Growing season length

[beckybanbury]

@teixeirak @nkunert Initial results using climate data only from growing season months are here. This is so far only using CRU data, as this is in the easiest format to match with month-by-month growing season data.

[beckybanbury]

I've now updated this using monthly mean productivity during the growing season - doing this actually seems to control out the effects of climate.

[teixeirak]

That's not too surprising. Actually, to make a more appropriate growing season correction, we could also correct for daylight hours of the growing season, which shouldn't be complicated. You should be able to find a formula fairly easily, including in my Environmental Biophysics book (smallish, mostly black, probably on the shelf in my office, otherwise somewhere int he lab).

This correction (divide by daylight hours) would probably re-introduce temperature and other climate sensitivities (i.e., high latitudes will have lower productivity per hour because growing season days are much longer there than in the tropics).

This would actually provide a nice robust test of the hypothesis (e.g., Michaletz) that productivity per growing season hour is independent of temperature, and give us more to discuss.

[teixeirak]

Becky, the way you're setting this up now is ideal for comparison with some of these other studies, but an alternate way to set it up would be to include fraction of year with growth-suitable conditions (or such), which would be [daylight hours during growing season]/[total hours of year]. You could then include this as a regular variable and compare its performance with others. Of course, you'd still want to include only growing season climate of the other variables. We can then test whether this model gives us better results than the annual values.

Also note-- if you're able to get monthly solar radiation, that could substitute daylight hours.

[beckybanbury]

I have monthly solar radiation; how would you use that to substitute daylight hours?

I'm looking at deriving daylight hours currently.

[teixeirak]

Monthly solar radiation should probably correlate closely with daylight hours... (I think!). One difference would be driven by cloud cover. Ultimately, the plants are responding to solar radiation, not daylight, so solar radiation may be better. However, daylight hours of the growing season would capture more information in one variable.

[beckybanbury]

I've been trying to work out how to get total daylight hours for the growing season, and it's fairly complicated. I think it's possible, but it could take me a couple of days easily to write the code to derive the values. Before I really start to do that, I wanted to check that you think it's worth me doing!

[teixeirak]

Let's try first with solar radiation.

Also, maybe we could use a rough approximation at first to see if it would give interesting results?

[beckybanbury]

I have total solar radiation for the growing season and monthly average solar radiation during the growing season. I have run monthly growing season productivity against monthly growing season solar radiation; the graphs are saved here. What further analysis had you considered with this?

Regarding calculation of daylight hours on the growing season, I'm worried that we would be using a very coarse estimate of growing season (on a monthly resolution), and then using that to derive a variable on a very fine scale (an hourly resolution). I don't know that we can support analysis of hourly productivity, when our estimates of the growing season are such broad approximations? What do you think?

I'm also getting stuck with the circularity of this kind of analysis: growing season length is based entirely on temperature variables, and so it makes sense that this would control out temperature (and similarly if we included a measure of water availability it would control out that). But because climate determines growing season it is clearly still a key factor driving productivity.

A quick look at how total growing season solar radiation varies with latitude suggests that the highest levels of total solar radiation occur around lat = 20, and this declines to higher latitudes. There is no latitudinal pattern in average monthly growing season solar radiation (so this doesn't seem to support the idea that longer growing season day length in higher latitudes would mean higher light availability and potentially bring back in climate variables as explaining variations in productivity).

[teixeirak]

There is no latitudinal pattern in average monthly growing season solar radiation (so this doesn't seem to support the idea that longer growing season day length in higher latitudes would mean higher light availability and potentially bring back in climate variables as explaining variations in productivity).

Interesting, and makes sense. Despite longer days at high latitudes, the sun is at a lower angle, so average monthly growing season radiation is less. I agree-- this implies that the growing season hours correction is not appropriate. Let's drop that idea.

[teixeirak]

I'm also getting stuck with the circularity of this kind of analysis: growing season length is based entirely on temperature variables, and so it makes sense that this would control out temperature (and similarly if we included a measure of water availability it would control out that). But because climate determines growing season it is clearly still a key factor driving productivity.

I agree with you regarding the circularity as well. The motivation behind studies that do this has been to understand the role of temperature during months when trees are actively photosynthesizing. It's an interesting question whether productivity in warmer climates is higher because growing seasons are longer or because productivity inherently increases with air temperature during growing season months.

[teixeirak]

Regarding steps forward,

1. let's test growing season months as a "climate variable". Does it perform better than other temperature variables?
2. let's consider including the results of this analysis that you've just done.

The main point is to address the question as to whether accounting for growing season length gives us any improved productivity. (1) will tell us whether that's a useful variable. (2) could open up some interesting discussion.

[beckybanbury]

Growing season months as a climate variable is significant for all carbon fluxes, but doesn't come out as a better predictor than temperature for them (results are added here )

[teixeirak]

Thanks! And could you please check if it does better in the multivariate models? (Given the attention that has been given to this in the literature, I think its useful to test that.)

[beckybanbury]

It doesn't appear to (results table here ). It comes out as a better predictor for GPP, but looking at the delta AIC values, there is <2 difference between LGS and MAT, so this isn't super significant. In addition, marginal r2 is lower for LGS than for MAT. It also comes out as a better predictor for R_auto_root. Otherwise it doesn't improve on our previous results.

[teixeirak]

Okay, thanks. It will be good to present the test of this hypothesis. I'm working on our draft hypothesis table right now.

[beckybanbury]

If we are really wanting to test the ability of LGS to predict productivity, would it be worth using a definition of LGS that includes water availability? (several definitions I have read include this, and I think we have the data to calculate it). This would capture LGS more accurately in ecosystems where growth is limited by water availability for part of the year.

[teixeirak]

Yes, I think that would be good.

[beckybanbury]

I updated the definition of growing season to be months with mean minimum temperature > 0.5 and months with precipitation > 0.5* potential evapotranspiration. Unfortunately this definition leaves several sites with a growing season of 0 months, which is obviously incorrect!

I could try the definition used by Michaletz et al. which calculates a moisture index MI = (PPT−PET)/PET, and defines growing season as MI > 0.048.

Alternatively Kerkhoff et al. (2005) use a definition of growing season as min temp > 0 and moisture index > -0.95.

Do you have any thoughts on this?

[teixeirak]

What are some example sites? I take it the problem is with the moisture index?

[beckybanbury]

Yes, the problem is the addition of the moisture criteria - using only the temperature definition gave all but two sites in my analysis with growing season length of at least two months. Its a range of sites that are affected: e.g. Big Butte, Biliuti, Coconino, Dalongsi, Deer Canyon, Frazer, Manasi, New Brunswick. A lot of them are from Arizona/Colorado/California, so it makes sense that the moisture criteria is coming into play there; I wonder if the moisture criteria are too rigorous, or alternatively whether this is a problem with not having site-specific monthly data. There's also a good number from China.

[teixeirak]

I think the criteria are too rigorous. In that region, there can be very little precipitation during the summer, and snow melt/ groundwater can be a big factors. Note that in some places winters (when most precip occurs) could be eliminated by cold temperatures, and summers by low precip.

One reason I was hesitant about a moisture criteria is that its more complex...

[teixeirak]

I think that doing the moisture limitation really well would require accounting for lags driven by snowmelt and soil water storage, which of course is more complex than we can get.

How influential does inclusion of moisture on growing season length appear to be?

[beckybanbury]

I tried the Kerkhoff definition, where moisture index > -0.95 (Michaletz states that their definition is calibrated to their dataset, such that each site has at least one growing season month), and it doesn't change really very little from only using minimum temperature as a cut-off criteria. Kerkhoff et al. did find a similar result, where using only the minimum temperature criterion for growing season didn't quantitatively affect their findings.

[teixeirak]

-.95 would imply that there's a >95% deficit of PPT relative to PET, correct? That seems like a pretty extreme deficit! It's not surprising that that has little impact. We could exclude those just to say we did, but I think the moisture side is too complex to capture well with a simple metric like this, so maybe better just to leave out (and explain why).

Also, is this correct? (PPT−PET)/PET, and defines growing season as MI > 0.048 (Michaletz et al., above)? Or is it (PET-PPT)/PET?

[beckybanbury]

Also, is this correct? (PPT−PET)/PET, and defines growing season as MI > 0.048 (Michaletz et al., above)? Or is it (PET-PPT)/PET?

Michaletz actually used PPT/PET > 0.048, while Kerkhoff used (PPT−PET)/PET > -0.95.

-.95 would imply that there's a >95% deficit of PPT relative to PET, correct?

Yes, I think that is correct - in which case you're right, and that would have very little impact! Reading the methods of Michaletz, their cutoffs are very similar - the Kerkhoff cutoff is equivalent to 0.05 if using Michaletz's equation. So it seems that this is the maximum value which allows for at least one growing season month at all sites.

[teixeirak]

I think we could go either way on this.

On the side for including it, (1) we're engaging with two (+?) studies that have done this, and implementing the same method would make for more direct comparison of results (although that could be achieved by a simple statement that including dry season changes little), (2) I like the concept of including both temp and precip seasonality simultaneously (this fits well with hypotheses) (3) reviewers may ask that we deal with dry season

On the side against including it, I don't have any great faith that this metric they're using is accurate. To know this, it could help to look at some eddy flux studies from sites where a dry season is shortening growing season--or even just read some site descriptions. Do dry seasons noted in publications match those identified using this metric?

[beckybanbury]

On the side against including it, I don't have any great faith that this metric they're using is accurate. To know this, it could help to look at some eddy flux studies from sites where a dry season is shortening growing season--or even just read some site descriptions.

I've looked through a selection of the Fluxnet site data to identify sites where this might occur, but so far I haven't found any that don't align with what we would predict from temperature. Are there any sites you know of that you would expect to show a pronounced dry season that I should look at more closely?

[teixeirak]

Try Howard Springs (Australia)

[beckybanbury]

All the Australian sites I looked at generally show negative NEE in all months of the year. This varies a little from year to year but isn't consistent enough to conclude that there are months outside the growing season based on the flux data.

[teixeirak]

What are some sites where growing season is shortened due to aridity? They don't necessarily need to be flux sites. I'd just look at the site/ climate descriptions.