Based on Support Vector Machine（SVM）

a technique simulating radiative transfer forward model is presented. Using European Centre for Medium-Range Weather Forecasts （ECMWF） RTTOV

8

7

radiative transfer forward model and 60L

S

D profile database

we simulate the brightness temperature received in AMSU-A instrument. Combine this brightness temperature datasets and correspondence profile datasets as training and validation database. After training the SVM network

the simulating technique is validated. The results show that SVM network describes the nonlinear projection relationship between input space and output space very well

and the simulated brightness temperature of channel 5—14 is precise. The RMS error of channel 6—14 is less than 0.1K and the mean standard deviation is less than 0.01K. In order to find whether SVM simulated brightness temperature is appropriate for temperature retrieval

muti-regression retrieval method is used to retrieve temperature profile. Experiment result shows that the SVM simulate brightness temperature is appropriate for the purpose

and the retrieval precision is not only equally but also a little more precise than the RTTOV

8

7

simulated brightness temperature.