CO<sub>2</sub> Fertilization to Climate Limitation: Shifting Drivers in a Dryland Forest

Irob, Katja; Chang, Liling; Preisler, Yakir; Fiedler, Sebastian; Büchner, Matthias; Grünzweig, José M.; Sheffer, Efrat

doi:10.5194/egusphere-2026-2448

Preprints

https://doi.org/10.5194/egusphere-2026-2448

Preprints

05 May 2026

| 05 May 2026

CO₂ Fertilization to Climate Limitation: Shifting Drivers in a Dryland Forest

Katja Irob, Liling Chang, Yakir Preisler, Sebastian Fiedler, Matthias Büchner, José M. Grünzweig, and Efrat Sheffer

Abstract. Dryland forests represent a significant but uncertain component of the terrestrial carbon sink, where rising CO₂ and intensifying drought and heat stress exert opposing controls on carbon and water fluxes. Using ED2.2-hydraulics calibrated against 21 years of flux tower and inventory data from Yatir Forest, an arid pine plantation in Israel, we simulated carbon fluxes, biomass dynamics, and water-use efficiency under current climate and high-emission scenarios (SSP3-7.0, SSP5-8.5) through 2100 using five CMIP6 projections.

CO₂ fertilization initially enhanced productivity and biomass accumulation despite periodic drought, with the strongest response under SSP3-7.0. Under SSP5-8.5, compound heat and drought stress suppressed productivity below SSP3-7.0 levels despite higher CO₂ concentrations, triggering stand collapse in two projections and eliminating approximately 40 % of accumulated biomass. GAM-based driver decomposition showed that forest responses shifted from CO₂-dominated to interaction-dominated by late century, with compound heat and drought stress explaining 25–63 % of variance. Apparent biomass gains under severe scenarios reflected accumulation in fewer, larger surviving individuals as stand density declined, masking structural deterioration in aggregate metrics.

Functional decline (NEP, WUE) preceded structural changes by 12–37 years depending on scenario, providing an early warning window before visible deterioration. These results show that CO₂ fertilization benefits cannot compensate for the compound climate extremes that accompany high emissions, and that functional indicators must be monitored alongside structural metrics to detect forest vulnerability in dryland afforestation systems.

Received: 28 Apr 2026 – Discussion started: 05 May 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1341 KB)

Supplement (3496 KB)

Download & links

Katja Irob, Liling Chang, Yakir Preisler, Sebastian Fiedler, Matthias Büchner, José M. Grünzweig, and Efrat Sheffer

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-2448', Anonymous Referee #1, 15 Jun 2026
Comment to “CO₂ Fertilization to Climate Limitation: Shifting Drivers in a Dryland Forest” by Irob K et al.
This study projects future drought impacts on the Yatir Forest afforestation using the ED2 ecosystem model under high-CO₂ and warming scenarios. Understanding whether dryland afforestation can maintain productivity and carbon sequestration under intensifying drought is critical for climate adaptation policy in water-limited regions. However, I raise a main concern that the model's calibration/validation may be inadequate—specifically, that GPP and ET dynamics are not well captured even in the improved hydraulic version—which fundamentally undermines confidence in the projections.
Here are the main points to address:
The model was calibrated against 20 years of FLUXNET data, including ET and carbon fluxes (Section A2), but critically, the validation metrics (R², RMSE, bias, correlation for GPP, ET, soil moisture, LAI) are not reported in the main text or as a table. Readers cannot assess whether ET is well-captured (R² = 0.8?) or mediocre (R² = 0.5?), or whether GPP peaks are matched or missed. These numbers appear only in the supplementary repository, requiring readers to download and run code to assess model performance. This is problematic because readers cannot evaluate whether key fluxes are well-captured or whether discrepancies might propagate into projections.

The authors acknowledge limitations in GPP representation (Figures 2 and S2 show seasonal peaks that are missed, particularly in spring and early summer). However, the quantitative extent of these mismatches is unclear in the main text. Line 278 reports that 'model–data agreement was strongest for gross primary productivity (GPP; 89.3%, Fig. 2c),' but what do the seasonal error metrics show? For example, if spring GPP RMSE is ±25% while annual RMSE is ±10%, this distinction matters: spring is when the forest builds up carbohydrate reserves critical for drought survival. Without explicit RMSE, R², and bias reported by season in the main text, readers cannot assess whether the model's representation of pre-drought productivity is sufficient to trust projections of forest resilience. Given that GPP mismatches propagate into all downstream carbon projections (NEP, NPP, mortality timing), quantifying these errors explicitly would strengthen confidence in the study's conclusions—or appropriately flag where uncertainty is largest. Similar considerations apply to ET validation.

ED2 was developed for tropical and temperate forests and is rarely applied to semi-arid ecosystems. While the authors added a hydraulic module to improve water-stress representation, the model's core structure (patch-level demography, implicit hydrodynamic coupling) may not be optimal for extreme drought environments where water availability is the primary limiting factor. The calibration difficulties (see above) raise the question: would a simpler, water-focused model (e.g., dynamic global vegetation model with explicit soil-water balance) or a site-optimized model be more appropriate? The authors should justify why ED2 is suitable despite these constraints, or acknowledge this as a limitation.

Other comments:
Several decision thresholds need supporting references and justification. The 15-year rolling window for defining 'significant decline' (lines 237–239) is presented without explanation—why 15 years rather than 10 or 20? Does literature on forest productivity dynamics support this window, or does sensitivity analysis show robustness to alternative windows? The 0.5% threshold for NEP collapse similarly needs supporting references. Additionally, the time-lag assumptions discussed on line 242 should cite physiological or phenological literature documenting these delays in Mediterranean pines or similar ecosystems.

The choice of SSP5-8.5 (very high emissions, weak mitigation) also warrants discussion. This scenario is now considered less likely than moderate pathways (SSP3-7.0, SSP2-4.5) given recent climate policy. While upper-bound risk scenarios are valuable, how sensitive are key findings (dieback onset, NEP collapse timing) to scenario choice? A brief comparison across scenarios—or explicit framing of SSP5-8.5 as an upper-bound risk rather than a most-likely forecast—would help readers contextualize the conclusions.

Line 338 describes 'spikes' in variables following dieback events, but the mechanism is unclear. Do remaining trees compensate through increased growth? Is this an artifact of the model's mortality algorithm releasing light and water to survivors? What is the magnitude, and is it biologically realistic? This needs explicit explanation because it affects interpretability of post-disturbance dynamics.

Line 346 states that NEP decline onset is later under SSP5-8.5 than under milder scenarios (e.g., SSP3-7.0). This is counterintuitive: we would expect the most severe warming scenario to show earlier and more severe decline, not later. The authors note "larger variability between simulations" under SSP5-8.5, but this does not explain why the mean decline onset shifts later under the warmest scenario. Is this due to fewer ensemble members converging under SSP5-8.5, creating a statistical artifact? Are there nonlinear dynamics (e.g., early hydraulic failure followed by brief recovery before collapse) that delay mean onset? Is this an error in temporal alignment or scenario attribution?

The paper provides valuable analysis of which drivers (CO₂, temperature, water stress) explain variance among simulations (Figure 4). However, equally important would be understanding what drives the mean projected change and how important each driver is to the central estimate. A complementary analysis (e.g., a figure similar to Figure 4 but showing driver contributions to mean NEP and GPP trajectories) would clarify whether the ensemble mean decline is primarily driven by water stress, temperature, or declining CO₂ benefit would be really appreciated.

Suggestion: The paper analyzes vulnerability—when and how the forest declines—providing important information for understanding forest resilience. An analysis framing these projections in terms of “net carbon benefit” would strengthen the manuscript and enhance its value to policymakers. Over the full projection period, what is the cumulative carbon stored in the Yatir stand under each scenario (Mg C/ha or Tg C total)? If collapse occurs in 2060 under SSP5-8.5, is the cumulative carbon benefit positive or negative relative to the cost of establishment and management? Even a straightforward calculation (e.g., 'SSP3-7.0 yields ~250 Mg C/ha by 2100; SSP5-8.5 yields ~150 Mg C/ha before collapse') would frame results in terms of afforestation viability and help land managers and policymakers weigh uncertainty. This would transform the analysis from a purely vulnerability assessment into decision-relevant information about whether this afforestation is 'worth it' under climate scenarios
Final comment:
This study addresses an important question regarding the viability of dryland afforestation under climate change. The analysis is interesting and novel for this ecosystem type, and projecting potential dieback under intensifying drought is a valuable contribution to climate adaptation research. However, the solidity of the results needs to be better established. The validation metrics—R², RMSE, bias, and seasonal error estimates—are currently relegated to the supplement repository (not suppl. info), making it difficult for readers to assess whether model discrepancies are acceptable for the study's key conclusions about forest vulnerability and carbon dynamics. These metrics must be brought into the main text so that the credibility of the projections can be properly evaluated. With the revisions addressing model validation and result interpretation, the work could provide an interesting contribution for understanding afforestation resilience in dryland regions.
Citation: https://doi.org/10.5194/egusphere-2026-2448-RC1
- AC1: 'Reply on RC1', Katja Irob, 30 Jun 2026
  
  The comments have meaningfully strengthened the manuscript, particularly regarding the presentation of validation metrics, model justification, and policy-relevant framing of the results. Below we summarise our responses and outline the corresponding manuscript changes. A full revised manuscript with track changes will be submitted upon the editor's invitation.
  Main Concern: Validation Metrics (Comments 1 and 2)
  Standard validation metrics have been added to the main text. Table 1 reports R², RMSE, and bias for all evaluated variables; Table S1 reports seasonal breakdown for GPP and ET. Key metrics: GPP R² = 0.86; NEP R² = 0.79; soil moisture R² = 0.89; LAI R² = 0.92. ET showed moderate overall agreement (R² = 0.70, r = 0.84); summer ET underestimation is large in relative terms (66%) but small in absolute terms (bias: −11.4 mm month⁻¹ against an observed mean of ~17 mm month⁻¹), reflects conservative hydraulic downregulation under peak VPD stress, consistent with observed stomatal behaviour in P. halepensis (Klein et al., 2016; Preisler et al., 2021, 2022), and has negligible impact on carbon projections given near-zero summer carbon fluxes. Spring GPP underestimation (38%) constitutes a conservative bias on early-century CO₂ fertilization benefits, as acknowledged in the Discussion.
  Comment 3: Model Choice Justification
  The reviewer's concern about implicit hydrodynamic coupling accurately describes standard ED2, not the version used here. ED2.2 with hydraulics explicitly resolves xylem water potential gradients, stomatal regulation, and species-specific vulnerability curves. Size-structured demography and sub-daily temporal resolution are requirements, not preferences: the lag dynamics and cohort-level mortality central to our findings cannot be reproduced by simpler annual-timestep models. Few dryland modeling studies evaluate performance simultaneously across carbon fluxes, water fluxes, vegetation structure, and radial growth; achieving R² of 0.70 to 0.92 across all four domains substantially increases confidence in the model's process representation. Clarifying statements have been added to the Methods and Discussion.
  Comment 1: Decision Thresholds
  The 15-year rolling window filters dryland interannual variability while detecting trends within the projection period, consistent with comparable forest productivity studies (Anderegg et al., 2020; Reichstein et al., 2013). The 0.5% threshold applies to AGB growth stagnation, not NEP collapse as stated in the review: it reflects the point where new biomass production no longer exceeds turnover and mortality losses. Time-lag citations (Anderegg et al., 2019; Hartmann et al., 2022; Klein et al., 2016; Tatarinov et al., 2016) have been added to the Methods section.
  Comment 2: SSP5-8.5 Scenario Framing
  The manuscript already frames SSP5-8.5 as 'the upper end of plausible radiative forcing.' SSP5-8.5 was standard practice at the time of submission and was only formally removed as a CMIP7 priority scenario in April 2026, after submission. A sentence has been added to the Methods making the upper-bound framing explicit.
  Comment 3: Post-dieback Spikes
  The LAI and WUE spikes following dieback reflect resource release to surviving cohorts, a mechanistic process in ED2.2's demographic structure, not a model artifact. The WUE spike is partly a ratio effect as transpiration falls faster than GPP after density reduction. The pattern is consistent with experimental thinning results in Israeli P. halepensis plantations (Calev et al., 2016). A clarifying sentence has been added noting that the recovery is transient.
  Comment 4: Later NEP Onset under SSP5-8.5
  NEP decline was detected in only three of five SSP5-8.5 models; the mean first year decline in 2053 ± 16 is computed from those three only, as is standard when not all ensemble members trigger the threshold. The later mean onset reflects two factors: gfdl-esm4's very late detection (2071) due to abrupt collapse, and CO₂ buffering of carbon assimilation in surviving individuals under reduced competition, delaying sustained functional decline despite more severe climate stress. The pattern has been verified against Table S2 and is robust. A clarifying paragraph has been added to the results.
  Comment 5: Driver Contributions to Mean Trajectory
  We attempted three statistical approaches to produce a mean driver attribution for NEP and GPP. All three failed to reliably attribute water stress to NEP, because water stress operates primarily through nonlinear interactions with CO₂ and temperature rather than as an additive main effect. With only five GCMs, fitting interaction terms on the mean trajectory lacks sufficient statistical power.
  The cross-simulation variance decomposition in Figure 4 nonetheless addresses the reviewer's question indirectly. All GCMs share the same CO₂ forcing pathway per scenario, so variance across models reflects differences in their climate trajectories. The driver decomposition therefore captures the forest's sensitivity to water stress, temperature, and CO₂ across the plausible range of future climate conditions, making it informative about the drivers of the central estimate as well as the spread between models.
  Suggestion: Cumulative Carbon Benefit
  Despite periodic source years, over the period 2015 to 2100 the forest accumulated net carbon under both future scenarios, with CO₂ fertilization driving cumulative sequestration despite intensifying drought stress: 106 Mg C ha⁻¹ under SSP3-7.0 (range 77 to 139 Mg C ha⁻¹) and 70 Mg C ha⁻¹ under SSP5-8.5 (range 62 to 85 Mg C ha⁻¹). We have added these figures to the Discussion to provide the policy-relevant framing the reviewer suggests.
  We believe these revisions substantially improve the manuscript's credibility and policy relevance. We look forward to the continued discussion and welcome any further comments from the referee or the broader community.
  
  Citation: https://doi.org/10.5194/egusphere-2026-2448-AC1
RC2:
'Comment on egusphere-2026-2448', Anonymous Referee #2, 01 Jul 2026
Summary:
This paper addresses the response of dryland forests to future climate projections under a range of scenarios. To do so, the authors undertook the substantial task of developing a new P. halepensis PFT for the Ecosystem Demography model (ED2). The paper is timely, as our understanding of vegetation dynamics in drylands, and in particular the ability of models to represent them under current and future climate, remains limited. Most DGVMs, including ED2, lack both the parameters and the processes needed to capture dryland vegetation adequately.
That said, I think the paper would benefit from a more in-depth analysis of model performance, as I outline below. Future projections are only informative if we can first agree that the model reproduces current observations well. I hope the comments below are helpful to the authors.
Major comments:
SSP5-8.5 is one of the most extreme scenarios. Recent work suggests such high-forcing pathways are increasingly viewed as unlikely, and CMIP7 has retired them from its core set (https://gmd.copernicus.org/articles/19/2627/2026/). I am not asking the authors to redo the analysis, but a strong justification for why SSP5-8.5 remains informative in this context should be provided.

In the Results, it is unclear why the model was averaged over 20 years (e.g., Figure 2 and A2). Dryland forests are highly sensitive to environmental variability such as precipitation, as the authors themselves emphasize. It would be more informative to see whether ED2.2 with the new PFT captures seasonal patterns when they matter most, for example during exceptionally wet or dry years. Additional analysis of temporal dynamics such as phenology, trends, and interannual variability would strengthen the validation. The main text also states that R² and RMSE values are provided in SI A2, but I could not locate them there.

The reported agreement scores are somewhat misleading. LAI shows 86.6% agreement, yet Figure 2 makes clear that the seasonal dynamics, which are the key feature, are not captured well. GPP has the highest agreement score, but Figure 2 suggests this is driven by low-activity months where GPP is near zero, while during the active season the simulation falls well below observations. A more robust validation is needed before moving to future projections.

Section 3.2.1: dieback events appear near 2080 in some projections. The authors attribute this to CO₂-driven biomass accumulation producing large trees with a negative carbon balance. I am not familiar with the site, but I doubt this is a real ecological signal. It more likely reflects a model artifact, for example from allocation equations that may not be appropriate for this PFT. This warrants careful investigation. The Discussion refers to "maintenance respiration costs of the largest trees eventually exceeded their carbon assimilation," but this is never explicitly demonstrated in the main manuscript. Figure S8 shows time series of the relevant fluxes but does not demonstrate the proposed mechanism directly. A simple plot of GPP against total respiration over time would show whether respiration ever exceeds GPP, if that is indeed the driver.

In the same figure, some simulations show AGB dropping while LAI increases abruptly at the same time (Figure 3a.4). This is not a plausible physiological response and again suggests a model issue. One would also expect a sharp change in structural variables to be mirrored by functional variables such as GPP, but this is not the case. Functional responses (e.g., WUE) show noisy behavior across the century with no consistent pattern across models or scenarios.

Section 3.4: some results here need more careful examination. For instance, how can CO₂ explain 85% of the variance in DBH but close to 0% for AGB, given that these variables are closely correlated? The early/mid/late categorization also appears somewhat arbitrary: CO₂ explains 83% of the variance in mean DBH in the mid period but drops abruptly to 32% in the late period, which is difficult to reconcile with a smoothly evolving system.

The Discussion provides limited treatment of model performance itself, which is a notable gap given that this study introduces a new halepensis PFT for ED2.2. Section 4.4 attempts to address this but reads as a mix of literature review and brief acknowledgment of results, and I have mixed feelings about how it is currently structured. A diagnosis of how the identified limitations affect the simulations under this new PFT would be more valuable. For example, the spring GPP underestimation is attributed to "fixed physiological parameters," but which ones? Vcmax? Stomatal conductance? Phenology triggers? Without this, the diagnosis is not actionable for others adapting the model. I would also encourage the authors to offer a prioritized roadmap for model development rather than an undifferentiated list of missing processes.

Minor comments:
Abstract: define GAM at first use.

The manuscript alternates between "forest" and "dryland forest" throughout. This is confusing when processes and consequences are described, especially for readers less familiar with dryland systems, which differ from other forest types in important ways. Paragraphs three and four of the Introduction, for instance, describe processes without clarifying whether they apply to dryland forests specifically or to forests in general. If the latter, it reads as overgeneralization. Since dryland forests are the focus, the Introduction should stay in that context as consistently as possible.

Are these planted forests actively managed? If so, please state it in the paper.

Lines 174–175: was the "ISIMIP3BASD v2.5 statistical bias adjustment" applied by the authors, or is it an external product used as delivered?

Lines 194–198: the thresholds used for the drought severity index need justification. The same applies to the temporal periods (2015–2040, etc.). As written, these choices appear subjective.

Lines 208–212: a dense block of statistical tests is listed without explanation. A brief description of each test and why it was chosen, either here or in the SI, would help the reader.

Section 2.6.1: how was the drought decomposition actually done? This is not clearly explained.

Lines 237–240: the rationale for the specific threshold values is unclear.

Figure 3: the two shades of orange used for the two scenarios are difficult to distinguish. A more contrasting color scheme would help.

Section 3.3 would benefit from a graphical time series showing the onset of functional decline followed by structural changes (e.g., LAI). This would make the early warning argument much more tangible for the reader.

Thanks!
Citation: https://doi.org/10.5194/egusphere-2026-2448-RC2

Katja Irob, Liling Chang, Yakir Preisler, Sebastian Fiedler, Matthias Büchner, José M. Grünzweig, and Efrat Sheffer

Supplement

https://doi.org/10.5194/egusphere-2026-2448-supplement

Model code and software

ED2 GitHub page ED2 developers https://github.com/EDmodel/ED2

Katja Irob, Liling Chang, Yakir Preisler, Sebastian Fiedler, Matthias Büchner, José M. Grünzweig, and Efrat Sheffer

Viewed

Total article views: 388 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
308	64	16	388	28	13	10

HTML: 308
PDF: 64
XML: 16
Total: 388
Supplement: 28
BibTeX: 13
EndNote: 10

Views and downloads (calculated since 05 May 2026)

Month	HTML	PDF	XML	Total
May 2026	272	50	10	332
Jun 2026	36	14	6	56
Jul 2026	0

Cumulative views and downloads (calculated since 05 May 2026)

Month	HTML	PDF	XML	Total
May 2026	272	50	10	332
Jun 2026	36	14	6	56
Jul 2026	0

Viewed (geographical distribution)

Total article views: 382 (including HTML, PDF, and XML) Thereof 382 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 05 Jul 2026

Short summary

Dryland forests are promoted for climate mitigation, but their persistence under future climate is uncertain. Using a calibrated forest model and 21 years of field data from an arid pine forest, we simulated carbon uptake and growth through 2100. Rising carbon dioxide initially boosted productivity, but compound drought and heat stress later overrode these benefits. Declining carbon uptake preceded visible structural deterioration by up to 37 years, an early warning window for managers.


Total:	0
HTML:	0
PDF:	0
XML:	0

CO2 Fertilization to Climate Limitation: Shifting Drivers in a Dryland Forest

Supplement

Model code and software

Viewed

Viewed (geographical distribution)

CO₂ Fertilization to Climate Limitation: Shifting Drivers in a Dryland Forest