Markov-Switching Linked Autoregressive Model for Non-continuous Wind Direction Data

• Published in: Journal of agricultural, biological, and environmental statistics Vol. 23; no. 3; pp. 410 - 425
• Main Authors: Zhan, Xiaoping, Ma, Tiefeng, Liu, Shuangzhe, Shimizu, Kunio
• Format: Journal Article
• Language: English
• Published: New York Springer Science + Business Media 01.09.2018; Springer US; Springer; Springer Nature B.V
• Subjects: Agriculture; Animal behavior; Atmospheric models; Autoregressive models; Biostatistics; Computer simulation; Data processing; Geography; Health Sciences; Markov chains; Mathematics and Statistics; Medicine; Model accuracy; Monitoring/Environmental Analysis; prediction; Statistics; Statistics for Life Sciences; Switching; Time dependence; time series analysis; Wind; Wind direction; Prediction accuracy; Circular regressive model; Non-continuous wind direction; Mean circular prediction error
• ISSN: 1085-7117; 1537-2693
• DOI: 10.1007/s13253-018-0331-z

In this paper, a Markov-switching linked autoregressive model is proposed to describe and forecast non-continuous wind direction data. Due to the influence factors of geography and atmosphere, the distribution of wind direction is disjunct and multi-modal. Moreover, for a number of practical situations, wind direction is a time series and its dependence on time provides very important information for modeling. Our model takes these two points into account to give an accurate prediction of this kind of wind direction. A simulation study shows that our model has a significantly higher prediction accuracy and a smaller mean circular prediction error than three existing models and it is illustrated to be effective by analyzing real data. Supplementary materials accompanying this paper appear online.