Precise retrieval of land surface temperature from satellite remotely sensed data need atmospheric correction and a known effective emissivity of the pixel. Various split window algorithms have been used to solve the first problem

but they all need a known surface emissivity. LSF model can calculate the effective emissivity of the nonisothermal and heterogeneous pixel

but the data must be atmospherically corrected when using satellite images. In this paper

we have developed a model based algorithm that can correct the atmosphere effects and retrieve component temperatures using ATSR 2 dual angle observation. In this algorithm

QUAD algorithm is used to perform atmospheric correction

and LSF model is used to calculate the directional effective emissivity

by iteration

atmospheric correction and component temperatures retrieval can be completed synchronously. Good linearity was found between the difference of the directional emissivity and the difference of the directional brightness temperature after atmospheric correction. Although the range of the retrieved component temperatures is large

it is still clear that the component temperatures of vegetation and soil are separated. Further analysis of the uncertainty and sensitivity for the two component temperatures show that if only the most sensitive sample is used in inversion

the results tend to be more robust.