the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Nov 2025
Exploring the Potential of LPJmL-5 to Simulate Vegetation Responses to (Multi-Year) Droughts
Abstract. Climate change is expected to increase the frequency and severity of Multi-Year Droughts (MYDs), but their impacts on vegetation remain poorly understood. While satellite records offer valuable insights, they cover only recent decades, limiting the number of MYDs available for analysis. Dynamic global vegetation models (DGVMs), such as LPJmL-5, can help overcome this limitation by simulating vegetation dynamics over longer timescales. However, their ability to capture drought impacts has not yet been systematically evaluated. In this study, we benchmark LPJmL-5 against MODIS-derived gross primary production (GPP) to assess how well it captures vegetation responses to drought. We find that LPJmL-5 reproduces GPP reasonably well in some regions, but improvements can still be made in the Southern Hemisphere and for croplands. During MYDs, LPJmL-5 captures the key temporal and spatial GPP drought dynamics observed in MODIS. However, the model tends to overestimate vegetation response at the onset of MYDs and shows some rapid recovery behaviour, resulting in muted overall drought impacts. Vegetation responses also vary by type: croplands show relatively good agreement, while boreal and temperate vegetation underestimate positive and negative impacts, respectively. These discrepancies appear to be linked to simplified model representations of vegetation stress and mortality, which limit long-term vegetation loss. Our results highlight the need to improve how LPJmL-5 simulates vegetation stress and recovery, especially under prolonged drought conditions, in order to better capture ecosystem vulnerability in a changing climate.
- Preprint
(7825 KB) - Metadata XML
-
Supplement
(4819 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-4966', Anonymous Referee #1, 02 Dec 2025
-
AC1: 'Reply on RC1', Denise Ruijsch, 15 Dec 2025
We thank the reviewer for the detailed comments regarding the use of SPEI and MODIS GPP.
1. SPEI is a widely used to characterize drought conditions particularly because it is standardized and as such always relates to the local climatic context (see Slette et al. 2019). However, this also poses some issues as outlined in Zang et al., (2020). The main argument of Zang et al., is that negative SPEI (e.g. -1 to -2; often characterized as moderate drought) does not necessarily always coincide with "actual" water shortage as determined by the amount of evapotranspiration subtracted from the precipitation. This seems to be the case especially in wet regions of the world such as the tropics or the boreal forest. The consequence is that SPEI may indicate drought conditions when in reality ample water is available (as determined for example by the Maximum Climatic Water Deficit commonly used in tropical regions).
We thank the reviewer for highlighting this important point regarding SPEI and its interpretation in wet regions. In our study, we use the term “drought” to refer to periods of below-average water availability relative to the local climate. This reflects a relative deviation from normal conditions. To address the reviewer’s point, we will examine whether soil moisture is low during the multi-year drought periods identified in this study. This will help confirm whether the droughts indicated by SPEI correspond to actual stress experienced by the ecosystem.
2. The second issue is the use of MODIS GPP as a stand-in for "observed GPP". Satellite-derived GPP is as much a model as it is based on satellite observations. In the case of MODIS GPP, GPP is calculated by an algorithm using remotely sensed FPAR, a land cover classification, a parameter for the conversion efficiency of PAR, and some climate inputs (Tmean, VPD). Consequently, it is not really accurate to consider satellite-derived GPP from MODIS as "observed". This obviously does not invalidate the comparison with simulated GPP from LPJmL-5, however, it does warrant some further discussion and potentially the use of an alternative product (e.g. SIF) to strengthen the message of this study.
For the use of MODIS GPP, we agree that it is a model-derived product based on satellite observations and not a direct observation of GPP. We will clarify this in the manuscript and expand the discussion on the limitations of satellite-derived GPP.
Regarding land cover, LPJmL-5 was run with prescribed land cover, including crop distributions, which ensures that the land-cover types used by LPJmL-5 match those of MODIS. Further details on the differences between prescribed and dynamic land cover in LPJmL-5 are provided in Supplement Section S4.
Finally, we have already explored the consistency across MODIS vegetation indices in our comparison of GPP and EVI in Supplement Section S2. Also, Ruijsch et al. (2025) analysed the correlation between GPP, EVI, and LAI during multi-year droughts. Both analyses show broadly similar patterns across the different vegetation indices. To further strengthen the study, we will extend this comparison by including SIF.
Citation: https://doi.org/10.5194/egusphere-2025-4966-AC1
-
AC1: 'Reply on RC1', Denise Ruijsch, 15 Dec 2025
-
RC2: 'Comment on egusphere-2025-4966', Anonymous Referee #2, 15 Dec 2025
This study presents a comprehensive set of analyses; however, its scientific motivation and added value remain unclear. The work mainly consists of descriptive comparisons between LPJmL-5–simulated GPP and MODIS-derived GPP in terms of mean states, variability, drought anomalies, and responses to SPEI across multiple timescales. Such comparisons, by themselves, do not address a clearly defined scientific question, nor do they provide new insight into vegetation responses to drought beyond documenting discrepancies between two fundamentally different GPP products.
Importantly, MODIS GPP is itself a model-based, remotely sensed product rather than a direct observation. As a result, differences between LPJmL-5 and MODIS GPP cannot be straightforwardly interpreted as model deficiencies or ecological mechanisms. As currently presented, the results primarily show that the two products differ substantially, but it remains unclear how these differences advance understanding of ecosystem drought responses or meaningfully address the stated limitation of short satellite records.
Although the authors argue that DGVMs can overcome the temporal limitations of satellite observations, the study does not demonstrate that LPJmL-5 provides additional or improved insight into vegetation responses to long-term or multi-year droughts. No independent evaluation at longer timescales is provided, nor is it shown that the model improves the representation of drought processes or stress regulation. Consequently, the claimed advantage of using LPJmL-5 to assess long-term ecosystem vulnerability is not sufficiently supported.
Overall, the study lacks a clear hypothesis, mechanistic advancement, or demonstrated improvement in modeling drought impacts. Without evidence that LPJmL-5 adds robust, independent information beyond existing GPP products, the analysis remains largely diagnostic and does not justify the broader conclusions drawn.
Some specific comments
1- I question whether the comparison between LPJmL-5 GPP and MODIS GPP is meaningful in the first place, given that the two products differ substantially in spatial resolution, temporal aggregation, and conceptual formulation. LPJmL-5 operates at a coarse grid scale and simulates GPP through process-based parameterizations, whereas MODIS GPP is a remote sensing–based product with its own empirical assumptions and limitations. Under these circumstances, differences between the two cannot be unambiguously attributed to vegetation response characteristics.
Moreover, the choice of MODIS GPP as the sole benchmark appears arbitrary. Using an alternative satellite-based product with different resolution and algorithms (e.g., VPM GPP at 500 m and 8-day resolution) would likely lead to different comparison outcomes. This raises a fundamental question that the study does not address: what exactly do these inter-product differences represent, and on what basis can one conclude that one product provides a more correct or ecologically meaningful representation of drought response than another?
2-Although the use of SPEI-12 is common in studies of multi-year droughts (MYDs), its role as an optimal or sufficient drought indicator has not been systematically validated. Importantly, different ecosystems exhibit markedly different sensitivities to drought at various temporal scales. For instance, croplands and grassland ecosystems typically respond more strongly to short-term moisture anomalies (e.g., SPEI-1 to SPEI-6), whereas the strong smoothing inherent in SPEI-12 may obscure critical short-duration drought signals, potentially leading to biased interpretations of ecological responses. In addition, longer aggregation periods can delay the detection of drought onset and termination, which may affect the characterization of drought event dynamics.
Another concern is a drought event identified from an index such as SPEI only captures the potential drought stress on vegetation relative to the long-term normal period and may not necessarily reflect the actual water deficit of an event.
3- The comparison presented in Figures 4 and 5 does not provide a valid basis for any inference about ecosystem drought resistance. The observed differences between LPJmL-5 and MODIS GPP primarily reflect systematic discrepancies in GPP construction, temporal sensitivity, and model structure, rather than differences in vegetation response or regulation in an ecological sense. In particular, LPJmL-5 does not include any explicit parameterization or calibration targeting ecosystem resistance, buffering capacity, or drought regulation, making it conceptually unjustified to interpret shorter response timescales as reduced drought resistance.
Figure 4 explicitly demonstrates that the model produces sharper and more immediate GPP declines during dry periods, which is a direct consequence of the model’s tight coupling between soil moisture and photosynthesis. This behavior reflects model design choices rather than ecosystem properties, and therefore cannot be interpreted as evidence of altered resistance or response dynamics. Without explicitly separating response speed, response magnitude or demonstrating that these metrics are meaningfully constrained by observations, the presented analysis does not support any ecological conclusions regarding drought resistance. As currently formulated, this comparison is fundamentally methodological and cannot substantiate claims about ecosystem response mechanisms.Citation: https://doi.org/10.5194/egusphere-2025-4966-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 163 | 86 | 26 | 275 | 45 | 18 | 20 |
- HTML: 163
- PDF: 86
- XML: 26
- Total: 275
- Supplement: 45
- BibTeX: 18
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Ruijsch et al., present an in-depth evaluation of the ability of LPJmL-5 to simulate the impact of multi-year droughts on vegetation across the globe. Such an evaluation is much needed as the incidence of prolonged droughts is on the rise globally. The ability of DVMs to accurately simulate vegetation responses to drought is extremely important and global evaluation of this ability is largely lacking. The authors communicate their results well both through the text and the figures. In particular I would like to highlight Figure 4 as a I find the overlay of SPEI and GPP very informative.
That being said, I do unfortunately believe that this work requires some major improvements related to 1.) the identification of droughts using SPEI and 2.) the use of satellite-derived MODIS GPP.
1. SPEI is a widely used to characterize drought conditions particularly because it is standardized and as such always relates to the local climatic context (see Slette et al. 2019). However, this also poses some issues as outlined in Zang et al., (2020). The main argument of Zang et al., is that negative SPEI (e.g. -1 to -2; often characterized as moderate drought) does not necessarily always coincide with "actual" water shortage as determined by the amount of evapotranspiration subtracted from the precipitation. This seems to be the case especially in wet regions of the world such as the tropics or the boreal forest. The consequence is that SPEI may indicate drought conditions when in reality ample water is available (as determined for example by the Maximum Climatic Water Deficit commonly used in tropical regions).
To alleviate this issue which could potentially result in overestimation of droughts, I would suggest to use a second (potentially non-standardized) drought indicator to ensure that droughts are being accurately captured.
Please see Zang et al., (2020) for more detail: https://doi.org/10.1111/gcb.14809
2. The second issue is the use of MODIS GPP as a stand-in for "observed GPP". Satellite-derived GPP is as much a model as it is based on satellite observations. In the case of MODIS GPP, GPP is calculated by an algorithm using remotely sensed FPAR, a land cover classification, a parameter for the conversion efficiency of PAR, and some climate inputs (Tmean, VPD). Consequently, it is not really accurate to consider satellite-derived GPP from MODIS as "observed". This obviously does not invalidate the comparison with simulated GPP from LPJmL-5, however, it does warrant some further discussion and potentially the use of an alternative product (e.g. SIF) to strengthen the message of this study.
Some concrete issues that need to be addressed in regards to the use of MODIS GPP are:
1. How does the landcover type used by MODIS GPP differ from the landcover type used by LPJmL-5 on a gridcell basis? Could discrepancies between the two landcover products be the cause of some of the mismatch seen between modeled and satellite-derived GPP (e.g. the relatively poor performance of crops)?
2. Both LPJmL and MODIS GPP rely on climate inputs. To what degree does this pose an issue if both products model GPP, at least partially, on the same or similar inputs? Related to this, how would a mismatch between the climate inputs for LPJmL and MODIS affect results? For example, the climate data used for LPJmL indicates a drought in a given gridcell in a given time-period but the MODIS climate input does not? Will this skew results?
3. Yang et al., (2022) highlighted a divergence between DVM simulated GPP and satellite-derived GPP, especially in tropical regions of the southern hemisphere. They identified the uncertainty in tropical LAI data to be a major contributing factor. In particular, they highlighted that: "NOAA satellite orbit changes and MODIS sensor degradation might cause long-term satellite-derived LAI products inconsistent with each other. Xie et al. (2019) also suggested that satellite-derived LAI datasets can cause uncertainties in GPP estimations through model structure and the complexity of the terrain".
As stated before, I do not think this entirely invalidates the results of this study. Rather, I believe this study would be greatly strengthened by 1.) including a secondary remote sensing GPP proxy such as e.g. SIF and 2.) a thorough discussion of the limitations of satellite-derived GPP in the discussion section of this study.