Modeling vegetation and carbon dynamics of managed grasslands at the global scale with

Abstract. Grassland management affects the carbon fluxes of one-third of the global land area and is thus an important factor for the global carbon budget. Nonetheless, this aspect has been largely neglected or underrepresented in global carbon cycle models. We investigate four harvesting schemes for the managed grassland implementation of the dynamic global vegetation model (DGVM) Lund–Potsdam–Jena managed Land (LPJmL) that facilitate a better representation of actual management systems globally. We describe the model implementation and analyze simulation results with respect to harvest, net primary productivity and soil carbon content and by evaluating them against reported grass yields in Europe. We demonstrate the importance of accounting for differences in grassland management by assessing potential livestock grazing densities as well as the impacts of grazing, grazing intensities and mowing systems on soil carbon stocks. Grazing leads to soil carbon losses in polar or arid regions even at moderate livestock densities (

We cordially thank the reviewer for their thorough and constructive feedback as well as the 23 positive evaluation of our manuscript. 24 Below we provide a response to all detailed comments including proposals to achieve the 25 suggested improvements. where both C3 and C4 grasses are present. 31 We agree, even though we do not look into the effects of changing CO2 concentrations it 32 should be a part of this overview and we will add a brief description of its role. 33 line(s) 42: "high temperatures can lead to an increase of microbial decomposition". Only in 34 combination with sufficient moisture. In arid regions, decomposition comes more or less to a 35 stand-still during the dry season due to the water limitation that affects the microbial 36 community. Rains at the beginning of the wet season then lead to peak emissions when 37 microbial decomposition picks up again. 38 We will add a phrase to highlight that moisture is a necessary condition independent of 39 temperature. line(s) 44/45 "...may be beneficial for grassland productivity depending on its intensity". 41 Maybe add: "by removing moribund plant material and triggering growth (over-42 )compensation." 43 We will add the phrase the reviewer suggested. 44 line(s) 49: "for the species" -"for the functional types". I'd rather consistently keep the focus 45 on functional types. 46 We agree and will make the amendments throughout the manuscript. 47 line(s) 52: "indirectly through alterations of the resource limitations" -add: "…that can cause 48 shifts in the competitive balance between functional types". 49 We will add the phrase suggested by the reviewer.  The terminology for the naming of all sites was derived from the Koeppen Geiger climate 55 zones (in this case hot steppe). At the first mention we decided to add the form of grassland 56 management (pasture). We therefore will keep the naming as is but will add a phrase pointing 57 towards the term savanna rangeland in L105.  Thank you for this interesting suggestion. In this part of the manuscript, we only mention the 61 managements for which experimental data were available and that could therefore be used to 62 parameterize the sites. Not knowing experiments including fertilizer X defoliation 63 combinations, we would be grateful for information and very interested to include such data 64 and combinations in further studies. The additional scenarios are described in 2.5. With this 65 separation we distinguish between the scenarios that were predefined by the data and those we 66 selected for further analysis. When defining the scenarios for further analysis, we decided to 67 use extreme cases to test the effect of different limiting resources (e.g. infinite nutrient 68 availability) instead of choosing different fertilizer levels. Regarding the defoliation intensity, 69 we agree that analyzing a gradient of different intensities provides another interesting 70 experiment. However, we decided to put our main focus on the resources and believe that the 71 defoliation intensities of the experiment already cover a sufficient range.

77
The reviewer raises a very interesting point. LPJmL-CSR follows the concept of using a small 78 number of PFTs with fixed parameters. Therefore, for example SLA is fixed and each PFT only covers one point of the continuum. We also see the potential for interesting future work 80 following an individual based approach drawing trait values from a continuum similar to LPJ-81 FIT (Sakschewski et al., 2015) or aDGVM2 . However, the currently 82 implemented management routines of such models are less detailed compared to "classic" 83 DGVMs that include an agricultural component. We therefore see the necessity to continue to 84 improve grassland representation in both model types for the foreseeable future. 85 line(s) 120 "Overview of managed grasslands in LPJmL" -"Overview of managed grassland 86 representations in LPJmL" seems a more fitting title for this section. 87 We will adopt the recommendation of the reviewer.  We will change the unit to MgDM ha-1 yr-1 and add a phrase defining forage supply as 100 annual quantity removed through defoliation from mowing or grazing. 101 line(s) 166-168: Does this new scheme also account for root biomass distribution in different 102 soil layers, and therefore varying water availability between different soil layers? So that the 103 total water uptake is the biomass-weighted uptake sum across soil layers? Or is it simpler than 104 that? 105 We thank the reviewer for pointing out that this could be described more clearly. Root 106 distribution between different soil layers was already used to determine the water supply from 107 the different layers in the previous model version (Schaphoff et al., 2018). Our scheme retains 108 this approach and only distributes the sum over the supply from all soil layers based on the 109 root biomass. We will include this in the explanation of our approach.     Using "area-specific" as suggested by the reviewer is in our opinion less explicit since it does 131 not define which area. We propose instead to replace "available area" with bare ground area.   We will pick up on this in the discussion.  Indeed, only herbaceous PFTs are allowed to establish on managed grassland stands. We will 152 add this to the model description. We agree with the reviewer that the term sapling is 153 misleading in this context and will replace it with the term seedling throughout the 154 manuscript. In addition, since this may create some confusion regarding the sapling LAI 155 parameter, for which we have to keep the term, we will explain the origin of the parameter 156 name and its purpose. should be explained in more detail to prevent confusion with individual based approaches. We 165 will add a section in the methods explaining that each PFT can be seen as a representative for 166 a population with certain attributes that describe the population (e.g. number of average 167 individuals, individual biomass). In addition, we will discuss our approach in comparison to 168 an individual based approach to show advantages and disadvantages. 169 line(s) 203: age-dependent mortality: hard set (at a specific age), or based on an age-170 dependent likelihood? And: the age-dependency differs between the different strategy types?

171
Thank you for this comment. Actually neither is the case. Depending on the growth 172 efficiency, the number of average individuals is reduced (Appendix A3 L745-753). Actual 173 mortality is derived from the maximum mortality rate -which is the same for all strategy 174 types -and the growth efficiency. The growth efficiency is dependent on SLA, which differs 175 between the strategy types (Appendix A3 Eq. A10). We will extend the description in And what is the allowed maximum number of average individuals, and the maximum number 178 of grass-layer PFTs that can now coexist within one grid cell? 179 We thank the reviewer for this question. It made us realize that we did not include this in study. For future studies this number can be increased, however this will also increase the 186 computation requirements. We will mention this in the model description.   Table 2: Maybe add a column that specifies the predominant gradient associated with the 204 parameter. You mention it in the text of this section, but it would be helpful to also have it as 205 a brief overview in the table. I find the distinction between biotic and abiotic dimension a bit 206 arbitrary/confusing with respect of the definition. Referring directly to the respective gradient 207 (stress gradient for biotic, disturbance gradient for abiotic) would seem more intuitive for me. 208 We will abandon the terminology abiotic and biotic gradient. When writing the original draft, 209 we found that it provides a clear distinction between the parameters related to each gradient. 210 However, as the reviewer correctly noted, this creates an additional layer of terminology to 211 understand when reading the manuscript. For this purpose we will modify the manuscript to 212 follow the terminology stress and disturbance gradient as proposed by the reviewer and add a 213 column to the table.
214 Table 2: Hierarchy: How did you determine the hierarchy? Based on your expert assessment? 215 We will add a phrase stating that the qualitative hierarchy of the parameter values for each 216 PFT was derived from expert assessment by all co-authors. should have more transmission than low-SLA leaves. 219 We agree with the reviewer that transmissivity of single leaves and their SLA are correlated.  (biotic interaction), but it also relates to the stress gradient (abiotic) with respect to water 234 uptake capacity. This is an example illustrating why using "biotic" and "abiotic" as 235 dimensions is maybe not the best way to make the distinction. 236 We agree that there are cases were the distinction between biotic and abiotic is not so clear.

237
As already proposed earlier (reply in L209-214) we will abandon the terms and only retain the 238 terms stress and disturbance gradient. on the thickness of the leaf, one would expect kbeer to be correlated with SLA, which, unlike 242 kbeer, you define as abiotic dimension. It would be good if you sort this out more clearly. 243 We agree and refer to our proposal from the related comment in the reply in L220-229. We 244 will also describe more clearly, which parameters play a role for the stress or the disturbance 245 gradient or for both gradients. 246 line(s) 241/242: the leaf area index of a sapling represents the offspring size -What do you 247 define as "offspring size"? The height of the offspring, or its starting biomass, or its projected 248 foliar coverage? I'm not sure LAIsap is a good description of offspring size, as its meaning is 249 rather vague without a clearer definition. Whether a seedling/sapling of given leaf biomass 250 has a high or low LAI is a function of its SLA, so LAIsap for a given unit of leaf biomass 251 essentially is nothing else as another way to refer to SLA.

252
In LPJmL, the leaf area index of a sapling is only used to calculate the sapling biomass using 253 SLA. So instead of assuming a given leaf biomass, we assume a given SLA and calculate the 254 leaf biomass. Using the same SLA, a higher sapling LAI is equal to a higher sapling biomass. 255 We will change offspring size to offspring biomass and add an explanation of the relationship 256 to SLA. We will also revise the discussion to reflect both SLA and sapling LAI when 257 discussing offspring biomass.
258 Table 3: Flip order of columns "variable" and "site", as site is unique and variable is tied to 259 site and non-unique. 260 We agree and will apply the proposed change. 261 line(s) 287/288: "the current representation of some processes within the model" -which 262 processes specifically? 263 We here refer to section 4.1.2 where these processes are listed. We will change "some 264 processes within the model" to "the processes, listed in sect. 4.1.2," and remove the reference   We agree that additional information is needed. We first conducted a potential natural All processes are executed on a daily time scale. We also compute the outputs on a daily 277 timescale but aggregate to a monthly or annual resolution for some of the results. We will add 278 a sentence on this to the modelling protocol. Upon initialization, each PFT is established dependent on the respective establishment rate 284 and biomass (derived from sapling LAI, SLA and leaf to root ratio). Therefore, initially a PFT 285 with high values in both has a higher share in the community. However, if its strategy is not 286 suitable this will change over time. This means, that no data on initial community 287 composition or similar is needed. We will add this explanation to the model description. Figure 1: Please specify temporal reference frame for panels a, d, and g -is it the annual sum 290 (yield), the peak season leaf biomass (leaf biomass), the grazing period duration offtake 291 (grazing offtake)? 292 We will add a more thorough explanation in the caption.

293
General question on all scenarios that included animal grazing: Is preferential grazing, i.e.,  We agree that the decrease of the minimum canopy conductance is unlikely to be related to 308 grazing directly. More likely, the high and similar minimum canopy conductance of the 309 ungrazed scenario (C0) is an artefact of the parameterization. All parameters can be assigned 310 primary and secondary processes that they affect. The leaf to root ratio and the SLA are 311 different in the two scenarios and act as a compensation of defoliation from grazing (primary 312 process). However, to some extent these parameters also control access to and distribution of 313 resources (secondary processes). In the ungrazed scenario, these do not need to be adjusted to 314 compensate for the defoliation but can still play a role in the competition for water. Therefore, 315 more parameters can control resource access and distribution and it is likely that this will 316 affect the parameterisation of minimum canopy conductance. We will amend the description 317 of the parameters to account for the primary and secondary processes affected and add the 318 explanation to the discussion. That is a bit surprising? One would expect that irrigation reduces stress resulting from water 330 limitation, therefore opening the community more strongly for the C-PFT.

331
This was also surprising and counterintuitive to us. We already provide an explanation in the 332 discussion in L579-583, which we will now reference in the sentence in L442f.  We already touch upon this in L542-547 but agree that this can be discussed in more detail. 339 We will add a reference in L473f and extend the discussion in section 4.1.2. The reviewer raises several interesting questions that go beyond this study. We are currently 348 working on a globally applicable set of PFTs, which will form the basis of another study in 349 the near future. For that study, we retain the fixed PFT parameterization of classic DGVMs.  The faster growth compared to LPJmL 5 has two reasons: First, the new implementation of 357 biological nitrogen fixation led to less nitrogen stress and higher photosynthesis. Second, this 358 is also a result of the new parameterization, which was tailored to this site. We will add this 359 explanation after L494f. 360 line(s) 504: "but selected a livestock density of 1.0 cows ha-1" -use "livestock units" rather 361 than cows (how about steers, heifers, etc.); And: Is this to determine the amount of manure 362 input? The temperate grassland was not grazed but mowed, so livestock density does not 363 make much sense with respect to grazing off-take? 364 The livestock density refers only to the spin-up and the historical periods for which no data on 365 actual land use were available. Therefore, it is entirely unrelated to the transient simulations 366 that reproduce the mowing experiments. We will rephrase this paragraph to make this clear. 367 line(s) 506: Briefly describe the processes / mechanisms that lead to increased carbon input to 368 the soil in the CSR-version compared to the old version. 369 We identified three causes for the increased carbon input: First, the SLA longevity trade-off 370 we implemented led to an increase in turnover supplying more carbon to the litter layer. PFTs from your site-scale simulations. 378 We will adopt this suggestion (see also reply in L152-157).

385
We agree with the reviewer that this should be explained and share their opinion of the 386 underlying reasons. We will add a sentence to explain the dry wet dynamics of the site and 387 that these are independent of grazing, which therefore does not affect the water stress level 388 allowing the S-PFT to remain advantageous. 389 line(s) 540/541: You could test this by specifically allowing no other PFT than the S-type to 390 enforce a monoculture. 391 We discussed the possibility to investigate this further, but decided against because LPJmL   We touch upon this in section 3.4.2 L475f by saying that "the transition occurred within the 400 first one to two years", which is much faster than we would expect. We mention this when 401 discussing the change in soil organic carbon (L532-538) but we agree that this is very brief 402 and will add more detail and highlight the transition time more prominently. We will also 403 provide an explanation for the fast transition, which is related to the removal of competition 404 for water. In a water scarce environment, the S-PFT as a water saver was advantageous and Correct, but not in this case, because due to the irrigation you had drought eliminated.

415
The reviewer is correct. A plausible explanation is that the parameterization allows the R-PFT 416 to coexist with the C-PFT if the main resource limitation is removed. We will update the 417 explanation. (which should then be termed as "forage demand"? 423 We agree that our use of forage supply was ambiguous because we use it to define the amount 424 of biomass removed through mowing or grazing for the temperate grassland and the cold 425 steppe but also for the amount of leaf biomass available for grazing for the hot steppe. This 426 was an attempt to use common terms for all sites, which appears to be confusing instead of 427 helpful. We will therefore change the term forage supply to forage offtake for the temperate 428 grassland and the cold steppe and use the term leaf biomass for the hot steppe. We will add a 429 definition of forage offtake in the methods section and explain why we use a different term 430 for the hot steppe.  We agree with the reviewer that the model results provide strong evidence for overgrazing 444 and will add a phrase explicitly stating so. We will also add a sentence discussing the change 445 in community composition which shows an increase of the C-PFT (and also to some extent 446 the R-PFT) as shown in Fig SI 9 and 12. line(s) 562/563: You did not combine fertilization with irrigation, right? Do you expect that 448 fertilization in combination with irrigation would increase leaf biomass beyond the level 449 reached with irrigation alone? 450 Generally, irrigation alone already affects processes related to inorganic N inputs and losses.

451
Biological N fixation and mineralization increase with increasing soil moisture. However, 452 irrigation also leads to higher leaching. We therefore expect that the PFTs are still N limited 453 even though irrigation may already increase but could also decrease inorganic N availability.

454
Additional inorganic N from fertilization may remove the N limitation leading to an 455 additional leaf biomass increase but may also lead to higher maintenance respiration limiting 456 leaf biomass growth. Therefore, we cannot give an unambiguous answer and will add this 457 explanation in section 4.1.3. 458 line(s) 575: "Fertilization had no effect on SOC" -Not surprising, given that fertilization 459 without irrigation did not increase leaf biomass and therefore C-input to the soil. 460 We agree with the explanation of the reviewer and will add this to the sentence. biological nitrogen fixation which may take longer than the simulated time frame (see also 468 reply in L420-423). We will add this to the discussion. given strategy type. Since you seem to have no other mortality causes aside from age-481 dependent mortality in the model (at least not for the grass layer), you will not see this effect, 482 but it does exist, nonetheless. 483 We agree with the reviewer and will extend this sentence to reflect the limitation of our model 484 to age mortality and to discuss potential effects of other causes of mortality. Facing the challenge of adding new PFTs to a classic DGVM, our aim was to reduce 488 complexity as much as possible at first. This included restricting ourselves to add as little 489 PFTs as possible. Grouping N-fixers with non-fixers halved the number of PFTs. We believe 490 this is reasonable because the model will only fix additional N if the demand is not fulfilled.

491
In an approach with two separate PFTs, this would mean a change in community composition 492 and an increase of the N-fixer PFT at the expense of the non-fixer. In our approach, this 493 simply means an increase in biological nitrogen fixation. One could say, that implicitly the 494 PFT is a fixer if needed and not if not needed and could determine this status using the 495 biological nitrogen fixation output. We will add the necessary detail to the description of 496 biological N fixation in Appendix A4. 497 line(s) 622/623: So the assumption is that grazing is non-preferential, correct? I.e., grazers do 498 not favor one PFT over another, for example based on criteria that characterize palatability / 499 nutrition value. This is a simplification in the model that should be discussed briefly, as 500 herbivores usually do not function the same way as mowing (or fire) that removes biomass 501 indiscriminatingly.

502
Yes, grazing is not preferential. As proposed in our reply in L301ff we will include this in the 503 model description and include a brief discussion in the limitations. 504 line(s) 624: "tolerance or avoidance" -Avoidance would for example (aside from temporal 505 avoidance) be realized by being unpalatable. As your grazing is non-preferential, being a 506 grazing avoider type based on palatability would not make a difference in your model as the 507 animals would not discriminate against the avoider. This is a limitation you should mention. 508 We thank the reviewer for raising this point and will include grazing avoidance through 509 palatability in the limitations together with preferential grazing (reply in L503f).  We thank the reviewer for their suggestion and will add a sentence on this at the end of the 515 section (L631). happens during the growing season, and your only way to implement avoidance is through life 520 cycle adaptation, i.e., temporal avoidance, this will push avoiders to the non-growing season 521 as time when no grazing happens. But I don't see how avoiders could succeed by shifting their 522 existence focus to exactly the season when growth is not possible? 523 We will add a phrase acknowledging that the model will not be able to simulate the type of 524 avoidance that is likely successful. strategies that can be realized, for example by defining typical value ranges for the given 528 parameters of a strategy type. Within these continuous ranges, a strategy type can assume many trait value combinations that define its location within the trait space occupied by the 530 strategy type, and therefore allows more plasticity within a strategy type, e.g., a plant could be 531 a moderate, intermediate, or extreme S-strategy type. 532 We agree with the reviewer, that moving away from the fixed PFT approach is a suitable way 533 to circumvent many of these issues. As discussed in previous comments one necessity is to 534 follow an individual based approach as in aDGVM2 or LPJ-FIT. We see this as a promising 535 and intriguing topic for future model development of LPJmL-CSR and will emphasize this 536 more in the discussion. show in this study is to prove that the CSR-concept can work within a DGVM and is 541 ecologically sound in many points. But to make it general, you will have to move away from 542 the discrete parameterization of your PFT approach, for example by allowing an evolutionary 543 approach that self-selects successfull strategies via environmental filtering from a pool of 544 potential trait value combinations, where each trait is represented by a continuous range of 545 allowed values.

546
The generalization for a global application indeed poses a challenge. However, for the tree 547 PFTs, researchers managed to find a set for classic DGVMS that represents the broad range of 548 environmental conditions possible. We believe that for herbaceous PFTs it will also be 549 possible to find a suitable set that will improve the representation of grasslands in current 550 DGVMs We hope to present this in a separate study in the near future. In the long term, 551 additional model development including the step towards dynamic PFTs will further improve 552 the representation of different growth strategies. 553 line(s) 664/665: I do not really agree with this approach. The light extinction coefficient (as I 554 know it) is a constant that describes how much light a respective layer of leaves will absorb 555 and how much it will allow to transmit to the next lower leaf level. As such, it is a proxy 556 associated with leaf characteristics such as leaf thickness or SLA more than overall plant 557 stature. If anything, I'd deem LAI closer to stature than the light extinction coefficient, if you 558 do not have height available as state variable.

559
The reviewer is correct that the light extinction coefficient usually refers to the transmissivity 560 of a leaf layer. In theory, this is represented as one leaf with a given height and SLA per layer. Similar to the missing inclusion of preferential grazing (comment in L294-300), this is related 574 to the representation of grazing. We will add trampling to the discussion of the limitations of 575 the current grazing approach as well.

576
Minor editorial comments 577 We appreciate the thorough reading and will adopt all minor editorial comments below 578 without responding to each of those separately. 579 line(s) 10: "… a temperate grassland, a hot and a cold steppe…" => "… a temperate grassland