Atmospheric temperature profile retrieval using multivariate nonlinear regression

In this paper multivariate nonlinear regression was used to retrieve the atmospheric temperature profile from remotely measured microwave emissions of the atmosphere. In this method the nonlinear models for each layer of the atmosphere are at first established and the model for the whole profile is obtained by combining the layer models together. In model building we use the stepwise regression, which analyses the importance of all the predictor variables in the model and determines which of the predictor variables are allowed to enter the model under an entry and exit criterion, and finally calculates the coefficients of all the included predictor variables in the model using the least squares approach. Since the entry and exit criterion can be freely set, it is possible to find a compromise between the model accuracy and the model sensitivity to noise. Simulations were done for the region of the Weddell Sea in the southern ocean.