Visualizing Hurricanes

by Richard Wesley
Tableau Software, Inc.

Contents

• Introduction
• Data Processing
• Analysis
• Gallery

A few months back I came across a site called RealClimate that provides articles written by a group of climatologists. The authors wanted to provide a forum to present issues of climate change in a manner accessible to laymen (such as myself). Of late, the site has had a number of articles on the possible links between Anthropogenic Global Warming (AGW) and hurricanes ("tropical cyclones" or TCs in the jargon). One recent article entitled Storms and Global Warming II discussed a study published in Nature by Kerry Emanuel entitled Increasing Destructiveness of Tropical Cyclones over the last 30 years.

The basic idea in the paper was to integrate the cube of the wind speed over the lifetime of the storm. There are some simplifying assumptions made here, such as assuming constant values for things like drag from the ocean's surface and size of the storm. Most of these seemed reasonable to me, but I have to admit I find the last assumption troubling.

I thought it would be interesting in light of Katrina to see if I could reproduce his results using data from the National Hurricane Center on my desktop. What follows is a description of how I went about it.

The first thing I needed was records of Hurricane activity for the last 30 years or so. Happily, the National Hurricane Center has a text file that you can download which contains observational data for the North Atlantic going back to 1851.

Now that I had the data, I needed to get it into a database engine and clean it up a bit. Excel can't handle dates earlier than 1900, so I uploaded the data into a MySQL 4.1 database server. From there I was able to clean the data up, which consisted of:

• Defining numeric columns for latitude and longitude and populating them from the text values. A few values were outside the [-180,180] longitude range and had to be normalized.
• Defining a datetime column and populating it from the year-month-day hour entries. MySQL was particularly well suited for this because of its collection of date manipulation primitives. There were a few other minor problems, including a day that was 0 instead of 8 and the fact that the year is actually a season identifier, and in 1954 the season extended into January.
• A number of pressures readings were nonsensical (i.e. the surface of Mars and some were negative. I believe that the negative values are an encoding for NULLs and those were set properly
• Some wind speed and other measurements were also negative and mapped to NULL.
• Some of the storm categories were slightly mistyped.

Now that I had the data in a useful form, it was time to start crunching it. My main tool for this was a data visualization product that my company (Tableau Software) produces called Tableau. Using Tableau, I connected to the MySQL database and started implementing the analysis described in the paper. The basic equation was just the integral of the cube of the wind speed, but the wind speeds prior to 1970 had to be corrected for with a quadratic. I defined a calculation "VPrime" that contained this correction and another calculation called "Energy" that contained the cube of this value. Integration of a set of discrete values is just a sum, which Tableau can plot, giving an unsmoothed plot of energy against year:

The next step in the paper was to apply two 1-2-1 filters to the time series data. Unfortunately, Tableau will not (yet!) do filtering, so I exported the data into an Excel file and added two filtering columns. Tableau could then connect to the Excel spreadsheet and plot the smoothed results:

I am happy to report that this is the same plot obtained by Emanuel in his paper.

Tableau can be used to make all sorts of visualizations, and here is a little gallery of seasonal data plots from the same database. All the plots are the tracks of the hurricanes, with wind speed encoded as the width of the line. Since Tableau can render polygons, we should start with a plot of the North Atlantic to give some context for the rest of the data:

First, the infamous 2004 season:

Next, the very active, but mostly unremarked 1995 season:

Notice that just because a lot of energy is disappated in a season does not mean that it caused disproportionate amounts of damage to huamn settlements.

Last, the very active 1969 season, including Hurricane Camille: