Precipitation Data Assimilation System Based on a Neural Network and Case-Based Reasoning System

Jing Lu,Xiakun Zhang,Wei Hu

doi:10.3390/info9050106

Abstract

There are several methods to forecast precipitation, but none of them is accurate enough since predicting precipitation is very complicated and influenced by many factors. Data assimilation systems (DAS) aim to increase the prediction result by processing data from different sources in a general way, such as a weighted average, but have not been used for precipitation prediction until now. A DAS that makes use of mathematical tools is complex and hard to carry out. In our paper, machine learning techniques are introduced into a precipitation data assimilation system. After summarizing the theoretical construction of this method, we take some practical weather forecasting experiments and the results show that the new system is effective and promising.

Highlights

Data assimilation [1,2,3,4,5,6,7,8] is a process by which data from different sources are processed and adjusted in a general or a comprehensive way
The main contribution of this paper is to propose a precipitation data assimilation method that makes use of machine learning techniques
Since precipitation is a kind of data type that occurs after the numerical processing, it can be understood as “posterior”

Summary

Introduction

Data assimilation [1,2,3,4,5,6,7,8] is a process by which data from different sources are processed and adjusted in a general or a comprehensive way. Models are widely used for predicting future states of the atmosphere. These models are dependent on exact initial conditions. In a narrow way, data assimilation is initially considered to be a process of analyzing and processing the observed data that are conformed to certain spatial and temporal distributions for providing as close to exact initial fields as possible for numerical predictions. The main data assimilation methods from the middle of the last century to the present include the function fitting method [12,13] (objective analysis), the stepwise correction method (SCM) [14,15], the optimal interpolation method (OI) [16,17], the variability method (3Dvar, 4Dvar) [18,19], and the ensemble

Methods

Discussion

Conclusion