본문 바로가기
  • Home

Study on Rainfall Estimation of High Resolution Using Machine Learning

  • Journal of Knowledge Information Technology and Systems
  • Abbr : JKITS
  • 2017, 12(2), pp.295-303
  • DOI : 10.34163/jkits.2017.12.2.009
  • Publisher : Korea Knowledge Information Technology Society
  • Research Area : Interdisciplinary Studies > Interdisciplinary Research
  • Published : April 30, 2017

Min-Gyu Kim 1 Moohun Lee 1

1한국외국어대학교 차세대도시농림융합기상사업단

Accredited

ABSTRACT

Occurred rainfall as local torrential rainfalls among the climate change has a decisive effect on life. Rainfall is closely related to the living of people, its exact prediction is important to us. In this paper, we have improved the estimation accuracy of rainfall by comparing the applicable interpolation algorithms of meteorological field. The comparison algorithms are Kriging, MLP(Multi-Layer Perceptron), SVR(Support Vector Regression), Random Forest and Bagging. For the experiment, we used the AWS(Automatic Weather System) data which exists the event of rain on August 16th, 2015, 16:00~20:00. The AWS data is collected by a 1-minute cycle, the attributes related to rainfall include precipitation type, accumulated rainfall such as 10 minutes, 15 minutes, 30 minutes, and 1 day, etc. The dataset consists of training data and test data in 255 observation stations, Seoul. 230 stations are used as training data, the remaining stations are used as test data. Validation methods such as RMSE, R-square, Correlation Coefficient, CSI(Critical success index) and BIAS(frequency BIAS) were used to compare the algorithms. CSI and BIAS were measured for precision of rainfall and calculated using the rain contingency table. Our experimental results showed that bagging algorithm performed better than the others in estimating the rainfall information.

Citation status

* References for papers published after 2023 are currently being built.