Looking at Past Weather Data with Time Series Plots
Christmas is right around the corner and the news media is abuzz about retail sales forecasts for this year's holiday shopping season. As much as I hate to admit it, another type of forecast is also on the horizon—a forecast for impending winter weather!
Last year’s winter weather started early, and you might remember that New York City got hit with quite the winter storm the day after Christmas ‘10. In fact, New York City’s December snowfall total for 2010 was a whopping 20.1 inches!
While meteorologists will be releasing their winter weather predictions in coming weeks, I thought it might be interesting to use Minitab time series plots to graphically view past snowfall histories for NYC, Philadelphia, and Washington, D.C. You know what they say—“Learning about the past can help you prepare for the future.”
To start, I visited NOAA.gov to gather monthly snowfall totals for December in New York City (Central Park), Philly, and Washington, D.C. from 1991-2010. I recorded the data I collected for each city in a Minitab worksheet:
In Minitab, I chose Graph > Time Series Plot > Simple to get a graphical representation of the snowfall data. Time series plots are a good graphing option when you have a collection of data where each observation is uniquely determined by a single point in time. They’re also really helpful for quickly viewing and finding patterns (visually) in time series data.
What can we infer from this time series plot?
While the amount of December snowfall in NYC is variable from 1991-2010, I see a pattern that shows steep increases followed by steep decreases. It seems that after a December with a larger snowfall, the following December has significantly less snowfall. However, I’m sure there would be several exceptions to this pattern if we included more years of December snowfall data to our time series plot—say dating back through 1900.
You can see from the time series plot (with multiple series) of both Philly and Washington, D.C. that a similar variable pattern of sharp increases, followed by sharp declines exists. Also, notice that December snowfall amounts for both cities have either increased or decreased together. From 1991-2010, not once did one increase while the other decreased from the previous year! The correlation between the two measurements is 0.863, so I think it's pretty fair to assume that this December's snowfall amount will either increase or decrease in both cities.
I’ll leave the real weather forecasting to the professionals, but I think that due to the proximity in location between Philly and D.C., what happens weather-wise in both of the cities is likely to be highly correlated. (and the plot above and correlation calculation seem to reflect this)
Speaking of forecasts—one thing you can’t do with this data is develop a forecast model using the trend analysis feature of time series plots in Minitab. You can tell from the time series plots above that a trend analysis will not be helpful because there is no real trend showing the snowfall amount generally drifting up or down over time.
While it’s tempting to make forecasts based on the trend analysis graph below, this data doesn't lend itself to an overall trend in one direction or the other.
A valid trend analysis plot in Minitab might look like this:
This forecast is acceptable because the employment data collected in this example is trending upward (and not all over the place like the snowfall data).
How have you successfully used trend analysis in the past at your company?
Snow photo by jdurham, used under Creative Commons 2.0 license.