teixeirak
commented
4 years ago I fully agree that it's coarse, and to me that's part of the point. There have been a number of papers attempting to use this (including one I wrote way back in grad school!), and the goal here was to test whether that approach can provide a better predictor than annual climate values.
I don't know how possible it would be to provide more accurate resolution. Even with daily temperature data, differences in other aspects of climate and biology are going to make it impossible to derive great growing season length estimates. Trying to bring in moisture-defined growing seasons becomes even more complex because of lags from snowmelt/ ground storage. (We spent a bit of time trying to optimize growing season length calculations.) Perhaps a better approach would be satellite-based estimates based on greenness or such, but that's really beyond the scope here.
I would not change any analyses, but we do need to make sure the discussion around this issue is appropriate.
One of the main objectives of the paper is to compare MAT with growing season length as a predictor. But growing season length is calculated only as number of calendar months, which is quite coarse, and that coarseness alone will reduce r2. This raises the question of whether this is a fair comparison. The coarseness of the growing season calculations also reduces the expected explanatory value of all the climate metrics calculated for growing seasons.
After all, consider how much lower the r2 for MAT would be if we just grouped sites into 8 or 12 categories of MAT, and used the midpoint of each category as the predictor variable. Similarly, if the true growing season is April 14 to October 10, and the growing season used for calculations is April 1 to September 30, and so forth for different sites, then that’s obviously going to make the analyzed growing season means differ from the true ones.
At the least, this issue needs to be acknowledged in the discussion as something that could tip the scales against growing season length and growing season means as explanatory variables.
The ideal would be to recalculate growing season length and growing season climate states using daily data instead of monthly data. This could be done either using data on means by day of year (then growing season length could be something like the number of days of year mean minimum temperatures are above 0.5 C, or if using reanalysis data for every year, it could actually be the mean over years of the number of days minimum temperatures are above 0.5 C). I realize that Worldclim data are only monthly resolution, so this approach would be tricky. I see a couple possibilities in practice: Use the worldclim data, and apply additional analyses to it in order to estimate the number of days in each month temperatures are above 0.5 C. Or use a different dataset with higher temporal resolution but lower spatial resolution (e.g., one of the global reanalysis datasets). Or most complicated – use a combination of worldclim high spatial resolution and other high temporal resolution data to interpolate / estimate high spatial and temporal resolution climate for focal sites.
To be clear, I don’t see new analyses as critical for this paper. But I’m sure that the current monthly resolution biases against the growing season variables being good predictors, so this needs to be kept in mind in the interpretation.