Predicting heading date: growing degree days and day length

Knowing when plots will head pays off all season. It tells you when to send someone out to score, when the crossing block will be ready, and when spray timings tied to heading and flowering come up.

Most programs record heading as days from sowing. It's easy to write down, but a poor way to predict anything, because a plant doesn't count days. It counts warmth, and in long-day crops like wheat and barley it also notices how long the days are. Counting growing degree days (GDD), and allowing for day length, fixes most of that.

Why days from sowing mislead

Take a made-up line that needs 800 degree-days above 0 °C to reach heading. Sown into a warm spring that averages 16 °C a day, it gets there in 50 days. Sown into a cool spring that averages 13 °C, it needs about 62 days.

Nothing about the line changed. The weather did. Plan the cool year's scoring on 50 days and you send people out nearly two weeks early.

The same happens between sites. A southern station that warms up early heads the same lines in fewer days than a cooler northern one, and a late-sown trial goes into warmer days and usually heads sooner after sowing.

Days to heading still works for ranking lines within one trial, where every plot had the same sowing date and weather. It breaks down when you compare across sites or years, or try to look ahead.

Growing degree days, briefly

A day's GDD is its mean temperature above a base temperature:

GDD = (Tmax + Tmin) / 2 − base, and never less than zero.

A day with a high of 22 °C and a low of 8 °C has a mean of 15 °C. With a base of 0 °C, that day adds 15 GDD. With a base of 5 °C, it adds 10. Add the days up from sowing and you have the thermal time the crop has had so far.

Base temperatures differ by crop. For wheat and barley a base of 0 °C is widely used. Canola is often counted from 5 °C. Maize uses 10 °C, with the daily high and low held between 10 and 30 °C first, because development doesn't keep speeding up in real heat.

Pick one base per crop and stick to it: a requirement of 800 GDD only means something next to its base.

Where day length comes in

Wheat and barley are long-day plants. As days get longer in spring, they move toward flowering faster. On short days, the same warmth does less.

For lines sensitive to day length, that has two effects. A late-sown trial heads in fewer GDD than an early-sown one, because its warm days are also long days. And a line can need fewer GDD at a site further from the equator, where summer days are longer.

Lines differ in how much they care. In wheat, the best known genes for this are the Ppd-1 genes: lines with insensitive alleles head with less regard to day length. A sensitive line and an insensitive one can swap order between an early and a late sowing.

Crop models handle this with photothermal time: each day's GDD is multiplied by a factor between 0 and 1. On long enough days the day counts in full; on shorter ones, only part of the warmth counts. Many crop models measure day length from morning twilight to evening twilight, not sunrise to sunset, because plants respond to quite dim light.

One caution for fall-sown crops. Winter wheat and winter barley also need a cold spell (vernalization) before they can head. A heat sum alone doesn't capture that, so everything here is easiest for spring-sown crops.

Using it with your own trials

You don't need a crop model to start, just four things you probably have already.

A sowing date for every trial, and an emergence date if you note it.
One definition of heading, written down. For example, the day half the heads in the plot have fully emerged. The Zadoks scale numbers ear emergence as stages 50 to 59, so you can name the stage you mean.
Daily highs and lows for each site from sowing to heading. Your own station is best; a gridded source like NASA POWER will do (see using your own weather station).
Your past heading notes, ideally from several sites and years.

Work out each plot's GDD from sowing to the day it headed, and compare lines by GDD instead of days. A line's GDD to heading should be steadier across sites and years than its days were.

If it isn't, look at the pattern. Fewer GDD at your most northern site, or in late sowings, is day length showing through. If one site is off every year, check its sowing dates and weather data before blaming the plants.

Turning requirements into a forecast

Once you know a line's requirement, a forecast is mostly adding up. Count the GDD each site has had since sowing. The line heads when the running total reaches its requirement. The hard part is the weather you don't know yet.

One good way around that is to play the rest of the season out with past years' weather at that site. Take this year's days so far, finish the season with one past year's weather, then with the next year's, and so on. Each past year gives a slightly different date, so together they give a range that narrows as the season goes on. A short-range weather forecast can cover the next week or so.

The requirement itself is uncertain too. A line seen in two trials is less certain than one seen in twenty. Lines with few records can borrow from their relatives, which is where pedigree and DNA help (see pedigree and DNA together).

Test it before you trust it

Any method looks good on the data it learned from. The honest test is to leave one past site-year out, predict its heading dates from all the others using its real weather, and count how many days off you were. Do that for every site-year in turn, then run the same test with plain days from sowing.

If GDD doesn't beat days on your data, look for sowing dates that are really planting-plan dates, weather from a source far from the field, or heading notes taken on different definitions.

How Crossling does it

Crossling's Outlook page does all of this with your program's own notes, and refits every night.

It turns every days-to-heading score into the development units the plot had by that day: GDD, slowed by short days for long-day crops like wheat and barley when your data show it helps.
A mixed model splits each requirement into an average, the line's part, the site-season's part and plot noise. The line's part carries over to a new season; a line with few records leans on its relatives (pedigree, and DNA where you have it).
Day length is tried at a few strengths, starting from none, each tested by leaving every past site-season out in turn. A day-length setting is kept only if it cuts the error by at least 3% compared with plain GDD.
To forecast, it adds up each site's units so far, uses MET Norway's forecast for the next days, then each past year's weather at the place for the rest. Every plot gets a middle date and a range. Plots already scored this season tell the model how this site-season is running.
The model card shows how far off it was, in days, on seasons it hadn't seen, beside the same test for plain days from sowing.
Today gives a heads-up up to two weeks ahead, such as "Heading starts at Lacombe around 12 Jul", which you can turn into a scoring task in one tap.

GDD bases follow the crop, from 0 °C for wheat and barley to 10 °C for maize. Weather comes from your own station where you have one, and NASA POWER for every other day.

The forecast starts once a trait has been scored at three site-seasons and on 30 plots, in trials with a sowing date and a location with coordinates. You can also look back at a past season as it would have looked on any day, or plan a season ahead: which lines in a crossing block are likely to flower together, for example.

The demo's heading notes at three Alberta stations are simulated on each season's real weather, so the Outlook page has something to show. Try the demo to see it, or create an account and add your own trials.