Detection of Pressure Patterns Using Support Vector Machine: Winter Type Pressure Pattern

Abstract

In climate research, when climatologists need to know on which days a specific pressure pattern occurred, for example, “low in the west and high in the east (winter type)” or “high in the south and low in the north (summer type),” a huge number of surface weather charts must be scanned visually. To solve this problem, we propose an automatic detection method using pattern recognition developed in the computer science field. In this study, we used support vector machine (SVM), which is one pattern recognition method. Then, we classified pressure patterns into the “winter type” and “not winter type.” We confirmed the validity of the proposed method experimentally. In our experiments, we used the JRA-25 data from 1981 to 2000 for training data and test data.
The results showed that our SVM method achieves greater than 90% accuracy. Therefore, automatic detection of pressure pattern is possible by setting positive and negative examples based on a standard. Additionally, long-term data can be classified easily using our proposed method, and the results can be used to evaluate changes in the frequency of pressure patterns.