Ensemble Weather Models
Operational (deterministic) weather models bore me. Please put the excitement back into my life by discussing ensemble models. Can I find any of these models in the Victoria Secret catalog?
I could answer yes, but I’d be lying. Unless Victoria Secret is a clandestine weather modeling agency and a sobriquet for the National Centers for Environmental Prediction (NCEP), you are out of luck.
Besides sounding “Francophile cool” (oui!), ensemble models play a very important role in weather forecasting. Ensemble models are run as a series: each individual member, of which there are 20 or so (depending on the model of choice), of the ensemble uses slightly different initial conditions during each run. Consensus agreement among the majority of individual ensemble members’ output results in a higher confidence of finding the correct forecast.
Below we’ll look at some of the ensemble model suites commonly used in forecasting:
GFS and ECMWF ensembles: Forecasters use “spaghetti plots” to locate positions of major upper air and surface features of the atmosphere. A spaghetti plot looks exactly like its name: strands of spaghetti representing the track or location of each ensemble’s individual member output. Rather than attempting to explain this (or create pasta) from scratch, a great example of ensemble use is explained at Penn State University’s World of Weather site and go to the “5b Ensemble Forecasting” link in the Table of Contents navigator. Starting midway down the page see the images that contrast the differences in agreement among ensemble members at a model run time of 24 hours versus 168 hours.
Even with their differences, sometimes ensemble models will hint at or even lock onto a solution much earlier than their operational model counterparts. If you read anything on the CWG blog mentioning that ensemble models are consistently showing XYZ, pay attention--regardless of what the operational models are showing.
Weather conditions can change quickly. Are there any models that can help us hone in on an accurate, short term forecast?
Yes. No. Maybe. It depends. With the increase in computing power and understanding of atmospheric dynamics, short term forecasting products have become more of a priority in recent years. Here are short descriptions of a couple of the more frequently referenced short term forecast tools:
SREF (Short Range Ensemble Forecast) – The SREF is a great example of an ensemble model suite. It only forecasts out to 87 hours, but it does so in 3 hour increments. It is updated four times daily (03Z, 09Z, 15Z, and 21Z).
The SREF is available from different sites, one of which is included at Penn State University’s E-Wall (prepare your eyes for overload). In this particular iteration, all 21 SREF ensemble members are displayed by clickable parameter and region (surface, 500mb, temperature, etc.), including their mean (average) solution.
Additionally, the SREF’s output provides something known as a “plume” chart which you’ll see referenced frequently on the CWG blog, especially in winter when frozen precipitation type and amount becomes a concern. A plume chart is a data (temp, precip type/ amount, etc) versus time graph that plots selected SREF members’ output for that particular data type. It’s possible to plot all 21 SREF members’ output on the same graph to see how closely they agree. The areas of the chart where all the lines are close to or overlay each other show a high confidence forecast; if the lines are all over the place, then the forecast confidence is low or non-existent. Rising lines means data are accumulating (changing); flat lines mean the data remain unchanged.
I’ve learned to forecast like a pro using ensemble model plume and spaghetti plots with no sauce. What if I want something very short term so that I can accurately forecast conditions for the next few hours and impress my neighbor’s dog?
This FAQ is officially going to the dogs now, eh? Your neighbor’s dog would probably be more ingratiated to you if you gave it something tasty and edible rather than short range forecast printouts. Still, there is something you can use to pretend you know what’s going to happen near-term:
HRRR (High Resolution Rapid Refresh) – After serving as an experiment for some time, the NCEP’s HRRR was officially granted operational status at the end of September 2014. (Some subroutines of the model remain experimental.)
The beauty of this model is its high resolution. It can “see” the atmosphere in 3 km horizontal chunks – very high resolution indeed. Coupling that with its hourly update cycle, discrete parameters (lifted index, lightning, vertical velocities, etc.), and optional regional view, results in a very powerful package. It is a short term model, however, and covers only 15 hours worth of forecast time. There is also a regional version available that shows certain forecast parameters in 15 minute increments (primarily temperature, radar reflectivity and wind info)
This is the go-to model to see if your late afternoon outdoors event will get washed out. There are times where the HRRR gets too conceited for its own good and shows a solution that looks like it’s out to lunch. Sometimes it is, but other times that lunch winds up being surf and turf. Its high resolution can nail forecasts that other models miss completely.
Frequently Asked Questions (FAQs)
An Introductory Guide