The Department of Atmosphere, Oceanic and Space Sciences Nov. 24 hosted NOAA Physical Science Laboratory John Albers for the AOS Colloquium Series. Even though his job is with NOAA, Albers specified that his seminar was not given as a representative of NOAA, rather as a program alum.
Albers’s research focuses on using applied math for atmospheric sciences, specifically he uses mathematical programs to enhance the accuracy and reliability of forecasts in the three to four week range. Climate predictions are fairly accurate on the eight to 10 day outlook, but, as you get further away from the present, it becomes more difficult to accurately predict weather — with the three to four week range being the upper-limit of decent predictability, Albers said.
Forecasts ranging two weeks to two months in advance are called the ‘subseasonal’ range, according to NOAA. These forecasts have lower “skill” numbers, which is a measure of how well a particular forecast did relative to a historical standard, according to the National Weather Service Climate Prediction Center.
“Week one [forecasts], the skill is great. Week two, everywhere is still above 0.5 [out of 1]…For these timescales, that’s great…[Skill values drop] starting at about week three and really dropping off at week four,” Albers said.
A lot of signals in forecasting are predictable, for example El Niño Southern Oscillation, Albers said. ENSO is a driving factor for season changes, and it has characteristic patterns of rainfall and temperature, according to the World Health Organization. Yet, other signals are unpredictable, like changes to solar output or human activities, according to the Canadian climate consortium Climate Data.
Combinations of various climate models can compute a variety of possible outcomes to try to more accurately predict future weather conditions, according to Climate Data. Yet, it can be difficult to parse out what causes differences between the various combinations, Albers said.
In his job, he uses a mathematical model called the Linear Inverse Model, which allows him to separate out predictable signals from unpredictable signals. This allows him and forecasters to evaluate which parts they can feel confident about and which parts of the forecast might be subject to variability, Albers said.
“In the LIM, it’s a clean split between predictable dynamics and forecast uncertainty … We can take these two pieces and calculate the signal to noise ratio and then what we call the ‘expected skill,’” Albers said.
This means he can predict how good the forecast will be, based on the expected skill value, Albers said.
