Search results

1 – 1 of 1
To view the access options for this content please click here
Article

Srinivasa Ramanujam, R. Chandrasekar and Balaji Chakravarthy

The purpose of this paper is to develop an algorithm, using PCA‐based neural network, to retrieve the vertical rainfall structure in a precipitating atmosphere. The…

Abstract

Purpose

The purpose of this paper is to develop an algorithm, using PCA‐based neural network, to retrieve the vertical rainfall structure in a precipitating atmosphere. The algorithm is powered by a rigorous solution to the plane parallel radiative transfer equation for the atmosphere with thermodynamically consistent vertical profiles of humidity, temperature and cloud structures, together with “measured” vertical profiles of the rain structure derived from a radar.

Design/methodology/approach

The raining atmosphere is considered to be a plane parallel, radiatively participating medium. The atmospheric thermodynamic profiles such as pressure, temperature and relative humidity along with wind speed at sea surface and cloud parameters corresponding to Nargis, a category 4 tropical cyclone that made its landfall on May 2, 2008 at the Republic of Myanmar, are obtained by solving the flux form of Euler's equations in three‐dimensional form. The state‐of‐the‐art community software Weather Research and Forecasting has been used for solving the set of equations. The three‐dimensional rain profiles for the same cyclone at the same instant of time are obtained from National Aeronautics and Space Administration's space borne Tropical Rainfall Measuring Mission's precipitation radar over collocated pixels. An in‐house Micro‐Tropiques code is used to perform radiative transfer simulations for frequencies corresponding to a typical space borne radiometer, and hence to generate the database which is later used for training the neural network. The back propagation‐based neural network is optimized with reduced number of parameters using principal component analysis (PCA).

Findings

The results show that neural network is capable of retrieving the vertical rainfall structure with a correlation coefficient of over 0.99. Further, reducing the ill‐posedness in retrieving 56 parameters from just nine measurements using PCA has improved the root mean square error in the retrievals at reduced computational time.

Originality/value

The paper shows that combining numerically generated atmospheric profiles together with radar measurements to serve as input to a radiative transfer model brings in the much‐required synergy between numerical weather prediction, radar measurements and radiative transfer. This strategy can be gainfully used in satellite meteorology. Using principal components to reduce the ill‐posedness, thereby increasing the robustness in retrieving vertical rain structure, has been attempted for the first time. A well‐trained network can be used as one possible option for an operational algorithm for the proposed Indian climate research satellite Megha‐Tropiques, due to be launched in early 2011.

1 – 1 of 1