search

forc-db / Global_Productivity

Creative Commons Attribution 4.0 International
2 stars 0 forks source link

implications for models #125

Closed teixeirak closed 3 years ago

teixeirak commented 3 years ago

@beckybanbury and @hmullerlandau (because you may have a good suggestion here),

I just realized that we haven't done anything about one of R2's major comments:

"3. Implication for models. I think it is nice for the authors to report that carbon flux of mature forest is primarily dependent on MAT or latitude. But how much could this advanced knowledge help us improve global simulation of forest carbon cycling. I’d be very curious to see if the state-of-the-art models demonstrate a similar or different functional response of carbon flux to MAT/latitude, and what is the possible reason for that. I think by tapping into this question (even just in discussion) will make the study be of interest to a wider audience."

Doing a quantitative comparison against model output is probably not possible at this point (without asking for an extension, which I strongly prefer not to do). It should actually be easy to pull out records, similar to Fig. 2 in Helene's Tansley review. A few years back, @ValentineHerr downloaded some of these model outputs, so we should have them on our drive, although I don't know if they're recent.

At this point, my plan will be to put some statements in the discussion so that we can at least give a minimal response to this comment, but I could sure use help, as I'm not familiar enough with models to say anything meaningful off the top of my head.

beckybanbury commented 3 years ago

@teixeirak I'll be spending the rest of the day working on the manuscript - in terms of this issue what would be most helpful for me to do? (e.g. papers to read/check etc). Do you know where the outputs Valentine used are saved - is that on a github repo or an SI drive (I might not have access if so)?

teixeirak commented 3 years ago

I believe the data are on the SCBI server. If you want to try tackling this, I suspect that @ValentineHerr will be working soon and able to point you to them. (Valentine, this was part of the calculator project. I can try to provide more details.) Alternatively, it may be easier to use the same data set that @hmullerlandau used. That has the advantage of being recent, and with everything figured out.

@hmullerlandau , who extracted the data for that analysis?

teixeirak commented 3 years ago

This is the only real remaining issue. I've been reviewing/ cleaning up all the files, and the only piece left is the discussion. After that, I'll turn my attention to reviewing more of the relevant literature, especially wrt models.

ValentineHerr commented 3 years ago

A few years back, @ValentineHerr downloaded some of these model outputs, so we should have them on our drive, although I don't know if they're recent.

I think you need to refresh my memory @teixeirak

teixeirak commented 3 years ago

We downloaded MsTMIP simulations: https://nacp.ornl.gov/MsTMIP_simulations.shtml. http://nacp.ornl.gov/mstmipdata/

Apparently we downloaded the following:

CLM4 RG1 (no climate change): AbvGrndWood, CarbPools, NEE, TotLivBiom, TotSoilCarb. Monthly data. 1990-2010 for fluxes, 2000 for stocks

CLM4 SG1 (climate change, but no CO2):AbvGrndWood, CarbPools, NEE, TotLivBiom, TotSoilCarb. Monthly data 1990-2010 for fluxes, 2000 for stocks

Biome-BGC RG1 (no climate change): AbvGrndWood, CarbPools, NEE, TotLivBiom, TotSoilCarb. Monthly data 1990-2010 for fluxes, 2000 for stocks

Biome-BGC SG1 (climate change, but no CO2): AbvGrndWood, CarbPools, NEE, TotLivBiom, TotSoilCarb. Monthly data 1990-2010 for fluxes, 2000 for stocks

Stored in Box (with calculator maps).

Two problems:

teixeirak commented 3 years ago

@hmullerlandau used the Coupled Model Intercomparison Project 5 (Taylor et al., 2012).

teixeirak commented 3 years ago

Here's the description of methods from the SI:

Model-data comparison(Figure 2)ESMs contributing to the last IPPC Coupled Model IntercomparisonProject 5 (CMIP5; https://esgf-node.llnl.gov/projects/cmip5/) included complex dynamic vegetation modules that predicted a range of vegetation patterns. Here we focused on three key measures of forest carbon stocks and fluxes: AGB, AWP, and AWRT. (AGB and AWP are denotedin the model output aswAGB and wNPP, and we calculated AWRT as AGB/AWP.)To our surprise, only a fewmodels actually distinguish a woody compartment for tree biomass and production. We present results of one representative of every modelfamily that reports AGB and AWP: Model A in Figure 2 is CCSM4, B is IPSL, and C is NorESM1-ME.

To obtained predicted values, we downloaded the output for historical simulations runbetween 1850 and 2005. Historical simulations use as external forcing the observed levels of greenhouse gasemissions and incoming solar radiation (Tayloret al., 2012). The ESM simulates the entire climate system and, although the simulations may beout of phase with observed variability, they areexpected to produce realistic long-term statistics. We compared the output of the model with digitized matchups in the database of field measurements assembled by Galbraith et al (2013)across the tropics (n = 177). We retrieved model time series of wAGB and wNPP at the four nearest neighbors of each location, excluding model cells covered by water. We averaged monthly model output for the period 1982-2005 (as in (Carvalhaiset al., 2014)), which overlaps the period with available observations. We also averaged model time series across realizations when more than one simulation was available. Dry mass wAGB and wNPP measurements were converted to carbon units (consistent with Galbraith et al. 2013) assuming a constant carbon content of fC = 0.47 across the tropics (Martin & Thomas, 2011). Then, we used inverse distance weighted interpolation to retrieve wAGB and wNPP values at each location with observations and, lastly, calculated residence time (RT) as the ratio between average wAGB and wNPP.

teixeirak commented 3 years ago

@ValentineHerr , I wasn't even able to log into Box. I think that that those files are already so dated that I wouldn't bother with them-- unless they are on our server and well-documented there, with fluxes other than just NEE.

Our team's most recent experience with this would definitely be Helene's analysis. @hmullerlandau , if you're able to comment on this this AM, and if it would be very easy to extract data for the points in Becky's analysis, we might try that. However, I'm not sure which variables are available (would be silly to look at just woody productivity).

teixeirak commented 3 years ago

Another option might be those used in Collalti et al. 2020:

Outputs from TRENDY v.7. We used the simulations from eight Dynamic Global Vegetation Models (DGVMs) performed in the framework of the TRENDY v.7 project2,67 (http://dgvm.ceh.ac.uk/node/9; data downloaded 27 November 2019). Models that did not provide NPP and GPP at plant functional type level were excluded because of the need to analyse CUE in forests without significant con- tributions from shrubs, grassland or crops. The selection comprises the following models: ISAM, JULES, LPJ-GUESS, CABLE-POP, ORCHIDEE, ORCHIDE-CNP, JSBACH and SDGVM (for references on models see refs. 2,67 and Supplementary Table 6). All the models represent the surface fluxes of CO2, water and the dynamics of carbon pools in response to changes in climate, atmospheric CO2 concentration, and land-use change across a global grid. However, processes underlying the exchanges of water and carbon are based on different formulations in different models. In the TRENDY protocol all DGVMs were forced with common historical climate fields and atmospheric CO2 concentrations over the period from 1700 to 2017. Climate fields were taken from the CRU-JRA55 dataset2, whereas the time series of atmospheric CO2 concentrations were derived from the combination of ice core records and atmospheric observations. Land-use change was taken into account in the simulations (S3). However, similar simulations without land-use change (S2) were also tested, showing no differences. CUE was estimated as NPP/ GPP (where NPP is commonly obtained in models by subtracting Ra from GPP) for the forest plant functional types simulated to be present in each grid cell. The model outputs refer to the mean from 1995 to 2015 for comparability with the records used when showing global land analysis (Fig. 5 and Supplementary Fig. 4). At site level, the same dates as the observations were chosen from the model outputs.

teixeirak commented 3 years ago

According to this list, there are only 3 variables in our analysis that are included in the TRENDY model output: GPP, R_auto, and NPP. A comparison of these 3 variables could make an interesting figure, but wouldn't add very much conceptually beyond what's presented in Collalti.

So, @beckybanbury , if you think you can quickly extract data from one of these sources and make a figure, it could be really cool, but if not I wouldn't sweat it.

I'm going to return to reviewing/ fixing up the manuscript now, and will then dig into the literature to incorporate as much as I can before I need to start the submission (~3pm eastern today).

beckybanbury commented 3 years ago

@teixeirak I'll have a look at the data and see if anything's possible!

teixeirak commented 3 years ago

@beckybanbury , I think we should add some citations to / mention of modeling papers in some of the places where we make general statements, e.g., "flux decreases with latitude", and also in Table 1. Could you please work on this while I start the submission? I'm going to list a few relevant pubs in the next comment.

teixeirak commented 3 years ago
hmullerlandau commented 3 years ago

Sorry, just seeing these messages now (right now github notifications go to my gmail, which I often don't look at during the workday).

There is a LOT in this paper already. I wouldn't really want to add any more analyses of model results already. I would recommend we just cite other published results on this. For example, the supplemental material of Martinez-Cano et al. 2020 provide information on output of the new LM3-PPA-TV model in terms of predictions for tropical sites varying in precip and temperature. I'll check for others like this ... and take a closer look at the long thread of comments today.

Helene

On Fri, Dec 18, 2020 at 3:50 PM Kristina Anderson-Teixeira < notifications@github.com> wrote:

NPP: Cramer et al. 1999: https://onlinelibrary.wiley.com/doi/epdf/10.1046/j.1365-2486.1999.00009.x

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/forc-db/Global_Productivity/issues/125#issuecomment-748312027, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADVG5N7P3VARAINGY2DPO6TSVO6BXANCNFSM4VBCEHJA .

-- Helene Muller-Landau hmullerlandau@gmail.com +507 6471-5214 (cell)

teixeirak commented 3 years ago

Thanks, Helene. This is due today, and I am about to start the submission while Becky works on references. So, if you can point us to a few useful references that could be helpful, but beyond that we don’t have time to add much

hmullerlandau commented 3 years ago

Looking at this more closely, Martinez-Cano et al. 2020 only shows results for AGB, so not useful.

Maybe we can just reply to the reviewer that we agree this would be an interesting comparison, but analyzing existing model results is beyond the scope of the paper, and we're not aware of relevant studies reporting such analyses for the current generation of models.

teixeirak commented 3 years ago

We've decided against trying to do an analysis (today!). I think we're fine with just adding discussion, so Becky is working on finding relevant modeling papers.

Here's what we have at the moment:

We have added the following comments on models to the discussion:

beckybanbury commented 3 years ago

Also added:

The significant variation in C fluxes as a function of stand age has implications for ecosystem models. Ecosystem modelling approaches may neglect age-related effects, or assume stand equilibrium [see e.g. @yu_high_2014; @collalti_forest_2020]. Our results highlight the importance of incorporating stand age into ecosystem models; without this, models are likely to be vulnerable to bias in global C flux projections.

bpbond commented 3 years ago

Reasonable given the short timeframe :) That reads well.

teixeirak commented 3 years ago

I agree, I think we've done about as much as we can on this issue-- at least in one day. :) Thank you all! Let's call it done and submit!

hmullerlandau commented 3 years ago

Sounds good.

On Fri, Dec 18, 2020 at 4:22 PM Kristina Anderson-Teixeira < notifications@github.com> wrote:

We've decided against trying to do an analysis (today!). I think we're fine with just adding discussion, so Becky is working on finding relevant modeling papers.

Here's what we have at the moment:

We have added the following comments on models to the discussion:

  • Discussion: "We find no significant trend in the allocation of $GPP$ between production and respiration across latitude or climate ($NPP$:$R_{auto}$; Fig. S3), counter to the idea that tropical forests have anomalously low $CUE$ [@de_lucia_forest_2007; @malhi_productivity_2012; @anderson-teixeira_carbon_2016], as predicted by most models [@collalti_forest_2020]. In contrast, @collalti_forest_2020 found that forest production efficiency increased with temperature--a finding that is consistent in direction with the insignificant trends observed here (Fig. S3)."
  • Last sentence of discussion: "In the meantime, understanding the fundamental climatic controls on annual C cycling in Earth's forests sets a firmer foundation for understanding global-scale forest C cycling and benchmarking the models [@fer_beyond_2021] used to predict forest responses and feedbacks to accelerating climate change."

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/forc-db/Global_Productivity/issues/125#issuecomment-748325829, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADVG5N324E5IRXRXUHJBXS3SVPB3HANCNFSM4VBCEHJA .

-- Helene Muller-Landau hmullerlandau@gmail.com +507 6471-5214 (cell)