Thermal infrared remote sensing technology supplies an attractive way to measure land

s

uface temperature （LST） on a large scale simultaneously. NOAA

A

VHRR data is commonly used in the inversion study of LST. But the inversion is a challenge for the scientists because of the complexity of land surface. So far split

w

indow algorithm is a major solution to retrieve land

s

urface temperature from thermal infrared remote sensing data. But because of the obstacle to obtain the in

s

itu validation measurements simultaneously

including land

s

urface temperature and emissivity especially in pixel scale

it is confined to judge directly which one is more precious and more applicable.\;In this paper

using radiative transfer code LOWTRAN

7

and six standard atmosphere models supplied by it

we analyzed

by simulation

the sensibility of the six common split

w

indow algorithms to the atmospheric profile error

including atmospheric water vapor profile and temperature profile. The sensibility to spectral emissivity

including the average emissivity and the differential emissivity between AVHRR channels 4 and 5

is also analyzed. It turns out that

（1） algorithms of SOB and UVM have less sensibility to atmospheric profile

and have more satisfactory accuracy of better than 1.5K; （2） most algorithms are more sensible to the mean emissivity than to the emissivity difference. Sobrino （1991） algorithm has the least sensibility to emissivity. As a result

we obtained an indirect criterion priliminarily for the six split

w

indow algorithms.