Shih-Hao Su, Chiao-Wei Chang, Ting-Shuo Yo, Yi-Chiang Yu and Jung-Lien Chu
The precipitation associated with the synoptic fronts is not only one of the major water resources but also the main cause of natural disasters in Taiwan. Earlier studies indicated that the frequency of the seasonal front systems is correlated with large-scale variations and multi-scale climate oscillations. This suggested under climate change scenarios, the frequency and properties of the frontal systems may change along with the large-scale environment.
Su et al. (2018) proposed a machine learning weather classifier detecting the frontal influence frequency near the Taiwan region. This approach was shown to outperform the traditional objective method in both the accuracy and the hit rate (Chang et al., 2019). In this study, we used reanalysis/hindcast data and machine learning weather classifier to investigate the long-term statistical properties of front systems near Taiwan. Besides historical reanalysis data, we also applied the same technique to selected CMIP5 model dataset used for climate projection in the IPCC Fifth Assessment Report (AR5).
The preliminary results from MRI-CGCM3 (Yukimoto et al., 2012) showed that the derived statistical properties can be affected by model features such as model resolution and parameterization schemes. Compared to the reanalysis (NCEP-CFSR, Saha et al., 2010), the historical run of MRI shows a lower probability (5.6 days per year) of frontal occurrence. It also shows a decreasing trend of frontal event from current, near future (4.4 days per year) and far future (2.1 days per year).
In order to standardize the effects of model properties, the climatology anomalies were used as inputs to identified the weather events. The modified method is further applied to estimate the frontal precipitation changes/variations under different climate change scenarios.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner