Carbon storage in phosphorus limited grasslands may decline in response to elevated nitrogen deposition: a long term field manipulation and modelling study

In ecosystems where nitrogen (N) limits plant productivity, N deposition can stimulate plant growth, and consequently, promote carbon (C) sequestration by increasing input of detrital C and other forms of plant C to the soil. However, other forms of nutrient limitation such as phosphorus (P) limitation and N-P co-limitation are widespread and may increase in prevalence with N deposition. Our understanding of how terrestrial ecosystem C, N and P cycling may be affected by N deposition when N is not the sole limiting resource is fairly limited. In this work, we investigate the consequences of enhanced N addition on C, N and P cycling in grasslands that exhibit contrasting forms of nutrient limitation.

widespread, than N limitation [Fay et al., 2015;Du et al., 2020;Hou et al., 2020]. In a meta-analysis of 73 grassland nutrient addition experiments spanning five continents, Fay et al. [2015] found that 74 aboveground annual net primary productivity (ANPP) was limited solely by P in 8 sites and co-limited 75 by N and P in 25, compared to only 10 sites showing N limitation alone [Fay et al., 2015]. Similarly, P 76 additions in 652 field experiments increased aboveground plant productivity by an average of 34.9% 77 [Hou et al., 2020], while it is estimated that co-limitation of N and P could constrain up to 39% of the 78 terrestrial surface's productivity [Du et al., 2020]. 79 Furthermore, P limitation may be exacerbated by N deposition [Johnson et al., 1999;Phoenix et al., 80 2004], or become increasingly prevalent as previously N-limited ecosystems transition to N-sufficient 81 states [Goll et al., 2012]. For example, in parts of the Peak District National Park, UK, N deposition has 82 exceeded 3 g m -2 yr -1 , with further experimental additions of 3.5 g m -2 yr -1 leading to decreases rather 83 than increases in productivity of calcareous grasslands [Carroll et al., 2003], in contrast to previous 84 studies of N deposition enhancement of N-limited productivity [Tipping et al., 2019]. This makes P 85 limitation and N-P co-limitation critical to understand in the context of global carbon and nutrient 86 cycles. By definition, N deposition should impact N-P co-limited and P-limited ecosystems differently 87 to N-limited ones, yet there is little understanding of how N deposition impacts P and N-P co-limited 88 ecosystems. 89 While N deposition may worsen P limitation in some instances, plant strategies for P acquisition, such 90 as changes in root architecture and increased root exudation [Vance et al., 2003] require substantial 91 investments of N, suggesting that in some areas with P depleted soils, N may facilitate enhanced P 92 uptake [Long et al., 2016;Chen et al., 2020]. It has been shown that N deposition can stimulate 93 additional production of extracellular phosphorus-cleaving enzymes by plants [Johnson et al., 1999;94 Phoenix et al., 2004], thereby increasing plant availability of organic forms of phosphorus in order to 95 help meet plant P demand. This response could be driven by exacerbated P-limitation resulting from 96 N deposition or extra N availability making elevated enzyme production possible. 97 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.
These interdependencies of the C, N and P cycles make understanding an ecosystem's response to 98 perturbations in any one nutrient cycle challenging, particularly when ecosystems are not solely 99 limited in N. This highlights the need for integrated understanding of plant-soil nutrient cycling across 100 the C, N and P cycles, and in ecosystems that are not solely N-limited. 101 Process-based models have a role to play in addressing this, as they allow us to test our mechanistic 102 understanding and decouple the effects of multiple drivers. There has been increasing interest in 103 linking C with N and P cycles in terrestrial ecosystem models [Wang et  biogeochemical cycling are realised [Yuan et al., 2018]. Yet, few modelling studies have explicitly 106 examined the effects of P or N-P co-limitation, likely mirroring the relatively fewer empirical studies 107 of these systems. 108 Combining process-based models with empirical data from long-term nutrient-manipulation 109 experiments offers a valuable opportunity for understanding ecosystem responses to environmental 110 changes that may only manifest after extended periods of time, such as with changes in soil organic 111 C, N and P pools, which typically occur on decadal timescales [Davies et al., 2016a, Janes-Bassett et 112 al., 2020. Ecological data from these experiments can be used to drive and calibrate process-based 113 models, which in turn can disentangle multiple interacting processes involved in plant-soil nutrient 114 cycling, that otherwise makes interpretation of empirical experiments complex. This allows us to test 115 our assumptions of the key drivers, processes and pathways for carbon and nutrient cycling in 116 grasslands exposed to multiple environmental perturbations. 117 Here, we combine new data from a long-term nutrient manipulation experiment on two contrasting 118 upland grasslands (acidic and calcareous), both N-P co-limited to differing degrees (one more P 119 limited, one more N limited within the co-NP range), with the mechanistic C-N-P plant-soil 120 biogeochemical model; N14CP [Davies et al., 2016b]. We use this model and data to simulate the long-121 term nutrient manipulation experiment in both grasslands and then use the calibrated model to 122 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License. determine the long-term consequences of differing nutrient limitation on plant and soil C, N and P. To 123 do so, we allow modelled P-access conditions to vary and used the combinations of P-access variables 124 that most closely represented empirical data to simulate the grasslands. 125 Specifically, we aim to first explore how variation in P acquisition parameters, that control access to 126 organic and inorganic sources of P in the model, may help account for differing responses to N and P 127 additions in the empirical data on aboveground biomass carbon and soil C, N and P pools. Secondly, 128 we explore the effects of long-term anthropogenic N deposition at the site and the effects of 129 experimental nutrient additions (N and P) on plant and soil variables of the simulated acidic and 130 calcareous grasslands. This will help improve our understanding of how similarly P-limited or N-P co-131 limited grasslands may respond to similar conditions. We hypothesise that 1) flexible P-access within 132 the model may help in alleviating P limitation and that 2) grasslands of contrasting nutrient limitation 133 respond to N deposition and nutrient treatment in dissimilar ways, with N deposition exacerbating 134 nutrient limitation in more P-limited grasslands, in turn leading to declining productivity and carbon 135  [Morecroft et al., 1994]. There are two distinct grassland 149 communities occurring in close proximity; acidic (National vegetation classification U4e) and 150 calcareous (NVC CG2d) semi-natural grasslands (Table S2). Both grasslands share a carboniferous 151 limestone hill but the calcareous grassland sits atop a thin humic ranker [Horswill et al., 2008] and 152 occurs predominantly on the hill brow. In contrast, the acidic grassland occurs in the trough of the 153 hill, allowing the accumulation of wind-blown loess and the formation of a deeper soil profile. As 154 such, the acidic grassland shares the same limestone bedrock but sits atop a palaeo-argillic brown 155 earth soil [Horswill et al., 2008]. 156 The biomass in both grasslands show signs of both N and P-limitation, though they differ in the 157 relative strength of limitation by N and P. The acidic grassland is co-limited in N and P, as positive 158 biomass growth responses are observed with additions of both nutrients [Phoenix et al., 2003]. The 159 calcareous grassland, however, is more strongly P-limited, showing increased productivity only with 160 the addition of P [Carroll et al., 2003], though N and P co-limitation has been observed [Phoenix et 161 al., 2003]. 162 Nutrients (N and P) have been experimentally added to investigate the effects of elevated N 163 deposition and the influence of P limitation [Morecroft et al., 1994]. Nitrogen treatments simulate 164 additional N deposition to the background level and also act to exacerbate P limitation [Johnson et 165 al., 1999;Phoenix et al., 2004], whereas the P treatment acts to alleviate it. Nutrients are added as 166 solutions of distilled water and applied as fine spray by backpack sprayer, and have been applied 167 monthly since 1995, and since 2017 bi-monthly. Nutrient additions are in the form of NH4NO3 for 168 nitrogen and NaH2PO4.H2O for phosphorus. Nitrogen is applied at rates of 0 (distilled water control -169 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License. 0N), 3.5 (low nitrogen -LN) and 14 g N m -2 yr -1 (high nitrogen -HN). The P treatment is applied at a 170 rate of 3.5 g P m -2 yr -1 (phosphorus -P). 171 Data collected from the Wardlow grasslands for the purpose of this work are; aboveground biomass 172 C, SOC, and total N, which is assumed to be equivalent to modelled SON. This new data is combined 173 with total P data that was collected by Horswill et al. at the site [Horswill et al., 2008]. Summaries of 174 these data are available within the supplementary material (Table S4)  Phosphorus enters the plant-soil system by weathering of parent material, the initial value of which 213 (PWeath0 within the model) can be set to a default value, or made site-specific by calibrating this initial 214 condition to soil observational data (as in methods section 2.3.3). In principle, P can be added in 215 small quantities by atmospheric deposition [Ridame and Guieu, 2002] or by local redistribution 216 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License. [Tipping et al., 2014]. For the purpose of this study, P deposition is set to zero as its net contribution 217 to the total P pool in comparison to weathering is assumed to be minimal. Annual release of 218 weathered P is determined by a first-order rate constant, which is temperature dependent. Where 219 the mean temperature falls below 0 °C, it is assumed that no weathering occurs. 220 In the presence of sufficient N and where plant demand for P cannot be met by more accessible P 221 sources, plants can access P from the soil organic phosphorus (SOP) pool via a cleaving parameter 222 termed PCleaveMax, which is the maximum quantity of cleavable P within a growing season (g m -2 ). It is 223 PCleaveMax and PWeath0 that we allow to vary to account for discrepancies in empirical data. 224 Contributions of N and P toward the plant available pools are summarised in Figure 1. in the N14CP model. In the model, N can enter the available pool via atmospheric deposition, biological nitrogen fixation, coarse litter decomposition and decomposition of the soil organic matter pools. For P, the two main contemporary sources are the inorganic sorbed pool and from the turnover of soil organic matter. The former is derived initially from the weatherable supply of P, defined by its initial condition (PWeath0). Solid lines indicate input to another pool and a dashed line indicates either a feedback or interaction with another pool. These interactions include the downregulation of N fixation by N deposition, the dependency of N fixation on P availability, and the cleaving of organic P by plants when N is sufficient and other P sources are inaccessible.
Phosphorus access within N14CP is determined by a hierarchal relationship, whereby plants and 239 microbes access the most readily available P sources first and only move onto the next once it has 240 been exhausted. Out of the P sources available to plants (Fig 1), organic P is the least bioavailable 241 within the model hierarchy, hence a depletion in the SOP pool is indicative of severe P stress and low 242 P availability. To provide climate forcing data, daily minimum, mean and maximum temperature and mean 291 precipitation records beginning in 1960 were extracted from the UKPC09 Met office CEDA database 292 (Table S3). The data nearest to Wardlow was calculated by triangulating latitude and longitude data 293 and using Pythagoras' theorem to determine the shortest distance. These data were converted into 294 mean quarterly temperature and precipitation. Prior to this, temperature was assumed to follow 295 trends described in Davies et al. [2016b] and mean quarterly precipitation was derived from Met 296 Office rainfall data between 1960 to 2016 and held constant. 297 To simulate the sites' land use history, PFT was defined on an annual basis using data on Holocene 298 pollen stratigraphy of the White Peak region of Derbyshire [Taylor et al. 1994 The N14CP model has been previously calibrated and tested against a wide range of site data to 308 provide a general parameter set that is applicable to temperate semi-natural ecosystems, without 309 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License. extensive site-specific calibration [Davies et al., 2016b]. The majority of those parameters are used 310 here for both grasslands. 311 However, two parameters relating to P sources and processes were allowed to vary between the 312 sites: the initial condition for the weatherable P pool, PWeath0; and the rate of plant access to organic 313 P sources, PCleaveMax (Figure 1). 314 We allowed PWeath0 to vary for each grassland as variation in a number of factors including lithology 315 and topography mean that we should expect the flux of weathered P entering the plant-soil system 316 to vary on a site-by-site basis Davies et al. [2016b]. Indeed, we should expect that PWeath0 differs 317 between the acid and calcareous grasslands, as despite their proximity, they have differing lithology. 318 Davies et al. [2016b], show that variation in this initial condition considerably helps explain variance 319 in contemporary SOC, SON and SOP stocks between sites. However, it is difficult to set this 320 parameter directly using empirical data, as information on lithology and P release is limited at the 321 site scale. 322 We also allowed PCleaveMax to vary as this mechanism for P acquisition has been under-explored in 323 previous modelling studies. This is the first time that this model has been knowingly applied to N-P 324 co-limited or P-limited grasslands instead of N-limited sites. Soil organic P has been shown to be an 325 important source of P to plants in P-stressed environments [Balemi and Negisho, 2012]. However, 326 the rates of access to SOP and their controls are relatively poorly understood. We allowed the rate 327 at which P can be cleaved from this pool (PCleaveMax) to vary, to investigate how plant P acquisition 328 might change when more readily accessible P forms become scarcer. , which were more likely to be appropriate for these P-335 poor sites. We explored a range of values for PCleaveMax, from zero where no access to organic sources 336 is allowed, to 1 g m -2 per growing season -a rate in the order of magnitude of a fertilizer application. 337 The model outputs were compared to measured aboveground biomass C, SOC, SON (assumed 338 equivalent to total N) and total P (Table S4) for each grassland. We tested how these parameter sets 339 The outputs for the calibrated model are shown in Figure 2 against the observations for above-365 ground biomass C, soil organic C, and N for both the acidic and calcareous grasslands (Fig 2). The 366 model estimates of above ground biomass C are broadly aligned with the observations: capturing 367 variation between the grasslands and treatments (r 2 =0.58), and on average overestimating the 368 magnitude by 12.9% (SE ± 11.9) and 12.1% (SE ± 9.4) for the acidic and calcareous grasslands 369 respectively (Fig 2a). Soil organic C on average was slightly overestimated (7.1% with SE ± 3.3) for the 370 calcareous grassland (Fig 2b), with a larger average overestimate for the acidic grassland (39.9% with 371 SE ± 6.8). However, in this latter case the variation between treatments was better captured. 372 Simulated magnitudes of SON are well-aligned with observations for the acidic grassland, with an 373 average error of 2.3% (SE ± 3.2), whilst the SON at the calcareous grassland was on average 374 underestimated by 17.8% (SE ± 3.6) (Fig 2c). Finally, the model overestimated total soil P (defined in 375 the model as organic P plus sorbed P) by an average of 6.0% (SE ± 4.3) for the calcareous but 376 underestimated by 54.7% (SE ± 8.0) in the acidic grassland, which was the least accurately predicted 377 variable out of those investigated (Fig 2d). Raw data used for  soil organic carbon, c) soil organic nitrogen and d) total soil phosphorus from both grasslands, with simulated values from the model. The blue line represents a 1 to 1 relationship and the closer the data points are to the line, the smaller the discrepancy between observed and modelled data. All data are in grams per metre squared and all treatments for which data were collected are presented. The horizontal error bars represent the standard error of the empirical data means. The r 2 value of regression models fitted to the data are presented to assess closeness to the 1 to 1 line. https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

The limiting nutrient through time 382
The model suggests that the acidic grassland NPP remained N-limited from 1800 through to 2020 383 under most nutrient treatments (Fig 3). Nitrogen deposition increased the potential NPP through 384 time and the grassland moved toward co-limitation in the LN treatment (i.e. the N and P lines were 385 closer) but remained N-limited (Fig 3b). In the HN treatment, the acidic grassland shifted to P 386 limitation as N-limited NPP surpasses P-limited NPP (Fig 3c). 387 The calcareous grassland was also initially N-limited according to the simulation, but was driven 388 through a prolonged (c. 100 year) state of apparent co-limitation until clearly reaching P-limitation in 389 1950, solely as a result of N deposition (Fig 3). In the 0N treatment, the grassland remained P-limited 390 but the potential NPP values for N and P are similar, suggesting the grassland is close to co-limitation 391 (Fig 3e). The LN and HN treatment amplified pre-existing P-limitation, lowering the potential NPP of 392 the grasslands (Fig 3f, g). With the addition of P in 1995, P limitation is alleviated, and the ecosystem 393 transitions to a more productive N-limited grassland (Figure 3h). 394 Another way to interpret the extent of nutrient limitation within N14CP with specific reference to P-395 demand, is to assess the rate of P cleaving through time. These data corroborate the N and P-limited 396 NPP data, showing that in the calcareous grassland, the maximum amount of cleavable P is accessed 397 by plants in the 0N, LN and HN treatments from approximately 1900 through to the end of the 398 experimental period in 2020 (Fig S1, Table S14), highlighting its consistent state of P or N-P co-399 limitation. 400 Conversely, while P is cleaved in the 0N control treatment in the acidic grassland, it occurs at 401 approximately one third of the total rate, hence the grassland is not entirely P-limited ( Fig S1, Table  402 S10). The LN treatment increases the rate of SOP cleaving and HN causes it to reach its maximum 403 value, confirming the shift to P limitation suggested by the NPP data ( Fig S1, Table S10). Soil organic 404 P cleaving does not occur in the P-treated plots of either grassland. 405 406 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.  (1995). The value of the lines represents the maximum amount of productivity attainable given the availability of N and P separately. Due to a Liebig's law of the minimum approach to plant growth, it is the lowest of the two lines that dictates the limiting nutrient of the grassland and represents actual modelled productivity. Where lines share a value, it can be considered in a state of N-P co-limitation. https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

Modelled trends and responses to nutrient additions 408 409
The model allows the temporal trends and responses to nutrient additions to be further explored. 410 Tables S16 (acidic) and S17 (calcareous). 415 416 3.3.1. Acidic grassland 417 The modelled time series suggest that in the 0N (control) treatment for the acidic grassland, 418 background levels of atmospheric N deposition between the period 1800-2020 resulted in an almost 419 four-fold increase in biomass C, a near-twofold increase in SOC and SON and increased the size of 420 the SOP pool by almost a fifth (Fig 4). 421 Since initiated in 1995, all carbon and nitrogen pools responded positively to N but not P treatments 422 (Fig 5a, c, Tables S7, S8). The LN and HN treatments further increased aboveground biomass C by 423 36.2% and 61.7% (Fig 4a) and increased the size of the total SOC pool by 11.5% and 20.6% 424 respectively (Fig 4c). Similarly, the total SON pool in the acidic grassland increased by 9.7% in the LN 425 treatment and 36.6% in the HN (Fig 4e). 426 Responses of the total SOP pool are in contrast to those of the SOC and SON pools, with LN and HN 427 slightly decreasing SOP by 4.4% and 9.1% respectively, while P addition substantially increased the 428 size of the SOP pool by 76.7% (Fig 4g). Nitrogen treatments facilitated access to SOP from both 429 subsoil and topsoil, increasing plant available P and facilitating its uptake into biomass material (Fig  430   5e, Table S9). 431 432 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

Calcareous grassland 433
Model simulations for the calcareous grassland also suggest N deposition between 1800 and 2020 434 considerably increased aboveground biomass C, SOC and SON pools (Fig. 4), but to a lesser extent 435 than in the acidic grassland. Soil organic C and SON increased by almost half and biomass C more 436 than doubled. Soil organic P accumulated at a faster rate than in the acidic grassland, increasing by 437 about a third (Fig 4, Table S15). 438 Responses of the aboveground biomass C and SOC pools in the calcareous grassland differ greatly to 439 those of the acidic, declining with N addition and increasing with P addition (Fig 4). This response 440 was ubiquitous to all C pools, with declines in subsoil, topsoil and biomass C (Fig 5b, Table S11). 441 Biomass C declined by 2.4% and 7.3% with LN and HN addition (Fig 4b) and SOC declined by 0.5% 442 and 1.4% with the same treatments (Fig 4d). Phosphorus addition increased biomass C and SOC by 443 22.0% and 6.1% respectively (Fig 4b, d). 444 Nitrogen treatments increased the size of subsoil, topsoil and available N pools, but led to small 445 declines in biomass N (Fig 5d, Table S12). The P treatment slightly reduced subsoil and topsoil SON 446 compared to the control yet increased available N and biomass N, to the extent where biomass N is 447 greater in the P than HN treatment (Fig 5d, Table S12). Total SON increased by 6.4% and 15.0% with 448 LN and HN respectively and declined by 0.2% with P treatment (Fig 4f). 449 The response of the calcareous P pools mirrors that of carbon, with declines in subsoil SOP, topsoil 450 SOP, available P and biomass P with LN and HN addition (Fig 5f, Table S13). The calcareous grassland 451 SOP pool declined by 0.2% with LN and 0.5% with HN addition, with an increase of 20.0% upon 452 addition of P (Fig 4h). The P treatment substantially increased total ecosystem P in the calcareous 453 grassland, particularly in the topsoil sorbed pool (Fig 5f, Table S13).  https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

Summary of findings 461
This is the first instance in which N14CP, and to the best of our knowledge; any other integrated C-N-462 P cycle model, has explicitly modelled N-P co-limited ecosystems and investigated their responses to 463 N deposition and additional nutrient treatments. 464 The model suggests that the acidic grassland was characterised by high access to organic P, with 465 comparatively low inorganic P availability, whereas the calcareous grassland was the opposite, with 466 low organic and high inorganic P availability. The selected combinations of PCleaveMax and PWeath0 467 resulted in responses to nutrient addition consistent with N limitation in the modelled acidic, and P 468 limitation in the modelled calcareous grassland. This aligned with our empirical understanding of the 469 two real grasslands with co-N-P limitation being more towards either N or P limitation. 470 The modelling highlighted the contrasting impacts of experimental nutrient treatments on these two 471 grasslands, and provided a means for decoupling the effects of deposition and experimental nutrient 472 manipulation. Most notably, the responses of plant biomass C and SOC to N and P addition were in 473 contrast to one another. In the simulations, N addition led to a small decline in biomass and SOC in 474 the calcareous grassland but a substantial increase in the acidic. Nitrogen addition caused SOP to 475 decline in both grasslands as N treatment exacerbated plant P demand, and increasing P limitation in 476 the calcareous grassland.

Simulating grassland C, N and P pools by varying plant access to P sources 483
Although N14CP was not able to replicate a co-limited response for the acidic site, it produced 484 behaviours akin to the most dominant limiting nutrient for both grasslands across multiple variables, 485 with an average discrepancy between observed and modelled data of only 6.6% (SE ± 9.1) and 1.2% 486 (SE ± 4.4) for the acidic and calcareous grasslands respectively across all variables (Table S5) The differences between P access of the two modelled grasslands could reflect the relative 497 availability of different P sources at Wardlow. The acidic grassland forms in a hillside depression 498 where loess has accumulated, distancing the plant community from the limestone beneath. The 499 plant rooting zone of the acidic grassland is not in contact with the bedrock, so roots almost 500 exclusively occur in the presence of organic P sources which can be cleaved and utilised by plants 501 [Caldwell, 2005;Margalef et al., 2017]. Conversely, the calcareous soil rarely exceeds 10 cm depth, 502 and the rooting zone extends to the limestone beneath. This provides plants with greater access to 503 weatherable calcium phosphate [Smits et al., 2012]. 504 The rate of organic P access was sufficiently high in the acidic grassland to temporarily overcome P 505 limitation induced by anthropogenic N deposition. Due to its lower PCleaveMax, the calcareous 506 grassland was unable to meet additional P demand driven by N addition, and thus remained P-507 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.
limited. It should be noted that the model grossly underestimates the acidic TP observations (Fig 2d), 508 as few parameter sets where simultaneously able to simulate the magnitude of the empirical TP pool 509 and the N-limited response of the acidic grassland to nutrient manipulations. Data that distinguishes 510 between organic and inorganic forms of P would help in understanding this discrepancy. 511 N14CP has a number of mechanisms to account for N and P interdependence, meaning that in 512 principle, it is capable of simulating N-P co-limited behaviour. Indeed, we found signs of N-P co-513 limited behaviour in both grasslands as nutrient treatment altered the limiting nutrient. Available N 514 in the calcareous grassland was marginally greater in the P than 0N treatment (but less than LN and 515 HN) (Fig 5d, Table S12), suggesting plants may be using surplus P to acquire N when it becomes 516 limiting. Calcareous biomass N was also highest in the P treatment, though this reflects an absolute 517 increase in N resulting from stimulated growth, and not a substantial acquisition of N from another 518 pool (Fig 5d, Table S12). Similar behaviour was found in the modelled acidic grassland, where LN and 519 HN treatments increased N availability, promoting access to available P (Table S9) and facilitating 520 growth under N addition when it was largely P-limited (Fig 3c). 521 Nitrogen fixation remained unaffected by nutrient treatment in both grasslands (Tables S8, S12). 522 This may be an unintended outcome of another N-P interaction within N14CP, whereby N fixation is 523 downregulated by atmospheric N deposition [Gundale et al., 2013]. However, when N deposition 524 exceeds fixation (as at Wardlow), fixation is essentially nullified (as in Tables S8, S12), meaning 525 deposition becomes the sole source of N to the grassland. This in effect, removes the dependence of 526 N acquisition on P availability, and could make modelling 'true' N-P co-limitation [Harpole et al.,527 2011] under high levels of N deposition challenging. This could be further explored by allowing N 528 fixation limits in the model to adapt to P nutrient conditions. 529 530 531 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

The limiting nutrient through time 532
There is some evidence to suggest that modelled transitions of the limiting nutrient may be 533 representative of historical nutrient limitation at Wardlow. Recent (post 1995) strengthening of P 534 limitation (Fig 3g), transition to N limitation in the P-treated calcareous plots (Fig 3h), and transition 535 to P limitation in the acidic HN treatment (Fig 3c), are likely to be accurate representations of the 536 trends in nutrient limitations at the Wardlow grasslands. 537 Strengthening P limitation in both the acidic and calcareous grasslands under increased N input is 538 supported by observations of increased root surface phosphatase enzyme activity in LN and HN 539 treatments [Johnson et al., 1999;Phoenix et al., 2004] that indicate increased P demand. 540 Furthermore, N deposition acidifies soil [Horswill et al., 2008], potentially reducing the availability of 541 mineral P by facilitating the formation of iron and aluminium complexes which act to immobilise P 542 [Kooijman et al., 1998]. Indeed, the model simulated reductions in plant available P for the 543 calcareous grassland in response to the LN and HN treatments (Table S13) its P-limited condition under the HN treatment (Fig 3c), the acidic grassland continued to accumulate 564 biomass with N addition as the grassland's greater access to topsoil SOP (Table S9) allowed it to 565 acquire sufficient P to stimulate additional growth but not necessarily to alleviate P limitation. This is 566 consistent with the acidic grassland at Wardlow, where N treatment stimulated root surface 567 phosphatases, likely supplying more SOP to plants [D Johnson et al., 1999]. 568 In the acidic grassland, LN and HN addition increased SOC almost linearly (Fig 4c). Biomass C and SOC in the calcareous grassland responded positively to P addition, via similar 575 mechanisms to the N-response in the acidic grassland. However, in contrast to the acidic grassland, 576 N addition caused declines in calcareous biomass and SOC, the former of which has been observed 577 at the calcareous grassland at Wardlow [Carroll et al., 2003]. Reductions in calcareous biomass C 578 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.
(and consequently SOC) in the model are a combined result of reductions in bioavailable P (Table  579 S13), occurring via N-driven increases in stoichiometric P demand, in addition to an inability to 580 access sufficient P from the SOP pool (Table S14). Plants therefore cannot meet P demand and new 581 biomass is insufficient to replace senesced plant material, decreasing net biomass C input to the SOC 582 pool. 583 Our results are consistent with findings by Li et al. [2018], who show that N fertilisation of an N-P co-584 limited grassland reduced SOC stocks by 5-12%, which they attribute to changes in community 585 composition toward a higher proportion of forbs, whose lower tissue C:N increases the 586 decomposability of litter input to the soil, and more rapid microbial degradation of SOC [Li et  Such intricate interactions between soil microbes and N-driven acidification are not detailed within 598 N14CP, therefore, our conclusion that N addition decreases P-limited SOC stocks is attributable to 599 reduced C input rather than increased C output. and HN treatment, worsening P limitation in the calcareous grassland, and depleting the SOP pool in 616 the acidic. While SOP declined in both grasslands, the responses of available and biomass P to 617 nutrient treatments differed markedly between the grasslands. Due to the higher rate of PCleaveMax in 618 the acidic grassland, more P accumulated in the plant-available pool and hence P does not become 619 the limiting factor under N treatments (Table S9). Conversely, available and biomass P decline under 620 LN and HN addition in the calcareous grassland (Table S13), highlighting how calcareous PCleaveMax 621 capability is insufficient to meet increased P demand. 622 623 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.

Conclusions 624
We have shown that by varying two P-acquisition parameters within N14CP, we can account for 625 contrasting responses of two N-P co-limited grasslands to long-term nutrient manipulation with 626 reasonable accuracy. This suggests that current measures to account for co-limitation within the 627 model are to some extent sufficient and widely applicable, at least to N-P co-limited ecosystems that 628 are close to N or P limitation. Flexible organic P access allowed the modelled acidic grassland to 629 acquire sufficient P to match the available N from chronic deposition and prevent 'anthropogenic P 630 limitation'. However, the model suggests that this is an unsustainable strategy, as the SOP pool 631 rapidly degrades, and if N additions are sustained, P limitation becomes likely. Conversely in the 632 calcareous grassland, which was less able to access organic P, additional N provision exacerbated 633 pre-existing P limitation. 634 We further show that anthropogenic N deposition since the onset of the industrial revolution has 635 had a substantial impact on the C, N and P pools of both the acidic and calcareous grasslands, to the 636 extent where almost half of contemporary soil carbon and nitrogen in the model could be from, or 637 caused by, N deposition. Experimental N and P addition had contrasting impacts on the simulated 638 grasslands. In the acidic grassland, N treatment stimulated plant access to soil organic P pools, 639 promoting plant growth and soil carbon sequestration. However, in the calcareous grassland, further 640 N addition simultaneously increased plant P demand and reduced its availability, decreasing plant 641 carbon input to the soil and leading to degradation of soil carbon. Our work therefore suggests that 642 as N deposition shifts more ecosystems toward a state of P limitation or strengthens it where it 643 already occurs [Goll et al., 2012], we may see reductions in sequestration to the point where 644 declines in grassland SOC stocks -one of our largest and most labile carbon pools -may occur. 645 646 https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License. https://doi.org/10.5194/bg-2020-392 Preprint. Discussion started: 9 November 2020 c Author(s) 2020. CC BY 4.0 License.