Closed leeyap closed 2 years ago
> clean_trend
# A tibble: 10 x 5
ssp computation trend error p.value
<chr> <chr> <dbl> <dbl> <dbl>
1 ssp126 ClassicAF -0.0281 0.000137 0
2 ssp245 ClassicAF -0.0282 0.000140 0
3 ssp370 ClassicAF -0.0277 0.000146 0
4 ssp460 ClassicAF -0.0320 0.000134 0
5 ssp585 ClassicAF -0.0266 0.000147 0
6 ssp126 EmissFraction -0.000867 0.0000273 3.50e-218
7 ssp245 EmissFraction -0.000946 0.0000272 2.06e-259
8 ssp370 EmissFraction -0.000542 0.0000270 3.33e- 89
9 ssp460 EmissFraction -0.000447 0.0000274 1.36e- 59
10 ssp585 EmissFraction -0.000929 0.0000263 3.18e-268
It'd be interesting to put our results in the context of Figure 4 of Van Marle et al. 2022:
Is "slope of LULCC emissions" computed like we did the AF trend?
@leeyap I think it would be, yes.
Will push up in a bit, but results:
ssp trend error avg
<chr> <dbl> <dbl> <dbl>
1 ssp126 -0.144 0.00124 1.66
2 ssp245 -0.142 0.00127 1.67
3 ssp370 -0.133 0.00124 1.68
4 ssp460 -0.172 0.00133 1.63
5 ssp585 -0.133 0.00123 1.70
This doesn't feel right per the chart above. I took the mean luc_emissions
by scenario over all runs to find the "avg" values, and used the same method as above for the trend. But using those values plots us off the chart, and not where our AF trend tells us we should be.
I get a slope of -0.011 PgC/yr:
What were your lm
arguments?
Just luc_data %>% filter(year %in% 1960:2020) %>% lm(luc_emissions ~ year)
basically.
Oh, silly mistake on my part, I was computing the values per decade.
ssp luc_trend error avg
<chr> <dbl> <dbl> <dbl>
1 ssp126 -0.0144 0.000124 1.66
2 ssp245 -0.0142 0.000127 1.65
3 ssp370 -0.0133 0.000124 1.67
4 ssp460 -0.0172 0.000133 1.66
5 ssp585 -0.0133 0.000123 1.69
So we still have a weird thing happening - the LUC values put us out of the range of the graph in the top left, but our AF trend color bar would be bottom right, or towards the center if we use the carbon tracking values.
Tasks:
@bpbond Not sure if I did this incorrectly, but here is the decadal AF trend across all runs for the non-carbon-tracking method:
These are just the values of each SSP's trend, although I don't know why each one doesn't have exactly 500 runs.
run_number computation af_trend error p.value
<int> <chr> <dbl> <dbl> <dbl>
1 1 Airborne fraction -0.0281 0.00311 9.34e-13
2 2 Airborne fraction -0.0281 0.00311 9.34e-13
3 3 Airborne fraction -0.0281 0.00311 9.34e-13
4 4 Airborne fraction -0.0281 0.00311 9.34e-13
5 5 Airborne fraction -0.0281 0.00311 9.34e-13
6 6 Airborne fraction -0.0281 0.00311 9.34e-13
For runs 1-500, 501-1000, etc, the trend number was the same across runs.
Hmm. @leeyap this is coming out of lines 591-?
af_trend <- plot_data %>%
filter(year %in% 1960:2020) %>%
group_by(run_number, def) %>%
do(mod = lm(med ~ year, data = .)) %>%
summarize(run_number = run_number,
def = def,
broom::tidy(mod))
If yes what is med
above?
...if it's the median value of a group of runs (e.g. by scenario) then I think we need to back up and use the individual-run AF values.
Ah, okay, this looks better: So this is the trend across all runs, not accounting for differences across SSPs
WOOOO! That is great! Dang, we may have a new manuscript figure.
I think I'm set here - I updated figure captions/labels/axes to make things more clear for our presentation next week. Does the AF trend info and Rmd look good to you?
Lazy scientist: could you re-send me the cleaned up html file?
Never mind!
Looks great. See #47 but then merge when ready!
Following the discussion in #37, here is code that calculates the airborne fraction decadal trend from 1960-2020 by scenario and computation method.
Interesting to note that the mathematical definition returns trend numbers very similar to those in van Marle et al., although with a much lower error, but the carbon tracking method's trends are much smaller values.