This is part of my advisor's (Dr. Jenni Evans) Tropical Meteorology class. It is a group project of three people - Jayma Hamilton, Chris Johannesson, and I. Files are in PDF format.
There are lots of challenges in forecasting tropical weather:
Tropical Cyclone Track and Intensity Forecast: Public views can be over critical on the errors in tropical cyclones (aka Hurricanes or Typhoons) forecasts, despite of the big improvements of TC forecast skill last two decades. Tropical cyclone intensity spends nearly most of its life in open ocean where data collection sites (buoys, airplanes and ships) are sparse located. Even with the vast improvement of satellite coverage and satellite data analysis techniques (i.e. satellite calibration, improved remote sensing technology, better understanding to radiative transfer physics, etc.), satellite data are not perfect. Computer models may lack to the correct physics (i.e. Cumulus parameterization, boundary layer physics) or resolution to resolve tropical cyclone structure properly.
Forecasting with skill vs. Forecasting without skill: The tropics are calm and weather rarely changes dramatically (except near the vicinity of a tropical cyclone). Weather is always like that... Hot humid mornings, chance of thunderstorm in late afternoon or evening.... There is hardly any skill in such forecasts, but is there a better possible forecast? How about "Hurricane Andrew is forecasted to move NW at 15 knots next 12 hours towards Miami" or "satellite data indicates the Southwest Monsoon is strengthening, expect monsoon rains to arrive at Mumbai in the next 2-3 days"? Forecasting skill is determined by the ability to forecast events that are unique and hard to predict. In the book "Statistical Methods in the Atmospheric Sciences : An Introduction" by D. Wilks, it has a good story about skill in Tornado forecasting in the Great Plains in the US. Basically, one person claims he has great skill in forecast Tornados, and has statistics to "back" him up. The problem is his statistics are skewed by the "No Tornado" forecasts - i.e. it is better bet that a tornado will not develop today than a tornado will develop. Statistics are only useful if 1) it is completely honest (a difficult task indeed) 2) it is completely unbiased, and the analysis technique are designed to be that way (i.e. use the correct method for the right problem).
Click here to return to main page.