Articles | Volume 21, issue 6
https://doi.org/10.5194/npg-21-1159-2014
https://doi.org/10.5194/npg-21-1159-2014
Research article
 | 
01 Dec 2014
Research article |  | 01 Dec 2014

An improved ARIMA model for precipitation simulations

H. R. Wang, C. Wang, X. Lin, and J. Kang

Abstract. Auto regressive integrated moving average (ARIMA) models have been widely used to calculate monthly time series data formed by interannual variations of monthly data or inter-monthly variation. However, the influence brought about by inter-monthly variations within each year is often ignored. An improved ARIMA model is developed in this study accounting for both the interannual and inter-monthly variation. In the present approach, clustering analysis is performed first to hydrologic variable time series. The characteristics of each class are then extracted and the correlation between the hydrologic variable quantity to be predicted and characteristic quantities constructed by linear regression analysis. ARIMA models are built for predicting these characteristics of each class and the hydrologic variable monthly values of year of interest are finally predicted using the modeled values of corresponding characteristics from ARIMA model and the linear regression model. A case study is conducted to predict the monthly precipitation at the Lanzhou precipitation station in Lanzhou, China, using the model, and the results show that the accuracy of the improved model is significantly higher than the seasonal model, with the mean residual achieving 9.41 mm and the forecast accuracy increasing by 21%.

Download
Short summary
This paper presents an improvement on the conventional ARIMA model for precipitation time-series forecast. The precipitation time series of 12 months is first classified into several clusters. The maxima, minima, and truncation means of each cluster are then predicted using the improved ARIMA models, which are further used to predict the monthly precipitation through a set of regression models. A case study demonstrates that the present approach could increase the forecast accuracy by 21%.