H41A-0771:
A Markov track model for simulating Typhoon Tracks in North-Western Pacific Ocean

Thursday, 18 December 2014
Byunghyun Song, Seoul National University, School of Earth and Environment Science, Seoul, South Korea; Korea Meteorological Administration, Seoul, South Korea, Balaji Rajagopalan, Univ Colorado, Civil, Environmental, and Architectural Engineering and Cooperative Institute for Research in Environmental Sciences, Boulder, CO, United States and Gyu-Ho Lim, Seoul National University, Seoul, South Korea
Abstract:
Typhoons typically occur during Jul-Sep, and the average number of occurrences per year in North Pacific and land falling over South Korea is ~25 and ~3, respectively. They cause enormous damage to infrastructure and lives and more so to developing countries in the region. Hence, to manage and mitigate this natural hazard, robust estimates of typhoon risks are necessary. 

Historical data is limited in length and incomplete in its variability. Thus, a methodology to generate a rich variety of realistic typhoon scenarios and the associated land fall risk along the coastal regions of their impact is important. 

To this end, we propose a spatial Markov track simulation model. In this, the domain is divided into 5x5 grids, and a typhoon in a grid box has ten states to transit to in the following 6-hour period – they are moving into one of the 8 neighboring grids, staying or fizzling out in the same grid. Based on historical 6-hourly typhoon track data (for the period 1977 – 2013) transition probabilities (i.e. probability of transitioning to one of the aforementioned ten states from the current 6-hour period to the following 6-hour period) are compute. 

Typhoon origination probabilities are also computed for each grid box. A track is initiated from one of the grid boxes based on the origination probabilities and it is propagated by the spatial Markov transition probabilities. A large number of synthetic tracks are generated for the spatial Markov probabilities from which estimates of land falling risks can be computed for different coastal segments. 

Wind speed magnitudes at each time step are generated by K-nearest neighbor resampling of the speeds of the historical tracks within each grid box based on the speed of the previous time step. A statistical distribution such as Weibull, can also be fitted for each neighborhood sample to simulate wind speeds. The Markov probabilities can also be estimated conditioned on large scale climate features that impact typhoon tracks to then generate track scenarios for any given year to enable seasonal projection. Extensions to simulating precipitation and consequently, streamflow and flooding risk are fairly straightforward. Our results show good representations of historical typhoon properties and thus promises to be an attractive methodology for natural hazard estimation.