I’m really impressed by the accuracy of medium-range weather forecasts. (I did some analysis of satellite-based observations relevant to medium-range forecasting in a previous job. The work got me interested in quantitative forecasting.) One forecast of particular interest to me is the National Weather Service’s (NWS) 5-day Quantitative Precipitation Forecast (QPF). Okay, so I haven’t kept of log of predictions vs actual but my sense is it’s pretty accurate – they predict a couple inches of precip over the next five days and we get something close, i.e., within a factor of two and much more accurate than if you just assumed the next five days would be the same as the last five or if one just assumed historical average. One’s sense of things can sometimes have little relation to reality however and I’d very interested to see a quantitative assessment of QPF accuracy. Someone must be doing it but no idea who. But I digress… The motivation for this post is that the NWS has introduced an experimental 7-day QPF!
Seven days. Wow. From a weather forecasting perspective that’s a long way out. The map is certainly visually appealing. It’ll be interesting to how it compares with reality.
The main thing is to get a dang NWS job! Nail it. Miss it. It don’t matter. The checks keep coming in. I got a gubmint job, but for some reason I have to hit the target ever dang time. Not fair.
Three-day ACC about 0.98? Five-day about >0.90? And that was in 2002! Granted, the stakes aren’t as high, but the numbers look pretty damn good to me. For what it’s worth, I often find reading the Forecast Discussion of the NWS forecasts more instructive than reading the forecasts themselves. The forecasters will get into what model outputs they believe or don’t believe and why. The “why” is what makes it interesting.
(And I’m pretty sure you need to know their secret handshake in order to get an NWS job. It’s still a mystery to me.)