Why do weather forecasts vary so much?

Chris Hovanic, Meteorologist, and Brian Koochel, Meteorologist, Weather or Not, Inc., Shawnee, Kansas

Predicting the weather remains challenging and complex. Increasing winter maintenance costs such as gas, materials and labor, and more importantly, public safety, demand that the public works professional keeps the margin of error as minimal as is humanly possible.

History of Modeling
The advent of Numerical Weather Prediction (NWP), or Modeling, in the 20th century added a new tool for meteorologists making weather predictions. They can give an overview of a large low-pressure system or a more specific view of a column of air over a city. NWP began shortly after World War II, but the largest advances in this field occurred during the 1970s and 1980s when supercomputing technology increased. Today's computers are much faster, the graphics more animated and the temptation to rely solely on the computer for weather answers even greater. While improved computer technology offers weather companies the option to hire fewer full-time employees (FTEs) such as meteorologists, is this a good business model for the public works professional to be relying on for safe, cost-effective winter road maintenance?

Meteorologists have a plethora of models to choose from to aid in their forecasting. Models are one of the many forecasting tools available along with satellite imagery, radar imagery, real-time surface observations, RWIS and good old fashioned communication from drivers in the field or folks in the town where the storm is coming from. As such a tool, models are generally referred to by forecasters as "guidance."

However, the atmosphere remains a difficult entity to model as there are a large number of quick-changing variables to take into account. Some of these cannot be accurately accounted for in the basic equations which make up the "brains" of modern-day computer models. The key to intelligent use of such computer model forecasts comes from a working knowledge of which model handles different weather situations the best. So the key discussion to have with your weather vendor who is providing you with graphics, tables or text derived from a computer model-based forecast would be how the model takes your location's unique weather needs into account.

How Modeling Works
As you increase the computer's power and crunch more numbers, the forecast model will provide a large number of solutions. In order for a forecast model to have any hopes of verifying, it must have an accurate starting point. This is called "initialization" and involves surface and upper air data sets, along with remote sensing data gathered from satellites. Once this data is ingested, the computer model will run its analysis to create a forecast. It is possible that the model's analysis won't accurately represent the actual state of the atmosphere. For example, the initialization shows heavy snow in North Platte, NE when the reality is that the sun is shining there and the heavy snow is occurring over Denver, CO. Any error in the initialization increases exponentially with time. Experienced meteorologists know to look at this first so they can decide if they have any faith in the model or to throw it out of the possibilities of their forecast.

Different models have slightly different equations or parameterizations (its brains) that allow it to attempt to forecast the future state of the atmosphere. It is possible for some atmospheric features to be over-forecast or under-forecast by the models. Whatever the outcome, the model's output will be a function of its initialization coupled with the model's brains. It's up to the forecaster to decide which among all of the models will be correct or has a better chance of being correct. Experience matters greatly. For example, in the winter of 2004, models were making it look like 12-24" of snow were going to arrive in the Kansas City area over Super Bowl weekend. Some forecasters were shouting "Storm of the Century!" However, keen forecasters remembered that the model's projections had been off by 60-100 miles northward for the last several storms. Since this included the temperature aspect of the forecast, the wise meteorologists predicted a mixture of rain and ice with no snow in the southern portions of Kansas City and less than 6" inches in the Northland. This "live, experienced look" at the models saved thousands of dollars for public works departments.

Why Differing Forecasts
So, now it makes sense that when you turn on the television or go on the Internet, you could see a varying number of different weather forecasts at any given time. A forecaster can choose to follow a forecast model's output in several different ways. They can believe it explicitly, adjust it some to what they think will happen or disregard it completely. Forecasters can also look to see if there is agreement in the output between several different models or if there is consistency from model run to model run within a single model or among a group of models. In other words, the model did really well the last two snowstorms but not in an ice situation.

The information we're giving you here may put more variables in the hands of the Snow Boss than he/she wants. After all, you just want to know when, how much and what kind of precipitation to expect. The reality is that computers are continuing to place more data and graphics in front of you. The better discussions you have with your weather vendors, the more confidence you can have in their products.

Chris Hovanic and Brian Koochel can be reached at (913) 722-3955 or info@weatherornot.com.

What Questions Should Public Works Professionals Ask Their Vendor?

  • Have computer models been better or worse at forecasting snow versus rain, ice versus snow, ice versus rain for your location? What are your models' advantages regarding the kind of weather problem we face here? How long was the situation studied or how long has your Snow Boss been comparing the two?

  • How has their computer model for weather forecasting been adjusted for your geography and topography or is the model the same for every region in the United State and Canada?

  • Does the vendor's computer model know when a forecast is in error or isn't verifying? How does it adjust for this and when does the adjustment take effect, the next model run (6 or 12 hrs.) or when the model undergoes a major revision?

  • When did the weather model last undergo a major revision? Was it put to an objective analysis by the meteorologists and mathematicians that may have designed it and if so, how often? Annually, seasonally, etc.?

  • Do your vendor's forecasts have live input from on-duty meteorologists around the clock before the forecast reaches you or have the forecasts you're receiving been completely derived from their computer models?

  • If there is a question regarding the forecast you're receiving, can you call a live meteorologist and ask questions?