Data
Background Perturbation Ratio
To simulate realistic uncertainty in numerical weather prediction (NWP) models, the background field was generated by adding intentional systematic bias to the observed Best Track data.
The applied bias was designed to generate errors in the typhoon’s position, intensity, and movement characteristics.
Latitude and longitude biases induced a continuous northwestward deviation in the predicted trajectory, while the positive pressure bias produced a weaker typhoon intensity compared to the actual observations. In addition, velocity bias was added so that the background field moved slightly faster than the real typhoon.
This background field configuration was designed to reproduce representative systematic errors commonly found in NWP models. Therefore, it provides an environment for evaluating how effectively EnKF-based data assimilation can correct forecast errors and recover the typhoon trajectory.
Maysak data
We focus on performing trajectory prediction of Typhoon Maysak that happened in 2020
Background information:
Classification: A powerful tropical cyclone tat reached Category 4-equivalent intensity, officially classified as a Typhoon(TY) and labeled "Very Strong" at its peak.
Active period: August 28 to September 3, 2020
Peak Intensity: Recorded a minimum central pressure of 935hPa and maximum sustained wind speeds of 49 m/s.
Key Parameters for Modeling:
Trajectory: Transitioned from North-Northwest(NNW) to North-Northeast(NNE) direction, eventually making landfall on the southern coast of the Korean Peninsula.
Size: The Gale Radius (Radius of 15m/s winds) expanded significantly, peaking at 380km during its most intense phase.
Variable Movement: Movement speeds fluctuated greatly, from a slow 3km/h during early development to a rapid 70km/h during its extratropical transition.
This project utilizes Maysak's observation data--specifically latitude, longitude, central pressure, gale radius, direction vectors and velocity--to evaluate the performance of LSTM model against a linear extrapolation model.
LSTM Training Data
We use typhoon data from 2016 to 2024
1. Nanmadol
2. Talim
3. Prapiroon
4. Malakas
5. Rumbia
6. Soulik
7. Kong-Rey
8. Chaba
9. Noru
10. Trami
11. Haishen
12. Bavi
13. Danas
14. Francisco
15. Lekima
16. Krosa
17. Lingling
18. Tapah
19. Mitag
20. Jangmi
21. lupit
22. omais
23. chanthu
24. aere
25. songda
26. trases
27. hinnaminor
28. nanmadol
29. khanun
30. jongdari
31. shanshan
Variables we use:
-
Latitude & Longitude: These coordinates are essential for precisely mapping the typhoon's real-time position and historical movement across the ocean.
-
Radius of Gale-force Winds: This variable defines the extent of the area at risk, helping to estimate the potential scale of damage and the size of the impact zone.
-
Direction of Movement: Tracking the heading is critical for predicting future landfall locations and issue timely evacuation warnings to coastal regions.
-
Movement Speed: The translation speed determines how long a region will be exposed to extreme weather conditions, directly affecting the severity of flooding and wind damage.