EFFECTS OF SPECTRAL SHIFTS ON SENSOR RESPONSE 
59 
P.M. Teillet 
Canada Centre for Remote Sensing 
1547 Merivale Road, Nepean, Ontario, Canada K2G 4V3 
ABSTRACT 
The effect of spectral filter shifts on sensor 
outputs are presented for the case of the Thematic 
Mapper (TM) imaging vegetation and the Moderate 
Resolution Imaging Spectrometer-Nadir (MODIS-N) 
imaging vegetation and water. Errors in excess of 
5% per 5 nm spectral shift are noted for TM bands 
1 and 2, and the error decreases with increasing 
wavelength for the TM bands in general. Most 
MODIS-N bands in the solar reflective spectrum can 
be affected by the spectral shift problem, with 
the largest errors (several hundred percent) 
occurring for the 10-nm water vapour band centred 
at 0.935 micrometers. 
INTRODUCTION 
Spectral bandpasses made by interference filters 
are susceptible to changes in response charac 
teristics with time and due to changes in environ 
mental conditions. It is not uncommon for the 
entire bandpass to shift toward shorter wave 
lengths and/or for the long-wavelength side of the 
bandpass to shift toward shorter wavelengths. 
Suits et al. (1988) have developed an analytical 
approach to examine the effect of spectral 
response shifts on the outputs of several sensors. 
They comment that the linear assumption implicit 
in their approach may not be appropriate for non 
linear conditions such as those considered here. 
Markham and Barker (1985) have noted a 9-nm shift 
in the long wavelength cut-off of the band-3 
filters in the fourth Landsat Multispectral 
Scanner System (MSS) after storage at normal 
pressure, temperature and humidity over an 18- 
month period. The cut-on at shorter wavelength 
was unchanged because it was mainly defined by a 
stable absorption filter. The cut-off was 
provided by an interference filter, thus the half- 
width and therefore the total transmittance of the 
filter changed (Slater, 1989, personal communica 
tion). The same effect has been observed in the 
case of the SPOT-1 HRV filters. Here the effect 
was most noticeable as expected for the narrow 
filters. For example, Dinguirard et al. (1988) 
report a 27% transmittance change in multispectral 
band 2 with a change from air to vacuum 
conditions. The broad panchromatic band on the 
other hand exhibited only a 3% change in trans 
mittance . 
A shift of a narrow filter can produce a sig 
nificant change in measured radiance even when the 
filter transmittance remains constant. This is 
because of the presence of Fraunhofer absorption 
lines in the solar output. It can be shown that 
a 10-nm filter with a rectangular profile centered 
at 480 nm, when shifted through 5 nm, can cause an 
8% change in the recorded signal (Slater, 1989, 
personal communication). For this reason, the 
system must be radiometrically calibrated in 
flight using solar radiation and/or a sensitive 
wavelength calibration must be employed. It is 
also desirable to use interference filters that 
have been manufactured using ion bombardment to 
reduce the porosity and therefore increase the 
stability in refractive index of the dielectric 
layers (Slater, 1989, personal communication). 
As the spectral bands on satellite sensors become 
narrower and selected for specific scientific 
measurements, it becomes more important to assess 
the impact of any changes in the response charac 
teristics of the bands if interference filters are 
used. The flexibility and speed of the 5S 
atmospheric code (Tanre et al., 1986) lend 
themselves well to some initial studies of this 
problem. In particular, the code's principal mode 
of operation is to produce band-integrated 
results. It also allows the user to input 
spectral response profiles on a 5-nm grid. 
Therefore, it is a relatively simple matter to run 
different cases to determine the error incurred 
if, unknown to the user, a spectral band has 
shifted. In this study, the effect of partial 
(long-wavelength side only) and full spectral band 
shifts on radiances output from a radiative 
transfer code were examined for reflective bands 
from two sensor systems: (i) the Landsat Thematic 
Mapper (TM) and (ii) the planned Moderate 
Resolution Imaging Spectrometer-Nadir (MODIS-N), 
should interference filters be used to construct 
its spectral bandpasses. In both cases, it was 
found that spectral shifts introduce significant 
changes in sensor response for many of the bands. 
Landsat Thematic Mapper Bands 
The effect of partial (long-wavelength side only) 
and full spectral band shifts on outputs from the 
5S program were examined for the case of the 
reflective TM bands. The input conditions are 
listed in Table 1 and assume, in particular, a 
vegetated surface. For each of the six reflective 
TM bands, forward runs of the 5S code were made 
for no spectral shift, for partial spectral shifts 
of 5, 10, and 15 nanometers (Figure 1), and for 
full spectral shifts of 5, 10, and 15 nanometers 
(Figure 2) toward shorter wavelengths. (A forward 
run of the code computes apparent reflectance at 
the sensor given the surface reflectance.) The 
apparent reflectances at satellite altitude output 
from these 5S runs were then used as input values 
for reverse runs (Teillet, 1989) with no spectral 
band shifts of any kind. The predicted surface 
reflectance differs substantially from the 
original surface reflectance in many cases 
(Figures 3, 4, and 5). The greatest differences 
occur in TM band 1, with a 25 percent error for a 
full shift of 15 nm and 12 percent for a partial 
shift of 15 nm. For the TM bands in general, the 
error decreases with increasing wavelength. 
MODIS-N Bands 
MOD IS is an instrument that will be part of the