If software estimators ran the National Hurricane Center
Posted by Robert Merrill on August 29, 2008 under Estimation, Project Set-Up | 
2 Comments to Read
As I was writing this, soon-to-be Hurricane Gustav of 2008 was intensifying near Jamaica, and workers were already being evacuated from oil platforms in the northern Gulf of Mexico. This was causing a loss of production (and profits) for those producers, and driving an increase in the price of oil on world markets.
Evacuating the offshore platforms was a decision. It was based on some decision support, provided by the fine folks at NOAA’s National Hurricane Center.
Tropical Storm Gustav Forecast Track and Error Cone
The solid line was the forecast track of Gustav’s center for the next three days. The dashed line extends that to five days. The white solid and hatched areas showed the expected errors in the track. Based on past performance of NHC forecasts, the center of Gustav had a 1/3 chance of going outside the threat area. This is the meteorological equivalent of a software project estimate, only presented with a lot more subtlety for use by much more sophisticated decision-makers.
As a decision-maker, there are a number of ways you could have misinterpreted this:
| Misinterpretation | Fact |
|---|---|
| Gustav will pass near New Orleans, LA Tuesday morning. | That was the most-likely scenario at this point, but it wasn’t significantly more likely than the center of Gustav being near Mobile, AL, Lake Charles, LA, still well offshore, or inland over Mississippi (not only are you learning about Software-Intensive Businesses, you’re learning the geography of the Gulf Coast). |
| I live in Brownsville, TX (the southern tip) so I have nothing to worry about. | Careful, there’s a 1/3 chance that Gustav will escape the cone. Actually, your probability of gale-force (40 mph or more) winds within five days due to Gustav is around 5%, or 1 in 20. Now you can decide if you need to take action yet, or not. |
| I’m on an offshore rig at 28 North, 89 West, and I need to be shut down before gale-force winds arrive. But Gustav’s a long way off. I should wait a couple of days before deciding what to do. | It all depends on how long it takes you to shut down operations and get your people off. Your probability of gale-force winds within 48 hours is nil, within 3 days, 7%, within 4 days, 36%, and within 5 days, 46%. If preparations take three days to complete, you need to decide if a 7% chance of being caught is worth it. Trust me, it’s not. That’s why producers were already evacuating those platforms. |
| Look at the size of the error cone! These forecasts are useless! The Government should fire those forecasters and get us some who are able to make a commitment and stick to it, rather than being so wishy-washy. | What do you think is wrong with this interpretation? |
Notice the fine print on the chart. “Advisory 15.” The National Hurricane Center updates its forecasts every six hours, because new information is coming in all the time. If you called them and complained, “But yesterday you said!” they would have explained this very politely and referred you to the current advisory.
If the National Hurricane Center were run like most software projects, they would have issued one forecast track, as soon as a storm formed, with no error cone. When the storm deviated from the track, the offshore platform managers would blame the storm or the forecasters. But wait, you say. We can control where a software project goes. We can’t control a hurricane!
Oh, really? Influence, sort of. Control, no.