An original approach of the forward radiative transfer modelling of sp
ace radiometric observations is presented and applied to the atmospher
ic vertical temperature sounder SSM/T (Special Sensor Microwave/Temper
ature) measurements. This approach relies on multi-layer neural networ
ks and is compared to ''classical'' methods. This study has been reali
zed not in a simulation context but with real data from radiosondes an
d the SSM/T instrument. Copyright (C) 1996 Elsevier Science Ltd