Data Assimilation by Artificial Neural Network using Conventional Observation for WRF Model

Abstract

In this paper, artificial neural network(ANN) are introduced to data assimilation for WRF model, which is a mesoscale complex model. A particle swarm optimization optimized Multilayer Perception data assimilation (MLP-PSO-DA) model is proposed in order to emulate the ensemble square root filter (EnSRF) analysis. Multilayer Perception is employed and the optimal parameter configurations are automatic obtained by particle swarm optimization (PSO) algorithm. The MLP-PSO-DA is integrated with WRF modeling system for assimilation cycle. The EnSRF analysis fields from July of 2004, 2005 and 2006 are taking as samples to train the model. The ANN-based data assimilation is conducted at July, 2007 with interval of 6h. The prognostic variables analysis fields of MLP-PSO-DA and EnSRF are very similar and the difference between two method is within a small scope. The results prove the effectiveness of MLP-PSO-DA. Meanwhile, the MLP-PSO-DA model has great advantage to speed up the DA process.