Study on Data Transfer in Meteorological Forecast of Small and Medium-Sized Cities and Its Application in Zhaoqing City. (10th September 2019)

• Record Type: Journal Article
• Authors: Huang, Tianwen; Jiao, Fei
• Is Part Of: Computer journal. Volume 63:Number 7(2020)
• Page Start: 1076
• Page End: 1083
• Publication Date: 2019-09-10
• Subjects: transfer learning -- data transfer -- autoregressive integrated moving average -- temperature forecast -- precipitation forecast; Computers -- Periodicals; 005.1
• DOI: 10.1093/comjnl/bxz087
• Languages: English
• ISSNs: 0010-4620

Abstract

Abstract: Using historical data, a machine learning model is usually built to forecast the future meteorological elements such as temperature, precipitation, etc. However, for numerous small and medium-sized cities, it is a challenging task because the maintained data of these cities are usually very limited due to historical or infrastructural reasons. So it is difficult to build an accurate forecast model in small and medium-sized cities. Aiming at this problem, a forecast method based on transfer learning method is proposed. Using instance-based transfer learning, this method extends the data of the target city by transferring the data from related cities and then builds a forecast model based on the extended dataset, so that the problem of insufficient samples in machine learning is solved. As a case study, the proposed technique is applied in Zhaoqing City, China. In the experiments, the data of temperature sequence and the precipitation sequence of Gaoyao weather station in Zhaoqing district are extended according to the data of related cities. The transferred temperature data and precipitation data are collected from 1884 to 1997 in Hong Kong and 1908 to 2016 in Guangzhou, respectively. Then temperature and precipitation forecasting models are built based on least square method and autoregressive integrated moving average. The experimental results have been verified by the actual situation. The results justify the effectiveness of the proposed method in building accurate meteorological forecasting model with limited data, and the superiority over existing techniques.