The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
422
and meteorological observations are prior knowledge for renew
SI model.
Fire points are identified according to its temperature that is
much higher than the background. Generally, the temperature of
forest fire is around 600K. However, some points do not reveal
such a higher temperature in remote sensing than that of in the
reality because of the spatial resolution of the image. So we
divided forest fire points into two types, which were higher
temperature forest fire points and non-high temperature forest
fire points. They were treated in different ways with different
channels of MODIS.
2. LST RETRIEVALTITLE
Land surface temperature (LST) retrieval, which is one of the
important parameter in environment remote sensing, is also a
most difficult one in the field of remote sensing. Many scholars
worked on this subject and achieved many researches results,
such as thermal radiative transfer equation method, mono
window algorithm, split-window algorithm, and multi-channels
algorithm. (Price J C, 1984; Becker F, 1987; Cooper D I, 1989;
Becker F, 1990; Sobrino J A, 1991; Prata A J, 1993; Wan Z,
1997; Li Zhaoliang, 1999; Shunlin Liang, 2001; Ma X-L, 2000)
Generally, these models are divided into two categories,
physical models and statistical models. Our experiments in
Chongqing show that the physical model has a better
performance. Ma’s physical model is used in our experiments
duo to its good LST retrieval performance (Ma X-L, 2000). It is
an integrated inversion algorithm with land surface and
atmosphere parameters retrieved together.
For a cloud-free atmosphere under local thermodynamic
equilibrium the RTE (Radiative Transfer Equation) in the
thermal infrared region may be expressed as below. (Ma X-L,
2000)
r (Yj > M) = B{Vj,t s )£(yj,iu)T{Vj,ii,p s ) + R a (Vj, p)
1 2
(2-1)
+ t (v■, p, -p G , 0)E 0 (v )f r (//; -p ', <j> ’)
In this formula, ^ V J’^ j s the mean spectral radiance
V
measured in a band whose mean effective wave number is J
and the cosine of local zenith angle ^ is ^ j»t s )
the Planck function of the surface skin temperature
t e(v ,p) .
s , 1 is the effective surface emissivity, and
t(v p p )
J ’ ’ s/ is the transmittance from the surface pressure
level ^ s to the top of the atmosphere along the observation
angle @. The first term of Eq. (1) represents surface emission
to space (less atmospheric absorption). - g ^
upwelling radiance contributed from atmosphere to space.
R d( v j >№> №>0) deices t he atmospheric downwelling
emissive radiance being reflected by the surface upward to
space; its incident direction is represented by and
^ (where the minus sign indicates that direction is always
downward). (Ma X-L, 2000)
Though we have improved Ma’s model and got a higher
precision of the retrieved target parameter LST, the improved
inversion method is not as efficient as needed. A significant
disadvantage is its large quantity of calculation, which makes
the method too complex to be widely used in operational
application. Furthermore, Chongqing city has its own
characteristics, mountainous terrain and cloudy climate. Ma’s
model may not be applicable in such a reality, so it is better to
develop a new linear statistical model as simple as it could.
On the base of an intensive comprehension of the physical
model, we deduced a linear experiential expression with both
MODIS and observation data in July and August, 2006.
LST = 2.7932(5^, - 57^)-0.178157;, +354.3806
(2-2)
j^j 1 ^32
where 1 ~ ’ are the brightness temperatures of
MODIS bands 31 and 32.
LJ & Jl *<№cq„COimey «hf
Figure 1-1 LST retrieval results on August 7 th , 2006
Compared with the meteorological data through July and
August, 2006, LST retrieved with the linear model matched
quite well with the observation data in Fengjie, Youyang,
Fuling, Liangping, Shapingba, and Wanzhou. Figure 1-1 is the
LST inversion result on August 7 th , 2006 (The white area is
covered with cloud).
However, it is better to combine the physical model and
statistical model together, so as to get a better retrieval precison
of the target parameter in a practical way. We are working on
this purpose now, and the result will be presented soon.