Page tree
Skip to end of metadata
Go to start of metadata


One of the biggest factors in determining weather impacts is probability precipitation.  It turns out that its also the most easily misunderstood metric in our weather suite.  In this article we examine the nature of the metric "Precipitation %" and discuss how best to use it and how to avoid making incorrect conclusions based on the data provided.  Read below to view the definition directly from the source. Once you understand the technicals of this information, we can discuss the practical impact of the information and how ultimately it affects consumer behavior.

Boy, is Clint right. In the context of forecasting precipitation's impact on sales and consumer behavior, it's more about whether the consumer thinks it will rain than whether, where and when it rained. Case in point - if you're a golfer, you check the weather forecast before you book a tee time. If you're like us, and you live in an area with hot, humid, rainy summers where it rains frequently, you might read anything less than 60-65% chance of rain as worth taking the risk to book the tee time. But what if that chance increases to 75%? Maybe we choose to forego the tee time and plan something indoors instead. These instances of consumer-behavior risk calculations happen constantly, as consumers evaluate near-future events. If it turns out that it doesn't rain on Saturday, we may do something outside, but it has definitely altered our planning in a big way. Additionally, if it does rain, people typically don't much care whether it's 0.1" or 1.1" of rainfall. In either instance, they're probably not golfing, mowing the lawn, planting flowers or anything else that requires spending a lot of time outside.

So forecasts matter because they affect behavior. This is basis for studying the level of impact on behavior (sales or not), not actual rainfall. But before we proceed, lets talk about one other tricky thing about analyzing weather patterns. When looking to past data sometimes we simply want to know: did it or didn't it actually rain? We've had customers say, "but I just need to know whether or not it actually rained!" This type of question highlights the both very localized and fluid nature of rain, seeing that rain storms move between 4-15mph. So "did it rain" depends highly on exactly which place time you're referring to. Because of this, rain data collection is based around a specified area, usually a zip code. Depending on where you are in the country, a zip code can greatly vary in size. If a rain storm is moving through a zip code at 10mph, and that zip code is 10 miles wide, this means that it's only going to be raining at this location for one hour out of a 24-hour day! Then we have to try to determine whether 100% of the area in the zip code was rained upon, or if it was only a portion. Rainfall measurements are good for determining that it rained, but not when it rained. They're also only able to say with 100% certainty that it rained at the point of measurement, not if it rained a mile away, unless there is another reporting measuring device in that particular location as well.

Learning all of this can feel a bit dry, but the important takeaway is that, when it comes to whether or not it rained, it doesn't matter! The biggest impacting factor is the % chance of precipitation. Determining whether or not it rained can be calculated in surrogate by assessing something similar to "if the % chance of rain was above 80% for this zip code, its likely rainy or it rained." The percentage you use as a cutoff will be based more on the region of the country. For example, if you live in the rainy southeaster United States, a 50% chance happens frequently, so it's not an effective cutoff to determine whether it actually did rain. In contrast, a 50% chance of rain in the extreme Arizona desert means it likely did rain, as 0-10% is the typical rain estimate for that region.

To sum it all up, precipitation % is a good surrogate and representative metric for rain and determining behavior, because it incorporates geography and timing into a easily digestible metric that people base their actions upon. With these few key points kept in mind, you'll be well armed in not overstating or understating the impact of these values when combining them with sales.  

Precipitation % Visualized - Proper & Improper Aggregation

Take a look at the two maps above.  The top left is out Year-Over-Year precipitation differential map.   The top right is our Days of Rain Map.  These are both two good examples of how we can use the metric of precipitation % to understand where its raining and how much for a give day or time period.  The first map indicates with red, where it is more rainy this year than the same period last year - in this case a day in June.  We can do this by aggregating up values at the zip code level to the county level and comparing to prior year before visualizing on the map.  A brief summary statement of this map would be "Its much rainier in the Deep South and Northeast / Midwest than it was on the prior year date 6/23/2016".  If you combine this with a sales heat map, for the same time period (assuming you have daily sales data at your disposal) you can see without any further deeper investigation whether sales were helped or hurt by the difference.  

Often times though you won't have daily sales data or perhaps just as likely - you're wanting to look at a sales impact over a longer time period.   The problem arises that you need to aggregate the data for rain each day into a longer running total.  This is what the Days of Rain Map above does.  It allows you to pick a longer time period - perhaps a week or two weeks (usually during peak season or weekends or holidays) perhaps and still evaluate the impact of the total number of rainy days and the affect it has on your sales performance.  Note: we've had customers (mis)use the Days of Rain map for a long period of time - 3 months or 6 months, for example, and come to some really erroneous conclusions.  

In general - weather needs to be evaluated against short term consumer behavioral adjustments.  If you live in Seattle, it rains all the time.  So much so, most people have umbrellas and rain gear.  So if you are an umbrella supplier, using the high number of days of rain over a 6mo period as a gauge for expected sales of umbrellas is fine, except where high rain amounts are common - and people have already adjusted to the weather pattern (they don't buy new umbrellas every week  for new rain).  It is something to be aware of - that, specifically for rain patterns, there are almost no non-recurring weather patterns that last more than 30 days - rainfall typically mean reverts over longer time periods.  The only known exception we've found to this is El Nino.  Most other rain patterns are short-lived - Hurricanes, Snowstorms, Blocking Low Pressure Systems - they all move in and out within a week and their behavioral impact typically is felt for a few weeks before and after at most.

Table of Contents

Learn About Popular Use Cases
  • Using Excel & Web Queries To Analyze Rain vs. Sales
    • Link to Article

Tips For Analyzing Precipitation

Ok, so a lot of talk, how then to begin to properly analyze precipitation?  The easiest place to start is with the weather report (112) named Weather Precipitation Histogram By Zip Code / Week and combine it with a Sales By Store (30) report or a YOY Sales By Item / Store / Week (56).  The 112 report is great because it gives you the number of days with precipitation % in buckets from 0-100% for each zip code and aligns it to your retailer's week (regardless of whether their week ends on a Friday, Saturday or Sunday).  Report 30 is good because it gives you store totals with zip codes.  Report 56 is also nice because it gives you store / item level by week.  Our goal is to match up the week in question by store, using the zip code.  Caution: you may be tempted to view sales for all your items. While this may be appropriate for some customers, we suggest starting with an item you have a high belief is impacted by rain patterns.  So start with just one or two items in either report 30 or 56.  Also, you may be tempted to run a report for the entire country - we suggest using the insight panels to identify specific regions of the country where rain has been identified and use those regions as subsets rather than the entire country - the report data will be smaller and there will be less data to confound you.  Remember - rain is temporal, and highly local - so make your impact analyses regional and defined short time periods with discreet items and you'll fare much better.

With these reports - we recommend making a webquery and getting them into Excel to do some vlookup and compare performance and rain values.  We've seen some customers combine the precipitation bucket values to create their own larger buckets - for example - combining any precipitation value > 70% and calling that rainy.  This is where Excel works best, this is also what webqueries are for (getting the data to Excel and refreshing easily) so use them!  

Once you have these two or three reports in Excel you can then begin looking for a correlation (either positive or negative) to increasing number of rainy days and decreasing sales performance during the week in question and vice versa!

Technical Definition of Precipitation % - Directly from the National Weather Service

Forecasts issued by the National Weather Service routinely include a "PoP" (probability of precipitation) statement, which is often expressed as the "chance of rain" or "chance of precipitation".

119 PM EDT THU MAY 8 2008

119 PM EDT THU MAY x 2008


What does this "40 percent" mean? ...will it rain 40 percent of of the time? ...will it rain over 40 percent of the area?

The "Probability of Precipitation" (PoP) describes the chance of precipitation occurring at any point you select in the area.

How do forecasters arrive at this value?

Mathematically, PoP is defined as follows:
PoP = C x A where "C" = the confidence that precipitation will occur somewhere in the forecast area, and where "A" = the percent of the area that will receive measureable precipitation, if it occurs at all.

So... in the case of the forecast above, if the forecaster knows precipitation is sure to occur ( confidence is 100% ), he/she is expressing how much of the area will receive measurable rain. ( PoP = "C" x "A" or "1" times ".4" which equals .4 or 40%.)

But, most of the time, the forecaster is expressing a combination of degree of confidence and areal coverage. If the forecaster is only 50% sure that precipitation will occur, and expects that, if it does occur, it will produce measurable rain over about 80 percent of the area, the PoP (chance of rain) is 40%. ( PoP = .5 x .8 which equals .4 or 40%. )

In either event, the correct way to interpret the forecast is: there is a 40 percent chance that rain will occur at any given point in the area.

  • No labels