Land Surface Temperature (LST) Retrieval from GOES Satellite Observations - Satellite-based Applications on Climate Change - page 285

Reflectance Radiometer). This method utilizes six thermal bands from TIMS and
five thermal bands from ASTER, focusing on emissivity retrieval. The MODIS
team (Wan et al. 1996) proposes a day/night algorithm that uses day and night
measurements in N MODIS bands with 2 N observations. The unknown parameters
include N band emissivities, daytime surface temperature, nighttime surface tem-
perature, and four atmospheric parameters (air temperature and water vapor content
at two times). The total number of unknowns N + 7 needs to be smaller than or
equal to the 2 N observations, ( N +7)

7. Seven MODIS
thermal bands (12.91, 12.25, 11.98, 8.6, 4.70, 4.11, and 3.74) are used to solve a 14-
equation set. The problem of this approach is that the pixel during nighttime may
not be the same as during daytime, and seven infrared window bands are required to
solve the equations. The alternative approach is to use Lookup Tables (LUT)
generated from radiative transfer model simulations. This will introduce errors
due to interpolation. Liang ( 2001 ) proposed an optimized algorithm for separating
land surface temperature and emissivity from MODIS and Advanced Spaceborne
Thermal Emission and Reflection Radiometer (ASTER). This method also needs
five (ASTER) and six (MODIS) thermal window bands, and emphasize emissivity
retrieval. Ma et al. ( 2002 ) proposed a physical algorithm for MODIS to retrieve
LST and surface emissivity simultaneously. This algorithm uses nine channels and
is computationally intensive.

2 N , which requires N

19.1.4 Validation Issues

The evaluation of LST retrievals from satellites has been difficult since satellites
measure skin temperature while global scale ground observations are from shelters.
The difficulty in obtaining ground truth has been addressed by Prata ( 1994 ) and
others. Weng and Grody ( 1998 ) tried to use shelter temperature in the early morning
(when the difference between surface skin temperature and shelter temperature is
the lowest) to validate LST retrieval from satellite SSM/I data.
Sugita and Brutsaert ( 1993 ) compared the land surface temperature derived from
the AVHRR and TOVS instruments on NOAA-9 and NOAA-10, the TM
instruments on Landsat-5, and VISSR instrument aboard GOES-7 with ground
truth from the First ISLSCP Filed Experiment (FIFE) (Sellers et al. 1992). For
clear condition, the root mean square differences from TOVS, TM, and VISSR data
are about 1-2 K; for AVHRR, it is of the order of 2-3 K. Prince et al. (1998)
compared surface temperature retrieved from the AVHRR with the BOREAS,
HAPEX-Sahel (Hydrological and Atmospheric Pilot Experiment in the Sahel),
and FIFE and showed RMS error of 3.5 K.
Existing approaches show that root mean square accuracy of 1-3 K can be
reached from the current operational and research satellite-borne visible/infrared
radiometers. While accuracy of 3 K is of marginal use, accuracy of 1 K or less is
desired for many applications. A main objective of this study is to develop new
algorithms to improve the accuracy of LST estimation from satellites.