Tropical Cyclone Track Probability
Historical probability of a tropical cyclone crossing various locations around the world
Robert Hart, Florida State University Meteorology, rhart@fsu.edu
Funding provided by Risk Prediction Initiative (RPI) of the Bermuda Institute of Oceanic Sciences (BIOS),
and the Florida Catastrophic Risk Managment Center.
[ MOE MAIN PAGE | TCPROB MAIN PAGE | HELP/DOCUMENTATION | IMPORTANT DISCLAIMER ]
Last updated: Thu Nov 21 08:48:17 UTC 2024
Summary:
This page uses the past several decades to century of historical tracks to estimate the likelihood of a tropical cyclone in a given location eventually crossing various land areas.
Method:
The 6-hourly historical best-track dataset for the Atlantic, Eastern Pacific, and Western North Pacific were linearly interpolated to 10-minute increments. A 0.0833 degree (~6-10km) land-sea mask for the world (from http://polar.ncep.noaa.gov/sst/ophi/Welcome.html) was used to determine when each 10-minute interpolated position was over water or land. Tracking continued over land, such that the probability shown is for crossing an area, not only landfalling. For example, a tropical storm crossing New England after making landfall on the Gulf Coast still counts as "crossing" New England for these probabilities.
Additional notes:
- Over-land tropical cyclone probabilities listed in the tables are Any Intensity / 64+ kt / 96+ kt
- A blank in the table means that all three probabilities are less than 1%.
- Shading in the table cells indicates "Any Intensity" threat at 10% (yellow), 25% (orange), 50% (red), and 75% (pink) thresholds.
- If a 1°x1° cell had less than 3 independent tropical cyclone occurrences, that cell was not used in the calculation shown here.
- A 9-pt smoother was applied twice to the 1°x1° grid, which leads to an underestimate of the risk close to the verification region. In other words, regions immediately near the coastline of verification regions should be much closer to 100%.
- Arrows in each table cell indicate the 12 hour change in the probability. ▽/△ indicates ±2 to 4%, ▼/▲ indicates ±5 to 9%, and ▼▼/▲▲ indicates ±10% or larger change. No arrow indicates means there has been little change in the probability in the past 12 hours, or the storm did not exist 12 hours ago.
- Landfall risk maps for Europe, Greenland,and Iceland have been added. These climatologies were derived by extending hurricane tracks using reanalysis datasets (1958-1978: JRA55; 1979-2016: ERAI). When the images for these regions state "Best-track", they are indicating the official TC best-track database extended here using the reanalyses stated.
- No account is made for the evolution of observing or tracking technology over the period of record. This means that probabilities over the ocean distant from land masses likely have considerable biases -- perhaps including an overestimate of risk given storms far out to sea prior to satellites were less likely to reach land if undetected.
- Please remember that a TC doesn't have to make landfall to greatly impact a region. As a result, the probabilities here do not account for historical situations where a TC brushed land, but the TC never actually made landfall.
- Although Atlantic and East Pacific INVEST regions are plotted on the maps and probabilities displayed, the probabilties for INVESTs are obvious overestimates since a greater percentage of INVESTs do not develop than TDs and, further, no archive of INVESTs was included in the historical tracks from which the probabilities are calculated here. In other words, the probabilities for INVESTs should only be considered valid if they become TDs or stronger.
- We greatly appreciate the funding support of the Risk Prediction Initiative (RPI) of the Bermuda Institute of Oceanic Sciences (BIOS), as well as the Florida Catastrophic Storm Risk Management Center at FSU.
- Users may be interested (if they are not already aware) in the famous "Hebert box" which subjectively diagnosed a similar approach for southern Florida hurricanes in the 1970s.
- This web page and its method are unconnected to the work of Brettschneider (2008; JAMC) [which was published while this work was well underway unbeknowst to this author in 2007]. The method published in his 2008 paper is more robust than that used here and web users of this page are encouraged to read it and to use that method for applications requiring more precise, robust results.