Sunday, December 11, 2011

All I Really Need to Know I Learned From Forecasting

Following on a post from earlier this week with warnings about not getting too literal with science analogies for daily life, here are some more lessons from "The Forecaster's Way" that I do think are useful.

These are less centered around forecasting as a science, but rather forecasting as an operational job in which deadlines are tight and pressure can be high. "Operational" means basically that someone is relying on the forecasters' products and expecting them to be there when the products are needed.

Perfect is the Enemy of Good (a quote from Voltaire)

See also, "perfect" is the enemy of "finished." Hydrologists need to triage and prioritize to get their jobs done. Don't we all have to do that though? Forecasting deadlines are hard and timeliness matters. Remember the scene in the movie "Trading Places" where the commodity trading pits go into a shouting frenzy when the results of latest orange juice crop survey are announced? The official didn't even finish "The cold winter has apparently not affected the orange harvest..." before the traders erupted into hair pulling and eye gouging.

It can be a little like that in the trading pits when Western US water supply forecasts are released because they affect the prices of hydropower and natural gas. I remember that the forecast release schedule was announced months ahead of time so that any individual trader didn't get an unfair advantage by being the first to know the forecasts were out. 

Therefore, the hydrologist spending all his time polishing a given forecast to perfection is probably not giving enough attention to all the other forecasts that have to go out the door. Pace yourself and plan ahead, but most of all, learn when to let it go. Furthermore, when trying to build new forecasting systems and adopt new technologies, someone may want to wait around for, say, the perfect model, but in the meantime he may be passing up opportunities to use perfectly adequate models.

Another way of looking at it is "Forecasts are never finished, only abandoned" (apologies to da Vinci). There are many variants on this, "Dissertations are never finished...", "Reports are never finished...". As an aside, problems do start when my idea of "good enough/done enough" is different from yours. Indeed, I seemed to have this problem every time I had someone do remodeling or house repair. I can't tell you how often I have been left with abandoned jobs.

Periodically Test Your Backup Plan 

Reliability and resilience are highly valued qualities in operational systems. When things are at their worst, people with operational responsibilities have to be at their best. Users do not usually accept hardware malfunctions as an acceptable excuse for forecasts being late or not being issued. It is your job to make sure that either your computer does not crash or that you do not lose all your work when it does. Naturally, there are tradeoffs- for example, US government agencies' stringent computer security policies are enforced as if by a stern martinet. I know examples where this has stifled innovation.

This plan should have been tested. 
This is not to say what your backup plan should be, but rather that you should test that it actually works. Few things are more frightening than pulling the emergency release handle and accidentally having it come off in your hand. Maybe you bought an external hard drive to backup your computer files? Have you ever actually tried to restore a file, in a non-emergency situation, just as practice to see if it works? You should! When backup plans fail is when problems turn into fiascoes.

This checking takes discipline. As Andrijanto, an Indonesian dam operator, said to me once, "The most important thing in water management is discipline."

No comments:

Post a Comment