A multilayer perceptron approach for the retrieval of vertical temperature profiles from satellite radiation data

Abstract

In this paper a multi layer perceptron neural network is used to retrieve vertical atmospheric temperature profiles from satellite radiation data. The training set consists of data provided by the direct model characterized by the Radiative Transfer Equation (RTE) 2) Direct problem Equation (I) represents the direct problem, in which IA is the spectral radiance, A is the channel frequency; 3 is the layer to space atmospheric transmittance function; subscript s denotes surface [9]; and B is the Planck function which is a function of the temperature T and pressure p given by equation (2), where h is the Planck constant, c is the light speed, and kB is Boltzmann's constant.The meteorological process of the atmosphere requires information of the vertical structure of the temperature and the water vapor, which are indirectly provided by satellite radiation data due to the lack of radiosonde observational stations over the globe. These pieces of information are especially important for weather analyses and data assimilation in numerical weather forecasting models in meteorology.Satellite measured radiation is interpreted in terms of meteorological parameters, requiring the inversion of the Radiative Transfer Equation (RTE) that relates measurements of radiation at different spectral frequencies to the energy at different atmospheric regions. There is a degree of indetermination associated with the spectral resolution and the number of spectral channels.Noises in the measuring process imply in instabilities to this solution [3,4], which led to the development of different methodologies and models to improve satellite data processing. Due to the difficulties to obtain correct RTE solutions, several approaches and methods have been developed to infer information from satellite data [5][6][7].In this paper a multilayer perceptron artificial neural network (ANN) is designed to solve the inversion of remotely sensed data in a multidimensional function approximation approach. The ANN retrieved temperatures are compared to those obtained [1] and [2], who used B2(T) = 'tc2/25 [hehlk,2T -ii (2)The solution of equation (l) maybe approached by discretization using central finite differences (3)

A High-Accuracy and Fast Retrieval Method of Atmospheric Parameters Based on Genetic-BP

“…A general BP neural network is a multi-layer neural network [45][46], whose structure is shown in Figure 1.…”