Post #G21-005: Your 70th percentile last frost date is actually your 90th percentile last frost date.

Posted on April 8, 2021


This is a short note on something that I’m sure is obvious to most seasoned gardeners, but not to me.  I’ll try to keep it short.

Edit:  My wife informs me that this is more-or-less incomprehensible.  Let me try to boil it down and save you the trouble of reading the full post.

A last-frost date tells you how likely you are to lose frost-sensitive plants if you blindly set them out on that date.  For example, the Old Farmers Almanac gives you the “70th percentile” last frost date, meaning, if you set your plants out on that date, there’s a 30 percent chance a frost will come along and kill them.

But if you don’t just set them out blindly on that date, and instead pay attention to the seven-day forecast on that date, you actually have a much lower chance of having your plants killed by frost.  Because, a) seven-day forecasts are pretty good and b) if frost is in the forecast, on that date, you’ll have the good sense not to set your plants out (duh).

The only real insight here is that, in effect, the labels attached to those last-frost dates — the likelihood you’ll lose your plants to frost — are wrong. (Or, really, they are only correct if you’re dumb enough to go ahead and plant with frost in the forecast).  The forecast lets you see with good accuracy seven days into the future, nobody is dumb enough to set out their tomatoes with frost in the forecast, and the combination of those two factors means that your actual risk of having frost kill your plants is much lower than the label on that date would suggest.

In my case (Vienna, VA, Zone 7), what’s labeled as the 70th percentile last frost date (nominal 30% risk of having frost kill my plants) actually only carries a 10% risk of having frost kill my plants.  And that’s because, on that date, you’ve got a real-time seven-day look into the future via the weather forecast.

I think that’s now all completely obvious.  As I say, the only real insight here is that the labels on the dates are, in a practical sense, wrong.  Those labels (e.g., 70th percentile) come from simply tabulating the last frost dates for the prior three decades.  They don’t account for the fact that you’ll have the good sense to wait, if frost is in the forecast as of that date.  And the second but important insight is that those labels are quite a bit wrong.  Just having the common sense to check the seven-day forecast (in my location) cuts that nominal 30 percent risk of frost damage down to 10 percent.

Two caveats:  The size of this “forecast” effect depends on where you live.  In some areas, the spread between these percentile last-frost dates is larger than it is here in Vienna, VA.  And, this doesn’t help you if the planting instructions tell you to (e.g.) plant two weeks before the last-frost date.  That’s outside the forecast window (or, at least, the reasonably-accurate-forecast window.)  This is really just about plants that should be planted directly after the last-frost date.

Back to the original and now totally unnecessary post.  You might want to read the final addendum, because if you actually work through the logic here, the likelihood of frost killing your plants is even lower than what I just said.

Most gardeners do spring planting around some chosen “last frost date”.  Anyone who has ever read a seed packet has seen some variation of the phrase “plant after all danger of frost is past”.  In practical terms, that really means the earliest date at which danger of frost is past.

In reality, of course, there’s no such thing as “a” last frost date.  Every reasonable date you choose involves some risk of frost past that date, despite how the date might be presented.  If you go to the right place, you can see the full array of odds, for a whole spectrum of dates, for your location.  (N.B., that website has the percentiles flipped around from the way most people present that, so their 10th percentile is what most sites will show as the 90th percentile date.)

(For what it’s worth, there’s nothing complicated about those numbers.  Those are just tabulated off the last three decades’ worth of observations, by the National Oceanographic and Atmospheric Administration (NOAA).  So the (e.g.) 50th percentile last frost date is just the median of all the last frost dates for that location, for the period 1980-2010.)

In Vienna VA, Old Farmer’s Almanac gives me a firm last frost date of April 18.  Firm, that is, until I start reading the fine print, and see that’s actually the 70th percentile last frost date, so there’s roughly 30 percent chance of seeing a frost after that date. 

So I’m looking at that and thinking, gee, that’s kind of risky.  I always thought that publication billed itself as a font of common sense.  And here, they’re advising me to plant on a date that gives me close to a one-third chance of losing all my frost-sensitive plants? 

Then I thought about it a bit, looked up the full chart cited above, and realized that planning to plant on that date, and paying attention to the current seven-day forecast on that date, means you actually have 90th-percentile protection against a freeze.  For the simple reason that if, on that date, the seven-day weather forecast calls for freezing weather, you’ll wait.

That occurs because, here in Vienna, VA, the difference between the 70th percentile date and the 90th percentile date is seven days.  (Shown below as the 30th and 10th percentile dates, because that source labels them that way.)  But you can eliminate that 20% of frosts merely by paying attention to a reasonably-accurate seven-day forecast.


The upshot is that the Old Farmer’s Almanac gives me a date that, if used unconditionally, incurs 30 percent risk of frost.  But if used conditional on the forecast at the time — which, obviously, any idiot would do — really only incurs 10 percent risk of frost.

I’m sure that’s something most seasoned gardeners understand.  You don’t actually use those dates, as listed, to achieve that level of risk.  You use them to achieve a lower-than-listed level of risk, because you will always choose not to plant if the forecast is for frost.  You always will use a conditional planting strategy, conditional on the forecast on that date.

I guess that’s obvious.  But it does have a practical implication.  Suppose you really are risk averse, and want to incur no more than a 10 percent risk of frost.  You’d think that you’d plant your seeds on (what shows on this chart as the) 10th percentile day.  But in fact, that’s not true.  The way to achieve a 10 percent risk  of frost is to plant on the 30 percent day, as long as the forecast says no chance of frost in the next seven days.  But to delay, by contrast, if the seven-day forecast predicts frost in the coming week, when you hit that date.

Conditional probabilities can be tricky.  But I think I have that right now.

I’m not sure whether this observation has any deeper implications or not.  Doesn’t this mean that, with a somewhat-accurate 10-day forecast available, you can plant even earlier with minimal additional risk?  I can’t quite get my mind around whether that’s true or not.

And so the only practical upshot, for me, is that what the Farmer’s Almanac date is actually substantially safer than the nominal 30 percent risk of frost that it is presented to be.  For the simple reason that nobody would be dumb enough to plant, on that date, if the seven-day forecast calls for frost.  With that conditional planting strategy, the numbers for Vienna, VA tell me that I only have a 10 percent risk of frost.   YMMV.

Addendum:  Conditional probabilities are tricky.  When you work through it, if you adopt a strategy of a) choosing a 70th percentile last frost date and then b) checking the seven-day forecast for frost, you will actually end up reducing your risk of frost damage below 10 percent. (In Vienna, VA — YMMV).

The reason is that, as you move your planting day forward, you get new seven-day forecasts.  And these will, in theory, allow you to avoid even more frosts that would have otherwise killed your plants if you had just blindly planted on the 70th percentile last frost date.

To calculate this exactly, you’d have to have go back to the detailed weather data.  Even to calculate this off tabulated data, you’d have to have better tables than are published, because this already pushes you past the 90th percentile last frost date (at least, here in Zone 7 Virginia).

So let me just characterize it this way.  Suppose you adopt a simple strategy.  If there’s frost in the forecast, on the 70th percentile last-frost date, you’ll wait a week.  If you do that, you’ll then be able to see seven days beyond the 90th percentile last frost data (here, Vienna, VA, Zone 7). 

I don’t know how sparse those percentiles get beyond the 90th, but by eye, it’s plausible that another seven days would get you to the 95th.  The upshot is that your actual risk of frost damage, with a strategy that starts with the 70th percentile date (30% chance of damage), actually yields something like a 5% real-world risk of frost damage.  Depending on what strategy you adopt, and where you live, those labels (based on the unconditional probability) may really overstate your true risk of frost damage.