International Archives of the Photogrammetry, Remote Sensing 
  
Table 1. Summary of datasets used for this study 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Viewi S Azim. 
A ges E. de iewing | Sun diff. 
Satellite Bands ; Date & angle angle 
Sensor used | time (UTC) (8.) (0) (l9. - $.| 
Landsat 02/23/2002 ; im 
- « . AR AO 3 C o 
ETM: 1-4 01:57:21 Nadir | 45.04° | 136.98 
ASTER 02/14/2002- E : 
-3 8.5 oO 3 oO 32 8 o 
VNIR | -: 02:32:16 8.59 51.39 132.84 
02/07/2003 
SPOTS -3: 2.07° A291 123.78° 
SPOT 5 ] 02:19:26 12.07? | 43.4 1237 
IKONOS jg [Gases 29.33°| 16.70?| 20.60° 
| À = 02:25 2.2 >. 24.0 
Next, we address the key issues in merging the datasets in the 
context of the physical environments at the time of the imaging. 
We will present a systematic way to refer the spectral and 
spatial dimensions into a single image plane. 
2.3 Spectral data processing 
Ideally, when the two spectrometer channels are calibrated to a 
single standard radiance unit (e.g. in Wm um ''sr^ ) or have pre- 
cross-calibrated ^^ nominal ^ counts, the instantaneous 
reflectance R,,. (4) of an object is the straightforward value of 
upwelling E, (4,obj) and downwelling radiance E, (4, sun) 
ratio, given that both measurements were taken at the same 
plane level. For spectrometers of uneven nominal radiance 
counts, cross-calibration is necessary. 
To describe the cross-calibration procedure, we follow the 
convention £ (wavelength, time, sensor ) for  radiances 
target 
measurements. Let Eg, (4.5,0bj) and E, (A,t,obj) be the 
radiances measured by the object sensor directed upward and of 
the standard respectively at the same time #, . However it is 
physically impossible to obtain such measurement 
simultaneously with one channel only. Hence the use of another 
sensor with the same spectral range is necessary. As mentioned, 
inter-calibration would be needed if channel nominal units do 
not match. E,,, (4,4, 5un) 7 kE, (At, 0bj) or in general: 
T Pres (4,14, sensor) 7 KE. (4.1, sensor2) , (1) 
where k is a calibration coefficient, therefore: 
E, (A.ts. 0b) 
Stat (2) 
E o (Ato, sun) 
Typically, the reflectance of the reference standard panel 
R,, (1) is available (e.g. a Spectralon" or Ever Color^). Our 
Rs (4) =k 
interest then is to find k : 
Eu sti sun) 
Ee (A.4.0bj) 
When the object sensor is directed at object target, at any time , 
R, (4) ES Ej, (A.t,0bj) (4) 
E, (Ast, sun) 
Further, if we correct for the effect of the air-water interface by 
introducing the factor, 7= [1 - n, (0,.0, )] / n, where 
r, (6,,0,) is the Fresnel’s reflectance due to differences in ray 
k= Ro (2) (3) 
entrance Ó, and exit angles 0, ; m, is the index of refraction of 
water (relative to air), and substitute k in Eq. (3), we have the 
and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
at-surface remote sensing reflectance from a cross-calibrated 
dual spectrometer: 
EA, (At, sun) rE,,, (A.t,obj) 
Ree) Ra NN 
"T ) sr ( FE E,, (At sun) ) 
— 
Cn 
2.4 Depth measurement and correction 
A multi-channel echosounder (DE-4000, Biosonics Inc.) 
equipped with GPS, capable of obtaining depth measurements 
to an average of 4 soundings per second is installed on a small 
(4-m length) motor boat. For purposes of tidal correction, a 25- 
hour tide level measurements was performed by deploying a 
depth logger (Diver™, Van Essen Instruments) at the reef area 
in a location with still water conditions during periods where no 
strong winds are prevailing) to obtain the actual tide 
amplitudeat time / in a given location within the reef. (7) is 
then applied to reduce the water depth measurement, d (/,) at 
specific time /, to its average depth, d or: 
d=d(1)-n(1) (6) 
If a satellite imagery was acquired at another period, /, , then it 
follows that d (t,) = d + nt, ) or in relation to Eq. (6) 
d(r,)=4(1,)-n(t)+n(e). () 
At the time of the satellite image acquisition, no actual 
amplitude distribution may be available. Instead, an 
interpolation method may be used based on the predicted tide 
tables simply by: 
a(t )= al) + 
where 7z(1,) and (1,) are the inclusive high and low tide 
f, — f, 
magnitudes respectively andr, <1, <1, . While this relationship 
may also be used to obtain n(t,) , measurement of d (1) a single 
location is necessary to fix the vertical datum. 
For purposes of comparison of the field spectral data with image 
values, we can reduce the spectrometer data measured into band 
values modulated by the spectral response function: 
^" S(b; A) R4 (4)dA 
Ro ( ) = I gr TH (9) 
[^ 5024 
Ay Sart 
where S(b;À) is the relative spectra sensitivity of the sensor 
for band ^ at wavelength A. 
2.5 Normalization due to differences in imaging geometry 
Previously, Paringit and Nadaoka (2003) studied the 
biderectional nature of shallow benthic cover reflectance and 
employed BRDF techniques to infer coral morphological 
characteristics from both in-situ spectra and satellite signal. 
BRDF for each benthic cover type was produced based on their 
pure reflectance R,(b) and transmittance 7, (b) spectra tO 
compute for the normalized reflectance R, (5) , and is applied to 
each of the reflectance values as: 
2G 
R. (b)  —|1- exp(-0.5F ) |x 
(5) el p( )] 
2[(z — 9)cos 9 e sin 9 | R, (P) - (sin 9— 9cos 3) vj, b) 
it 
(10) 
998 
Inte 
whe 
refle 
COS 
dire 
then 
The 
bent 
seag 
2.6 
To 
algo 
expl 
imag 
of ir 
prop 
data 
satel 
resol 
relati 
objec 
while 
Goin 
at th 
band 
surfa 
273 
estim 
Class 
spect 
neces 
is atte 
the di 
non-n 
consti 
the n 
comp 
R(b) 
where 
the re 
each E 
by th 
Square 
and th 
are th 
iteratis 
lo ar 
theref( 
In the 
ecolog 
is ado 
benthi