topographic slope effects. Field measurements of ground surface temperatures
made over 24-hour periods were used to design a thermal inertia correction incorporating day and night thermal infrared images.
In the resulting processed image, background temperature variations were reduced 30–50% without reducing the intensity of geothermal
anomalies, thus making it easier to distinguish geothermal activity from ‘false’ anomalies caused by non-thermal springs, topographic effects, and
variable rock, soil, and vegetation compositions.
© 2006 Elsevier Inc. All rights reserved.
Keywords: Thermal infrared; Geothermal; ASTER; Bradys; Nevada; Great Basin
1. Introduction
This paper investigates the ability of ASTER images to detect
surface temperature anomalies associated with geothermal activity
(hot springs and fumaroles) in the Great Basin of the western United
States. The Great Basin is a region of internal drainage
characterized by active faulting, crustal extension (Stewart,
1983), and high crustal heat flow (Blackwell, 1983), and contains
a number of high-temperature (N 150 °C) geothermal systems with
an installed electric power plant capacity of nearly 600 MW. The
Great Basin (Fig. 1) is well suited for remote sensing exploration for
geothermal resources because geothermal systems occur over a
large area, there is relatively sparse vegetative cover in an arid
environment, and because deep water tables sometimes impede the
formation of hot springs that otherwise would signal the presence of
subsurface geothermal activity.
⁎ Corresponding author. 1850 Prior Road, Reno, NV 89503, USA.
E-mail address: mfc@unr.nevada.edu (M.F. Coolbaugh).
0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2006.09.001
Remotely sensed thermal infrared (TIR) images have been
used for years to detect geothermal activity (Allis et al., 1999; Lee,
1978), but the success of those efforts in some cases has been
limited by the difficulty in modeling the diurnal heating effects
caused by the sun. An example of where this is a challenge is the
main sinter terrace at Steamboat Springs, NV, where a conventional pre-dawn thermal image does not detect a thermal anomaly
(Fig. 2a) in spite of numerous fumaroles being present. The
terrace has a relatively high albedo and reflects much of the sun's
energy during the day. It has a low thermal inertia because of its
high porosity and a currently low water table, and consequently
cools off quickly at night. A recent study by Coolbaugh et al.
(2000) demonstrated that it was possible to compensate for the
effects of albedo and thermal inertia to reveal the underlying
thermal anomaly (Fig. 2b) at the terrace. It should be noted in
passing that although hot springs and geysers were present at this
terrace as recently as 1987 (Koenig, 1989), the water table has
since dropped to at least 30 m below the surface (Platt, personal
communication, 2002) as a result of water withdrawal from
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
351
This area contains several geothermal occurrences, one of which, at
Bradys Hot Springs, has an extensive surface expression of
fumaroles, mud pots, and warm ground (Fig. 4) that follow surface
traces of Quaternary faults for a distance of 4 km.
The Hot Springs Mountains and the Truckee Range to the
west (Figs. 3 and 4) are faulted horst blocks consisting of
Tertiary basaltic and andesitic volcanic rocks interbedded with
lacustrine sediments including diatomite. Intervening valley
grabens (some of which are labeled as “sinks” in Fig. 3) contain
colluvium, alluvium, sand dunes, and playa evaporite deposits.
Vegetation in hills and mountains consists of relatively sparse
sage and salt brush: grass is locally present in wet areas or
marshes.
2. Material studied and software used
Fig. 1. Location map of ASTER and Steamboat Springs study areas. Grey circles
are known geothermal systems with measured or calculated temperatures of
100 °C or greater.
nearby wells. Steam still rises along open fissures to form weak
fumaroles, but observations by the authors confirm that the highly
porous sinter is essentially dry where observed within 1–2 m of
the surface, thus helping to explain its low thermal inertia.
The initial success of the Steamboat study, which used the
Thermal Infrared Multispectral Scanner (TIMS) from an airborne
platform, encouraged efforts to validate and refine the techniques
using satellite-derived images. ASTER images were acquired over
the Hot Springs Mountains in west-central Nevada (Figs. 1 and 3).
Day and night ASTER scenes were acquired on the same date,
Aug. 31, 2001, with the assistance of the Jet Propulsion
Laboratory (JPL) in Pasadena, CA. Weather was dry and hot,
with maximum temperatures near 35 °C and little wind; a light
haze was present due to distant forest fires in California and
visibility was approximately 60 km. Skies were clear prior to and
during the morning ASTER flyover, but isolated cumulus clouds
locally developed in the late afternoon. No rain fell, but the clouds
reduced the amount of late afternoon sun received in some areas.
After sunset, cloud cover increased from 20% to nearly 100%, but
then largely dissipated 1 h before the nighttime ASTER flyover.
Scattered clouds visible on the nighttime image were avoided
when the thermal anomaly algorithms were designed.
Preprocessed digital versions of the ASTER images were
downloaded from the EROS data center at http://edcdaac.usgs.
gov/asterondemand/index.html: those versions include the
ASTER higher-level data products AST07 (surface reflectance)
Fig. 2. Before and after enhancements of Thermal Infrared Multispectral Scanner (TIMS) images at the Steamboat Springs main sinter terrace. Darker shades denote
temperature anomalies. A pre-dawn thermal image (a) does not detect an anomaly at the Main Terrace. After processing to compensate for the cooling effects of high
albedo and low thermal inertia, a temperature anomaly related to geothermal activity is revealed (b).
352
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
Fig. 3. Color composite RGB image (converted to grayscale) of ASTER bands 3, 2 and 1 for the Bradys Hot Springs study area.
and AST08 (surface kinetic temperature). A discussion of the
ASTER data products is provided by Abrams (2000) and Abrams
and Hook (2002). At the time of download (Aug. 9th, 2002) these
were partially validated data products (version 3). Both products
had been preprocessed to correct for atmospheric absorption
effects; in addition, the AST08 product had been preprocessed to
convert radiance temperatues to kinetic temperatures using the
algorithms of Gillespie et al. (1998).
Topographic slope aspects were modeled using United States
Geological Survey (USGS) DEMs with 10-meter cells; these
were downloaded from the Geocommunity GIS Data Depot at
http://www.gisdatadepot.com/. USGS digital raster graphic
(DRG) topographic maps were used for georeferencing. Image
processing was performed using ENVI® v. 3.5 software.
2.1. Data quality and georectification
The daytime ASTER images appear of good quality, but a
slight haziness in the atmosphere was not completely removed in
the preprocessing stage, and field spectrometer measurements
were used to correct the images (Section 3.3.3 below). In a few
restricted areas of very high surface reflectance in several
diatomite pits and a couple small playas, ASTER bands 1 and 2
reached saturation levels, but because these pixels are limited in
Fig. 4. Fumaroles at Bradys Hot Springs. The view is looking southwest and the Truckee Range is in the background. Location of the photograph shown in Fig. 3.
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
number and extent, this does not appear to have significantly
affected the final processed image.
The nighttime images contain erratic pixels with low digital
values and some stripping effects. No attempt was made to filter
those effects from the imagery, to avoid degrading image quality
in areas where the effects were less evident. Low-value pixels are
most abundant in the western part of the image at high elevations
and are rare in the main area of interest in the Hot Springs
Mountains where the processing algorithms discussed in this
paper were optimized. Subtle stripping is widespread but
relatively uniformly developed. The image subsets used for
optimizing the processing algorithms included many stripes so
that local variations in stripping-related radiance values would not
significantly affect the statistics.
Geo-registration control points were identified in the field with
GPS units. Good control points were easily obtained for the
daytime images, but registration of the nighttime TIR images was
more difficult because of its 90 m pixel size, and only 4 control
points could be matched with the images. Registration error for
the night TIR data is estimated at 40–90 m in the central portion of
the study area including the Hot Spring Mountains, and near the
margins of the images, it locally reaches 90–120 m. Parallax
could contribute as much as 40 m of registration error at higher
elevations of the Truckee Range (Figs. 3 and 4) in the western
portion of the images.
3. Methodology
A land surface energy balance equation (Bastiaanssen et al.,
1998) can be used to model heat inputs to the ground surface from
different sources (the terms of that equation are rearranged here).
Under equilibrium conditions:
0 ¼ Q*−H−kE−G0
ð1Þ
where G0 is soil heat flux, Q* is net radiation, H is sensible heat
flux, and λE is latent heat flux (the energy of photosynthesis is
ignored). As defined here, sensible heat is heat lost from the
ground to the atmosphere by conduction and convection processes. Latent heat is heat lost through the evaporation of water.
Conceptually, ground surface temperatures can be modeled by
integrating this equation over time under non-equilibrium conditions, if the effects of subsurface conduction, convection, and heat
capacity are considered. Modeled temperatures can be compared to
TIR remote sensing measurements of surface temperature, and
anomalous differences between predicted and measured temperatures could point to geothermal heat sources.
Because of the difficulty in remotely mapping sensible and
latent heat flux, and because those terms will have relatively less
influence in the dry, desert environment being investigated, those
terms were not quantitatively modeled here (although the
qualitative effects on the processed images are discussed later).
Instead, efforts focused on modeling the effect of net radiation
flux (Q*) on surface temperatures.
Different surface materials with different physical properties
such as thermal inertia, albedo, emissivity, and moisture content
respond differently to solar radiation, resulting in different surface
temperatures at most times of a 24-hour (diel) day (Elachi, 1987;
353
Watson, 1973). Even during pre-dawn hours, appreciable differences in temperature persist due to the differential heating affects
of the sun the previous day. These temperature differences can
obscure underlying contributions of geothermal heat.
Quantitative theoretical modeling of the physical variables to
predict surface temperatures involves differential equations and
LaPlacian transformations that require iterative numerical solutions (Elachi, 1987; Kahle, 1977; Watson, 1973). A more practical
and empirical approach was adopted here that, although employing a number of assumptions, has the advantages of being easier to
understand, implement, and interactively monitor. It involves
minimizing temperature variances due to each of a number of
surface physical properties in a step-by-step process, as described
below.
3.1. Emissivity
Surface radiant temperatures are in part a function of surface
emissivity; low emissivities reduce the radiant temperatures measured with TIR sensors, making surfaces appear cooler in uncorrected radiant temperature images than they really are
(Sabins,1978). The presence of 5 thermal bands on ASTER
makes it possible to identify wavelength-dependent variations in
emissivity so that true kinetic temperatures can be estimated
(Hook et al., 1999). The higher-level data product AST08 (surface
kinetic temperature) provided by EROS and used in this study
employed the temperature–emissivity separation algorithm of
Gillespie et al. (1998).
Field measurements of temperature using soil thermocouples
were used to check the accuracy of the AST08 kinetic temperature
images at two playas (Fig. 3) at the time of the ASTER flyovers.
At the western site, thermocouple temperatures were respectively
4 °C and 3 °C lower than the corresponding day and night images,
whereas at the eastern playa, thermocouple temperatures were
respectively 2 °C lower and 0.6 °C higher than the corresponding
day and night images. Thermocouple measurements of known
sources (ice) compared with standard mercury thermometer measurements suggest thermocouple temperature accuracies ranging
from 0° to 2° at the temperatures measured. Thermocouple
Fig. 5. Field-measured diel temperature curves at two sites used for thermal
inertia adjustment. Measurements at site 1 were taken Aug. 15, 2001 with
thermocouples; average diel temperature for sandy soil was 37 °C. Measurements at site 2 were taken Oct. 18, 2001 with an Omegascope® hand-held
infrared thermometer; average diel temperature for bare soil was 13 °C.
354
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
Fig. 6. Thermal inertia variance curves for field sites 1, 2, and the ASTER day
and night temperature images. Error (variance) on the left-hand y-axis measures
ability of weighted day and night field temperatures taken at 12:00 pm and
10:00 pm to match the average field-measured 24-hour temperature difference of
surface materials. The sum of the day and night weighting factors equals 1.
and night AST08 temperature images together using different
weighting factors. Those weighting factors were systematically
varied in order to identify which weighting factors (which must
sum to 1) produced a combined day–night image with the least
variance (Fig. 6). Weighting factors of 0.76 for the night image and
0.24 for the day image produced the least variance in the combined
image (23% less variance than the corresponding night image); a
result which is in close agreement with the first approach. (where
thermal inertia effects are minimized in the combined temperature
scene, then one source of temperature variation would be
minimized, so that the overall variance should be at a minimum).
Although these weighting factors appear optimal for this particular
study, they may not be directly applicable to other areas or even to
this same area at different times of year, and it would be important to
independently estimate these parameters in any other studies.
3.3. Albedo and topographic slope
temperatures were measured at a 0.5 cm depth, and differ from
those at the surface measured with a radiometer by up to 3 °C,
depending on the time of day. We consider the difference between
surface and ASTER temperatures consistent within the measurement certainty, especially given the large (90 m) footprint of
ASTER compared to the size of the surface area measured with
the thermocouples (3 m).
3.2. Thermal inertia
The thermal inertia of a material is a function of its heat capacity,
density and thermal conductivity. If other factors are equal, mean
24-hour surface temperatures will be the same for materials with
different thermal inertias (Elachi, 1987). Therefore, if mean 24-hour
temperatures can be estimated, the effects of thermal inertia can be
minimized. If TIR images are available at the times of maximum
and minimum temperatures (near mid-day and pre-dawn respectively) on the same date, the average of those two temperatures can
be used to approximate mean temperatures (Coolbaugh et al.,
2000). However, this average is not exactly correct, because 24hour thermal inertia curves are not symmetrical (Elachi, 1987). The
situation becomes more difficult with the ASTER images because
the overflight times are at roughly 12:00 pm (noon) and 11:00 pm
(daylight savings time, Nevada).
Two approaches were used in this study to estimate mean 24hour temperatures. The first approach involved measurements of
temperatures at two field locations, each of which contained
materials with contrasting thermal inertias (Fig. 5). Temperatures at
noon and 11 pm (the ASTER flyover times) were combined using
different weighting factors, to see which combination predicted the
actual 24-hour mean temperatures with the least error. For both
sites, residual 24-hour temperature differences (errors) were minimized where a weighting of 0.75 times the night temperature
(11 pm) and 0.25 times the day temperature (noon) was used (a ratio
of 3 to 1, Fig. 6). Both sites yielded similar results even though
measurements were taken at different times of the year, with
different average temperatures, and with different surface materials
(Figs. 5 and 6).
For corroboration, a second, perhaps simpler, method of estimating weighting factors was used, which involved adding the day
The objective of this step was to minimize temperature variations caused by differences in surface albedo and topographic
slope. The first step was the creation of a radiation heat image
using visible and near infrared bands from ASTER as measurements of reflectivity, and a DEM from which topographic slope
aspect could be calculated. The DEM was also used to convert
ASTER measurements of reflectivity to albedo. We then scale this
heat image to an approximate surface temperature change caused
by that heat input. An equation from Watson (1973, p. 5908)
representing net surface radiation flux was used as a starting point:
Q* ¼ FSn þ FAn −FGn
ð2Þ
where Q* is the net surface flux at the surface, which equals FSn,
the absorbed solar flux, plus FAn, the absorbed sky radiation,
minus FGn, the re-emitted ground radiation. The mathematical
terms on the right of the equation are represented as follows
(Watson, 1973):
FSn ¼ f S0 ð1−AÞM ðZÞcosZ V
ð3Þ
FAn ¼ erTA4
ð4Þ
FGn ¼ erVn4
ð5Þ
so that
Q* ¼ f S0 ð1−AÞM ðZÞcosZ Vþ erTA4 −erVn4
ð6Þ
where f is a cloud cover factor; S0 is a solar constant; A is the
ground albedo; M(Z) is atmospheric transmission as a function of
the zenith angle Z; cosZ′ is the cosine of the angle between the
surface normal and the sun's rays; ε is the spectral average
emissivity of the ground; σ is the Stefan–Boltzmann constant, TA
is the effective sky radiance temperature; and Vn is the ground
surface temperature.
3.3.1. Heat energy model: simplification
In its current form, Eq. (6) is difficult to solve because of the
temperature dependency of the terms absorbed sky radiation
(εσTA4) and re-emitted ground radiation (εσVn4). Under cloud free
and low humidity conditions, the effective sky radiance
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
355
temperature is typically −30 °C during the day (measured). As
such the second term does not contribute a significant input to
surface heating and is ignored. The third term, surface cooling,
depends on surface temperature and emissivity. Local temperature
differences on the ground are on the order of just a few degrees in
most areas, including geothermal areas, so we make the approximation that cooling effects are not changing rapidly. Similarly, it is
assumed that surface variations in overall emissivity are relatively
minor, so that the primary cause of surface temperature changes is
the solar contribution. Thus, under cloud free conditions Eq. (6)
simplifies to:
Q⁎ ¼ S0 ð1−AÞM ðZÞcosZ V
ð7Þ
3.3.2. Topographic slope correction
The influence of topographic slope aspect on the heat flux Eq.
(7) is represented by the term cosZ′, which is the cosine of the
angle between the surface normal and the sun's rays. An ENVI®
software “shaded relief” option is available that calculates this
value directly from a DEM after the sun angle has been determined from input parameters of date, time, latitude and longitude.
3.3.3. Albedo calculation
Under ideal conditions of flat terrain and a normal atmosphere,
AST07 surface “reflectance” values are equivalent to albedo and
could be directly employed in Eq. (7) above. But because much of
the study area is hilly or mountainous, image brightness or
reflectivity is affected by local topographic slope orientation. To
derive albedo, the AST07 images were recalibrated using a
modified form of the empirical line method (Kruse et al., 1990) in
combination with a DEM and field-measured albedos obtained
with a spectroradiometer. This is similar to the method employed
by Gillespie and Kahle (1977). With the assumption that downscattered atmospheric radiance is negligible (Schowengerdt,
1997, p. 315), a simplified calibration equation can be written
as follows:
R ¼ Kw cosZ VðAÞ þ bw
ð8Þ
where R = AST07 surface “reflectance” (which is actually surface
reflectivity as treated herein) and Kw and bw are constants for each
spectral band “w”.
In Eq. (8), R and cosZ′ are known values, but Kw, A, and bw
are unknowns. This equation was solved for albedo (A) by directly
measuring albedo at four field locations with a FieldSpec®
spectroradiometer. The constants Kw and bw could then be
calculated using a best-fit linear equation of the form y = mx + b
(Fig. 7), where the product of cosZ′ times A is plotted on the xaxis against R on the y-axis. With the constants Kw and bw
determined, albedo can be calculated on an image basis using Eq.
(8), and this was done separately for ASTER bands 1, 2, and 3,
each with its own constants K and b. The weighted average
albedo was then computed for all 3 bands, using as weighting
factors the wavelength-dependent solar irradiance (at the top of
the atmosphere), atmospheric transmittance, and the band width.
If preprocessing had been successful in removing all
atmospheric absorption and scattering effects, the constant “K”
should be near 1 and the constant “b” should be near zero. Actual
Fig. 7. Relationship between slope, field-measured albedo, and AST07 band 3
reflectance. The best-fit line y = mx + b is used to calculate constants Kb (equal
to m, 0.888) and bb (equal to 0.092), which in turn are used with AST07
reflectance and a digital elevation model to solve for ground albedo.
values of K ranged from 0.78 to 0.89 and b ranged from 0.079 to
0.092. The significant positive values of “b” are attributed to
atmospheric haze caused by smoke from distant forest fires.
3.3.4. Heat energy model: approximate temperature change
An overall heat image was created by integrating the amount of
solar energy absorbed by each ground pixel over the course of a
day. In doing so, it is important to account for changes in the
intensity of light and changes in the position of the sun relative to
topographic slopes over the course of a day. Such integration can
be represented by a summation of M(Z) and cosZ′ values
weighted over discrete time intervals:
X
Eq ¼ S0 ð1−AÞ
½M ðZÞt cosZ Vt Dt &Dtt
ð9Þ
t
where Eq equals the solar heat energy absorbed per unit area over
the course of a day and Δtt represents the time interval for each
component of the sum. To this equation has been added the
variable “Dt”, a time “decay factor” that is a function of the
difference in time between a modeled position of the sun and the
time the images were acquired. The longer this time interval, the
less relevance the heat flux has on temperatures in the images,
and D was modeled as being inversely proportional to the
number of hours transpired between a given sun position and the
time of image acquisition: it ranges from 0 to 1. With this decay
factor the heat equation becomes a pseudo-temperature index,
and is roughly proportional to solar heat contributions made to
temperatures on the images. This method of accounting for heat
dissipation is admittedly highly simplified but the objective is to
find out if such a simple method will make it possible to remove
most topographic and albedo-related temperature phenomena
from the TIR images.
Heat energy images were constructed separately for the day
and night ASTER scenes, and each image is comprised of three
time-weighted products of M(Z), cosZ′, and Dt. Values for M(Z)
were taken from Finlayson-Pitts and Pitts (2000) based on latitude, longitude, time of day and time of year. The solar constant S0
was not factored into the images because it is a constant and is
implicitly accounted for when the heat energy images are later
rescaled and subtracted from the day and night temperature
images (see below).
356
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
removed by processing. Reductions in albedo-related temperature variations are illustrated by three open pit diatomite mines
(Fig. 10); topographic shading effects are illustrated by a series
of northeast-trending ridges (Fig. 11) and thermal inertia effects
are discernable at the sandy margin of the Hot Springs
Mountains (Fig. 12).
4.1. Bradys Hot Springs
Fig. 8. Variance of AST08 day and night temperature images after iterative
subtraction of solar heat images using different scaling factors. The minimum
variance identifies the scaling factor that minimizes topographic slope and
albedo effects in the ASTER temperature images.
3.4. Creation of final processed image
The day and night radiation heat energy (pseudo-temperature)
images were subtracted from their corresponding AST08 temperature images to better discern residual geothermal contributions. The heat energy images must be scaled properly so that
variations in their “pseudo-temperature index” match the actual
temperature perturbations caused by albedo and topographic
slope on the ASTER images. The correct scaling factor was
identified through iteration: the heat energy images were
multiplied by different constants and subtracted from the
ASTER temperature images until the best fit was identified. In
an earlier study at Steamboat Springs (Coolbaugh et al., 2000), the
optimal constant was determined visually by identifying the
image in which topographic shading effects most nearly disappeared. In the current ASTER study, statistics were used, by
choosing the residual image whose variance was at a minimum
(Fig. 8). This method was demonstrated empirically at Steamboat
(Fig. 2) and works as a rough approximation although conversion
of heat energy to temperature change is not rigorous.
After subtraction of solar heat energy from the day and night
AST08 temperatures, the resulting day and night images were
combined using the weighting factors for minimizing thermal
inertia (see Section 3.2) to produce a final temperature anomaly
image (Fig. 9a).
The geothermal anomaly at Bradys Hot Springs is visible in
both the final thermal image (Fig. 9a) and the single nighttime
image (Fig. 9b) with approximately equal clarity. Apparently the
distribution of surface albedo and thermal inertia properties are
such that they do not conceal portions of the anomaly in the
nighttime image, unlike Steamboat Springs (Fig. 2). However, the
amplitude of background temperature variations has been significantly reduced in the final thermal anomaly image (Fig. 9a), thus
reducing the number of “false” anomalies that could potentially be
mistaken for geothermal activity elsewhere in the image. By
reducing the number and strength of these anomalies, the field
time required to investigate potential geothermal occurrences is
dramatically reduced.
4. Results and analysis
A significant reduction in background temperature noise has
been achieved in the final processed image (Fig. 9a) where
compared to the night temperature image AST08 (Fig. 9b). This
visual observation is confirmed by statistics for subset areas 1
and 2 (Fig. 9b), in which scene variance has been reduced 53%
and 34% respectively, compared to the nighttime temperature
image. This is considered significant because even with perfect
removal of albedo, topographic slope, and thermal inertia effects, temperature differences will remain because of variations
in sensible, latent, and geothermal heat flux.
Examination of image subsets reveals the degree to which
albedo, topographic slope, and thermal inertia effects have been
Fig. 9. Fumaroles at Bradys Hot Springs are more conspicuous on the
processed image (a) compared to the single AST08 nighttime image (b). W =
warm ground on the lower eastern slopes of the Truckee Range; s =
groundwater springs in playas; v = “green” vegetation including salt grass; c =
clouds.
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
357
Fig. 10. Correction for albedo effects: (a) composite albedo image of ASTER bands 1, 2, and 3 (VNIR); (b) day temperature image; (c) night temperature image;
(d) final processed image after corrections for albedo, topographic slope, and thermal inertia. Highly reflective diatomite in three open pits appears cool in day (b) and
night (c) temperature images, but the temperature anomaly has been largely removed in the final processed image (d). In (b), (c), and (d), brighter areas indicate higher
surface temperatures. The cooler area in the northeast corner of (b), (c), and (d) consists of grassy meadows.
4.2. Ground moisture content and surface water
Image processing steps in this paper do not specifically address
sensible and latent heat exchange. Because the images were
acquired at the end of August after a period of generally dry hot
weather, soil moisture contents would have been low in much of
the Hot Spring Mountains and the Truckee Range, and most
surface temperatures would have been radiation-controlled, as
opposed to evaporation-controlled. Nevertheless, the effect of
evaporative cooling is noticeable in playas with locally high
moisture contents. For example, a broad strip of warm ground is
visible along the lower eastern slopes of the Truckee Range
(label w, Fig. 9b). These relatively warmer temperatures are
caused in part by ground moisture and elevation differences: the
Fig. 11. Correction for topographic effects on the daytime thermal image: (a) day temperature image; (b) final processed image after corrections for albedo, topographic
slope, and thermal inertia. Southeast-facing topographic slopes are warmer in the ASTER daytime image (a), but in the final processed image (b), residual temperatures
are largely a function of elevation and not slope orientation. Brighter areas indicate higher surface temperatures. Similar reductions in topographic shading-related
temperature anomalies were obtained with the nighttime images, where in unprocessed images southwest-facing slopes are warmer.
358
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
Fig. 12. Correction for thermal inertia effects: (a) composite image of ASTER bands 1, 2, and 3 (VNIR); (b) day temperature image after corrections for albedo and
topographic slope; (c) night temperature image after corrections for albedo and topographic slope; (d) final processed image after corrections for thermal inertia.
Because of its lower thermal inertia, sand in valleys appears warm in the daytime image (b) relative to outcropping basalt, and appears relatively cool in the night
image. After corrections for thermal inertia in the final enhanced image (d), temperatures appear warmer at lower elevations (meteorological effect). In (b), (c), and (d),
brighter areas indicate higher surface temperatures.
playa valley bottoms to the east are cooler because of higher
moisture contents (higher latent heat loss). To the west of the
anomaly, cooler ground temperatures are a result of higher
elevations where air temperatures are cooler and the atmosphere
thinner. In some places elevation and moisture effects do not
appear capable of explaining the entire anomaly, and a nighttime
radiation inversion or other meteorological effects could be
contributing.
Although most playas appear cool on the processed image, a
few areas are relatively warm (label s, Fig. 9b). Field inspections
suggest diverse causes, including 1) dry areas within playas, 2)
shallow ponds where convective circulation keeps surface water
temperatures warmer at night, and 3) upwelling of groundwater
to the surface. Most of these areas appear warm on the nighttime
image but cool in the daytime image, so that when the day and
night images are weighted together, these anomalies appear
cooler than they do on the night image alone, and thus it becomes
easier to distinguish geothermal anomalies from a variety of
surface and groundwater-related phenomena (Fig. 9).
4.3. Vegetation
The response of vegetation to the processing appears to depend
on the type of vegetation. For sage and salt brush, the thermal
inertia (Figs. 5 and 6) and albedo corrections address much of the
vegetation-related temperature variations. In contrast, areas with
denser vegetation growth such as salt grass near groundwater
seeps and springs show up as dark (cool) anomalies on the
processed image (label v, Fig. 9b). In these areas, higher transpiration rates, photosynthesis, and evaporative cooling act to keep
surface temperatures low.
5. Discussion and conclusions
Although the equations are simplified and the methodology
partly empirically derived, the processing steps removed much of
the spatial temperature variations caused by changes in ground
albedo, topographic slope aspect, and thermal inertia. By minimizing these temperature anomalies, it becomes easier to focus
on, and field check, anomalies that could have geothermal
sources. Similarly, this processing may facilitate mapping of
ground moisture contents.
In some places the processing described herein may not be
necessary, as at Bradys Hot Springs where the thermal anomaly
was detected equally well with the nighttime image (AST08)
and the final processed image. However, there are 3 situations
where the additional processing may be warranted:
1) Where it is desired to search for thermal anomalies over a large
region. In this case, the processing steps will reduce or
M.F. Coolbaugh et al. / Remote Sensing of Environment 106 (2007) 350–359
eliminate false geothermal anomalies produced by slope,
albedo, and thermal inertia.
2) Where topographic slopes are relatively steep and variable,
such as mountainous terrain. In these situations, it can be
difficult and tedious to distinguish thermal anomalies from
strong false anomalies caused by warmer sun-facing slopes in
uncorrected nighttime images.
3) Where the conjunction of surface geology and topography
serves to conceal or distort uncorrected thermal anomalies,
such as at Steamboat Springs (Coolbaugh et al., 2000).
There are many ways the equations could be improved, for
example, better time decay functions in Eq. (9) could be designed
and tested. Ultimately a more comprehensive treatment with
differential heat flow equations may yield more accurate results
but would require more sophisticated processing techniques not
as easily employed without custom software. One of the objectives of this paper was to explore to what extent simpler methods
would work.
The simplified equations notwithstanding, one of the most
important factors currently limiting the quality and completeness of
corrections is the geo-location accuracy; this would have to be
addressed before significant improvements could be attained. The
nighttime temperature image was the most difficult to register
accurately because of the 90 m pixel size and a lack of co-registered
VNIR bands. Improved geo-locational tools now available from the
JPL/ASTER web site should improve the accuracy of the nighttime
image location, but for registration of the daytime image, ground
control points may still be helpful. In mountainous terrain, it may be
necessary to use orthorectification. A higher resolution DEM
would also help in some areas.
Acknowledgements
This study would not have been possible without the help of
Anne Kahle, Leon Maldonado, and Elsa Abbott of the Jet
Propulsion Laboratory in Pasadena, California, who provided
data acquisition requests for the ASTER images and key
assistance predicting and programming dual day–night acquisitions on the same date. The research was funded in part by the
Nevada NASA Space Grant Consortium and by a U.S.
Department of Energy geothermal energy grant (instrument
number DE-FG07-02ID14311) through the Great Basin Center
for Geothermal Energy. Graduate students Sarah Mahoney and
Jessica Muehlberg helped with image processing as part of a
course in thermal remote sensing at the University of Nevada,
Reno, taught by Dr. James V. Taranik.
References
Abrams, M. J. (2000). The Advanced Spaceborne Thermal Emission and
Reflection Radiometer (ASTER): Data products for the high resolution
imager on NASA's Terra Platform. International Journal of Remote Sensing,
21(5), 847−859.
359
Abrams, M., & Hook, S. (2002). ASTER user handbook. Pasadena, CA: Jet
Propulsion Laboratory.
Allis, R. G., Nash, G. D., & Johnson, S. D. (1999). Conversion of thermal
infrared surveys to heat flow: Comparisons from Dixie Valley, Nevada, and
Wairakei, New Zealand. Geothermal Resources Council Transactions, 23,
499−504.
Bastiaanssen, W. G. M., Menenti, M., Feddes, R. A., & Holtslag, A. A. M.
(1998). A remote sensing surface energy balance algorithm for land
(SEBAL) 1. Formulation. Journal of Hydrology, 212–213, 198−212.
Blackwell, D. D. (1983). Heat flow in the northern Basin and Range province.
Geothermal Resources Council Special Report, 13, 81−92.
Coolbaugh, M. F., Taranik, J. V., & Kruse, F. A. (2000). Mapping of surface
geothermal anomalies at Steamboat Springs, NV. using NASA Thermal
Infrared Multispectral Scanner (TIMS) and Advanced Visible and Infrared
Imaging Spectrometer (AVIRIS) data. Proceedings, 14th thematic conference, applied geologic remote sensing (pp. 623−630). Ann Arbor, MI:
Environmental Research Institute of Michigan (ERIM).
Elachi, C. (1987). Introduction to the physics and techniques of remote sensing
(pp. 122−129). New York: John Wiley and Sons.
Finlayson-Pitts, B. J., & Pitts, J. N. (2000). Chemistry of the upper and lower
atmosphere (pp. 66−68). San Diego, CA: Academic Press.
Gillespie, A. R., & Kahle, A. B. (1977). Construction and interpretation of a
digital thermal inertia image. Photogrammetric Engineering and Remote
Sensing, 43, 983−1000.
Gillespie, A., Rokugawa, S., Matsunaga, T., Cothern, J. S., Hook, S., &
Kahle, A. B. (1998). A temperature and emissivity separation algorithm for
Advanced Spaceborne Thermal Emission and Reflection Radiometer
(ASTER) Images. IEEE Transactions on Geoscience and Remote Sensing,
36, 1113−1126.
Hook, S. J., Abbott, E. A., Grove, C., Kahle, A. B., & Palluconi, F. (1999). Use
of multispectral thermal infrared data in geological studies. In A. N. Rencz
(Ed.), Manual of remote sensing, 3rd Edition Remote Sensing for the Earth
Sciences, vol. 3 (pp. 59–110) New York: John Wiley and Sons.
Kahle, A. B. (1977). A simple thermal model of the Earth's surface for geologic
mapping by remote sensing. Journal of Geophysical Research, 82, 1673−1680.
Koenig, H. (1989). Recent geyser activity at Steamboat Springs, Nevada. GOSA
Transactions — The Annual Journal of the Geyser Observation and Study
Association, 1, 243−254.
Kruse, F. A., Kierein-Young, K. S., & Boardman, J. W. (1990). Mineral
mapping at Cuprite, Nevada with a 63-channel imaging spectrometer.
Photogrammetric Engineering and Remote Sensing, 56, 83−92.
Lee, K. (1978). Analysis of thermal infrared imagery of the Black Rock Desert
geothermal area. Colorado School of Mines Quarterly, 4(2), 31−44.
Platt, B. (2002), personal communication, SB Geo, Inc., Reno, Nevada, January,
2002.
Sabins, F. (1978). Remote Sensing: Principles and interpretation (pp. 119−124).
San Francisco: W.H. Freeman and Co.
Schowengerdt, R. A. (1997). Remote sensing, models and methods for image
processing (pp. 313−318). (3rd Ed.). San Diego, CA: Academic Press.
Stewart, J. H. (1983). Cenozoic structure and tectonics of the northern Basin and
Range province, California, Nevada, and Utah. Geothermal Resources
Council Special Report, 13, 25−40.
Watson, K. (1973). Periodic heating of a layer over a semi-infinite solid. Journal
of Geophysical Research, 78, 5904−5910.