Geography Reference
In-Depth Information
this theoretical background is necessary for choosing
whether the required tools and technical skills are available
to process and interpret the images.
Emissivity should be chosen to match the specific water
and sensor characteristics. Spectral libraries provide a
good source of emissivity and reflectivity data for most
applications. Care must be taken in interpretation of TIR
data under conditions that might affect the emissivity,
such as large observation angles and rough water.
To reduce errors that can occur at large observation
angles, images should be chosen to be near nadir ( < 30 ).
When multiple images are obtained they should have
similar observation angles.
While critical to accurately retrieving quantitative tem-
perature from TIR data, radiometric correction can be
time consuming and expensive. An alternative is to use
non-radiometrically corrected TIR images to assess rel-
ative spatial patterns within a single image, but this will
limit the applications for which the data can be used to
those not requiring absolute temperature information.
TIR radiation cannot be sensed through clouds or fog,
so standard remote sensing practices should be used
to identify and mask these out of the image before
quantifying water temperature.
profile can be complex because temperature from the
stream or river centre may not be representative spatially
across the stream. This necessitates approaches such as
weighted averages (Cristea and Burges, 2009) or median
filtering (Handcock et al., 2002).
Many TIR imaging sensors used for measuring water
temperature are designed to have multiple spectral bands
located at different wavelengths. These wavelengths are
typically determined from the TIR emissivity spectra
useful for geological applications (e.g., Gillespie et al.,
1984; Bartholomew et al., 1989). When available, multi-
ple bands have an advantage for checking the accuracy of
image processing when ground-based temperatures are
available, because of the physical constraint of there being
only one true temperature. The band or bands with the
least amount of instrument noise and atmospheric effects
can then be selected to calculate the final image temper-
atures. Alternatively, multiple bands can be averaged to
reduce noise due to atmospheric or sensor differences
and provide a better estimate of the actual temperature
(Handcock et al., 2006).
5.4.2 Accuracy, uncertainty, andscale
The issue of spatial scale is critical to the remote sensing
of rivers using TIR data, as the combination of river
width and pixel size will determine whether it is possible
to distinguish the river from the bank at the desired
levels of accuracy and uncertainty with the TIR imaging
sensor. The accuracy (bias) of a TIR measurement can
be compared to a known in situ reference value used
for validation, while its uncertainty (precision) is the
repeatability of measurements. The radiometric precision
of a TIR sensor is described by its NE
See also Table 5.1.
5.4 Extracting useful information
from TIR images
Once the TIR image data have been processed to deter-
mine T r , it is still necessary to extract information specific
to the thermal application. In this section we first discuss
how to calculate a representative water temperature. We
then examine how both the size of the river relative to the
pixel size of the TIR imaging sensor, and the near-band
environment, influences the accuracy of extracted water
temperature. We note that care is required for interpre-
tation of TIR images within their terrestrial and aquatic
context; a trained operator is required to reduce errors
associated with image interpretation. A detailed exami-
nation of this topic is beyond the scope of this chapter,
therefore we illustrate this complexity with examples.
T, or 'noise-
equivalent change in temperature,' which is theminimum
difference in temperature that the sensor can resolve as a
signal from the background noise.
Rivers often have a complex morphology of channels,
boulders, shallow areas, gravel bars, islands and in-river
rocks, and vary greatly in hydrological and hydraulic
characteristics such as ground-water inputs, water depth,
water velocity and turbulence fluctuations. Handcock
et al. (2006) quantified the accuracy and uncertainly
related to the TIR remote sensing of river temperature
across multiple spatial scales, imaging sensors, and plat-
forms, and showed that when the water was resolved by
less than three pure water pixels of a well-mixed river, the
measurements had low accuracies and high uncertainties.
In practice, it can be difficult to find three pure water
pixels as the edge pixels are frequently contaminated
Δ
5.4.1 Calculatinga representativewater
temperature
Thermal applications usually require fine resolution TIR
data to map thermal pollution sources or locate ther-
mal refugia. Extracting a longitudinal water temperature
Search WWH ::




Custom Search