the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Apr 2026
Spurious seasonality of Earth observation LAI across three northern evergreen needleleaf forests: Implications for analyses of the carbon cycle
Abstract. Leaf area index (LAI) is a key biophysical variable which quantifies the surface area for light capture and photosynthetic activity per unit ground area, giving a first order constraint on potential photosynthesis. LAI is tightly coupled to the carbon, energy, and water cycles of the global terrestrial system. Numerous Earth observation (EO) products provide estimates of LAI over time (LAIEO), delivering valuable information on plant phenology and canopy dynamics. Widely-used LAIEO, however, consistently exhibit unrealistic seasonality in evergreen needleleaf forests at the northern latitudes. Taking a model-data fusion approach, we show that naïvely assimilating biased, whole-year LAIEO (i.e., the business-as-usual (BAU) approach) at three well-studied evergreen needleleaf forests in Fennoscandia implies an ecosystem carbon cycle which is unrealistic and inconsistent with independent lines of evidence. We further demonstrate that the model-data fusion framework, CARDAMOM, is capable of diagnosing realistic seasonal amplitudes of LAI by assimilating localised information on leaf lifespan coupled with summer-only LAIEO (i.e., the alternative (ALT) approach). Important differences arise from the BAU and ALT experiments. The BAU experiment showed highly seasonal canopy dynamics and diagnostic leaf traits erroneously consistent with deciduous species. Conversely, the ALT experiment displayed canopy dynamics and functional characteristics more reflective of evergreen needleleaf species. For BAU, biases in LAIEO propagated throughout the carbon cycle, especially in the southern, more productive sites. This investigation highlights the need for improved LAIEO estimates in northern evergreen forests to enhance understanding of carbon cycle processes in this region of rapid warming and large carbon stores, and provides a mechanism for improvement using independent leaf trait data.
- Preprint
(1774 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 05 Jun 2026)
-
RC1: 'Comment on egusphere-2026-2222', Anonymous Referee #1, 25 Apr 2026
reply
-
RC3: 'Reply on RC1 about the MODIS ENF LAI in January', Hongliang Fang, 15 May 2026
reply
I happened to have noticed RC1’s comment for Line 52-54 “The Fang et al (2021) analysis is spurious. If you look at Yan et al 2016, their figure 4c shows the seasonality of ENF (their B7). It shows that the seasonal minimum is always above 1 LAI. It is perhaps that Fang et al (2021) did not strictly filtering low quality data based on the satellite QA.”
I would like to thank RC1’s attention to our work. There are some clarifications below:
1. In Fang et al. (2021), we did apply satellite QA and only used good quality retrievals from the MODIS (MCD15A2H C6) main algorithm (section 4.4 there). Note that we calculated all valid pixels including those of zero LAI for the ENF (Table 2 there). This probably leads to the relatively lower (but valid) ENF LAI in January.
2. Since Figure 4c in Yan et al (2016) is not about LAI, I did find that the ENF LAI in their Figure 7g for a particular site is always > 1.0. Note that Fang et al (2021)’s global average ENF LAI is at a pixel level. The difference between the higher site LAI and lower pixel LAI is understandable because of the scale difference.
3. Our earlier paper (Fang et al., 2013) showed that MODIS (MCD15A2 C5) ENF LAI (2003-2010) in January is around 0.80 (Fig. 2g there). Similarly, only MODIS pixels retrieved from the main algorithms were considered in the statistics. The slightly higher ENF LAI is possible because we calculated only positive (LAI > 0) pixels then (noted in Table 3).
Fang et al., 2013. Characterization and intercomparison of global moderate resolution leaf area index (LAI) products: Analysis of climatologies and theoretical uncertainties. Journal of Geophysical Research – Biogeosciences, 118(2): 529-548, https://doi.org/10.1002/jgrg.20051.
4. The ENF LAI in winter from Fang et al. (2021) is corroborated by other more recent statistics. For example, Zhang et al. (2024) showed that in DJF, the ENF LAI from the MODIS C6.1 main algorithm is around 0.50 (Fig. 4 there). Indeed, lower winter ENF LAI (~0.3) is observed from the backup algorithm (Fig. 4 there).
Zhang et al., 2024. An Insight Into the Internal Consistency of MODIS Global Leaf Area Index Products. IEEE Transactions on Geoscience and Remote Sensing, 62: 1-16. https://doi.org/10.1109/TGRS.2024.3434366.
5. Yan et al (2024) showed that the MODIS C6.1 ENF LAI could be <0.5 in winter although the field measured LAI is around 1.0 (the TUMB ENF site around 2019-2021 in Fig. S1 there). This lower ENF LAI is probably related to that both main and backup retrievals were included in the calculation. Their high-quality reprocessed HiQ-LAI has improved the winter ENF LAI to around 1.5, slightly higher than the field measurements (Fig. S1 there).
Yan et al., 2024. HiQ-LAI: a high-quality reprocessed MODIS leaf area index dataset with better spatiotemporal consistency from 2000 to 2022. Earth System Science Data, 16(3): 1601-1622. https://doi.org/10.5194/essd-16-1601-2024.
Sincerely,
Hongliang Fang
Citation: https://doi.org/10.5194/egusphere-2026-2222-RC3 -
RC4: 'Reply on RC3', Anonymous Referee #1, 15 May 2026
reply
Fang's comment: 2. Since Figure 4c in Yan et al (2016) is not about LAI, I did find that the ENF LAI in their Figure 7g for a particular site is always > 1.0. Note that Fang et al (2021)’s global average ENF LAI is at a pixel level. The difference between the higher site LAI and lower pixel LAI is understandable because of the scale difference.
Response: Thank you for pointing this out. I would like to clarify that there are two Yan et al. (2016) papers: Part 1 and Part 2. I mistakenly cited Part 2 in the reference list, but the intended citation should be Part 1. In Part 1, Figure 4c, the minimum value of ENF B7 in both C5 and C6 are around 1 LAI.
Yan, K., Park, T., Yan, G., Chen, C., Yang, B., Liu, Z., Nemani, R., Knyazikhin, Y., & Myneni, R. (2016). Evaluation of MODIS LAI/FPAR Product Collection 6. Part 1: Consistency and Improvements. Remote Sensing, 8(5), 359–16. https://doi.org/10.3390/rs8050359
I cannot verify exactly which paper has problematic values, or which paper may have issues in data processing, or perhaps all of them are correct but with different processing details. The Yan et al. paper is listed under the CEOS LPV Validation Hierarchy for MODIS Land products on the NASA MODIS Land website. Since it appears to be the most recent official reference there, I used it as the benchmark: https://modis-land.gsfc.nasa.gov/ValStatus.php?ProductID=MOD15.
When filtering MODIS products (such as LAI), it is important to use not only the algorithm path (main or back up) but also the State QA layer, which provides information on cloud contamination, snow cover, and other surface/atmospheric conditions.
Citation: https://doi.org/10.5194/egusphere-2026-2222-RC4
-
RC4: 'Reply on RC3', Anonymous Referee #1, 15 May 2026
reply
-
AC1: 'Reply on RC1', Timothy Green, 15 May 2026
reply
I find the overall motivation of the manuscript somewhat unclear. The authors argue that EO-derived LAI exhibits biased seasonality in evergreen needleleaf forests (ENFs). Based on this premise, I would expect the study to clearly demonstrate how such biases propagate into carbon fluxes, and how the proposed method improves those estimates.
The motivation for the manuscript is to investigate how assimilation of EO-derived LAI (LAIEO) time series with seasonal biases (that have been demonstrated by several previous studies) impacts our understanding of carbon cycling in northern evergreen needleleaf forest (ENF) ecosystems (BAU scenario) and to test whether it is possible to obtain more ecologically representative carbon cycle diagnostics through assimilation of summer-only LAIEO estimates alongside information on leaf life span for the principal tree species present (ALT scenario). These objectives are clearly reflected in the research questions specified in the introduction.
When the full time series of LAIEO are assimilated, the seasonal biases from LAIEO propagated not only in canopy carbon dynamics but throughout the carbon cycle. This is clearly demonstrated in our diagnostic retrieval of ecosystem carbon cycling for which the dynamics of the stocks and internal fluxes of BAU and ALT are very different in both their magnitude and seasonality.
In Section 3.2, we present the impact of the BAU scenario, for which we assimilated the entire LAIEO time series with its seasonal biases, on the canopy carbon dynamics, compared to our ALT approach, with clear impacts on the magnitude and seasonality of the canopy production and turnover (Fig. 3). We compared the results against independent estimates of canopy turnover from foliage litter traps at Fl-Hyy, finding the ALT scenario to provide much closer agreement.
In Section 3.3 and Fig. 4, we demonstrate that the impacts propagate beyond canopy carbon dynamics, showing there are fundamental impacts on the retrieved model parameterisation. The strong seasonality in LAIEO can only be achieved if leaves have a short lifespan (< 1yr i.e. deciduous). This is inconsistent with what we know about leaf lifespan for Norway spruce (5-13 yrs) and Scots pine (2-5.5 years) needles in Finland (Tupek et al., 2015). We also document impacts on retrievals of Potential Photosynthetic Rates (PPR), Leaf Mass per Unit Area (LMA), and fractional allocation of NPP between different plant tissues. Retrieved estimates of LMA were compared to independent estimates from the sites. Independent LMA was closer to median retrieved ALT LMA at two of the three sites (FI-Var and SE-Htm) and within the 95CI of ALT LMA at FI-Hyy.
In Section 3.5 we present mean annual carbon budgets for the two experiments across all three sites, showing the impact of the different data assimilation scenarios on all (time-averaged) carbon stocks and fluxes simulated by our ecosystem model. This analysis shows clear differences (quantified by the Hellinger distance metric) in the estimates of internal carbon cycling between ALT and BAU, with the number of differences among the sites aligning with the degree of seasonality in LAIEO quantified by absolute seasonal amplitude.
Moreover, the seasonality indicated by LAIEO (BAU) could only be achieved with deciduous-like carbon cycle dynamics and canopy traits. Assimilating leaf lifespan and summer LAI (ALT; i.e. the proposed method) resulted in carbon cycle dynamics and canopy traits that are consistent with our ecological understanding of northern ENFs, and consistent with the available observations across the sites for validation.
More detailed discussion placing these results in the context of the relevant literature is provided in:
- Section 4.1: Assimilated leaf lifespan generates realistic LAI seasonality
- Section 4.2: Earth observation LAI biases propagate into diagnosed functional characteristics and canopy carbon dynamics
- Section 4.3: Canopy biases from LAIEO propagate to whole-system carbon dynamics
Tupek, B., Mäkipää, R., Heikkinen, J., Peltoniemi, M., Ukonmaanaho, L., Hokkanen, T., ... & Lehtonen, A. (2015). Foliar turnover rates in Finland-comparing estimates from needle-cohort and litterfall-biomass methods. Boreal Environment Research, 20, 283–304
However, based on Fig. 3, while LAI may show seasonal biases (though this itself depends on resolving issues such as leaf-wood separation in field measurements), the differences in key carbon fluxes (NEE, GPP, and Reco) between the BAU approach (direct use of satellite LAI) and the authors’ alternative method (ALT) appear minimal. This raises the question of whether LAI seasonal bias meaningfully affects the simulated fluxes in practice.
We believe the reviewer is referring to Fig. 5, which, along with section 3.4, shows that BAU and ALT experiments produce nearly indistinguishable biosphere-atmosphere fluxes (NEE, GPP, Reco). This is not surprising, as NEE was assimilated in both experiments. However, while both scenarios provided similar model fits to the biosphere-atmosphere fluxes, only the ALT scenario (assimilating leaf lifespan, summer-only LAIEO) produced a diagnostic carbon cycle that was consistent with evergreen needleleaf forests.
The fact that similar NEE estimates are produced with significantly different underlying carbon cycle dynamics is an interesting result, and presents an important challenge that we describe in Section 4.4 ‘Divergent functional characteristics and internal carbon cycling yield indistinguishable biosphere-atmosphere fluxes’ where we state:
“sets of ecologically implausible, deciduous-like characteristics inferred from LAIEO and sets of more realistic evergreen characteristics were shown to be equally capable of generating acceptable agreement with eddy covariance observations. This could suggest validation approaches that rely solely on LAIEO and flux-based observations in northern evergreen needleleaf forests risk drawing flawed ecological interpretations and attributing false confidence in model performance… The incorporation of LAIEO in its current state can lead to correct model NEE outputs for incorrect reasons.”
Moreover, the alternative approach is relatively complex and heavily parameterized, yet the resulting improvements in fluxes are not clearly demonstrated. As such, it remains unclear whether the added model complexity is justified by a substantial gain in predictive performance.
We argue against the reviewer’s assertion that the alternative approach (ALT) is complex and heavily parameterised. Both BAU and ALT experiments were undertaken with the same model structure and number of parameters (Table A1). No new parameters were introduced in the ALT method, and model complexity is identical between the two approaches. The key difference between experiments is outlined in section 2.2 (lines 163-170) and summarised in Table 1. The BAU experiment assimilated the entire (biased) seasonal cycle of CGLS 300m LAI, whereas the ALT experiment assimilated CGLS 300m LAI in the summer period only along with an observation of leaf lifespan. As described in section 2.4.6, the site-specific leaf lifespan observation is formulated as a simple weighted average of species-specific leaf lifespans (which are specific to each location) weighted by their contribution to total basal area at the site. The differences in input data between the experiments impact the parameter calibration via the likelihood function (Eq. 2).
The reviewer is correct in stating that both experiments produced nearly indistinguishable biosphere-atmosphere fluxes (NEE, GPP, Reco) as described in section 3.4, presented in Fig. 5 and discussed in Section 4.4. As discussed above, the similar fits to NEE, GPP, Reco were produced with very different underlying carbon cycle dynamics, and only the ALT scenario (assimilating leaf lifespan, summer-only LAIEO) produced a diagnostic carbon cycle that was consistent with evergreen needleleaf forests.
Major comments:
The abstract (Lines 10–16) is somewhat confusing. It is unclear what is meant by “seasonal canopy dynamics”. Does this refer to LAI, carbon fluxes, or other processes? While LAI seasonality is known to be uncertain in ENF biomes, the extent to which these biases propagate to canopy processes remains debatable. For example, based on Fig. 3, the differences in NEE, GPP, and Reco between the BAU (direct use of satellite LAI) and ALT (the proposed method) simulations appear relatively small. This raises the question of whether the impact of LAI seasonality bias on ecosystem fluxes is important.
We appreciate the reviewer pointing out the ambiguity in the language of the abstract. Our phrasing is indeed unclear and will be corrected to “seasonal canopy carbon dynamics.”
We believe the reviewer is referring to Fig. 5 of the manuscript which shows simulated LAI, NEE, GPP, and Reco across the two experiments (BAU and ALT) as well as the observational data. As mentioned in the above responses, Fig. 5 illustrates that sets of ecologically implausible deciduous-like functional characteristics generated by BAU (Fig. 4) are equally capable of producing good agreement with biosphere-atmosphere flux estimates from eddy covariance compared to those generated from an ecologically more realistic set of traits (ALT). We note again that NEE was assimilated in both experiments, so the model was calibrated in both cases to reproduce the observed seasonality in NEE. However, it is clear that the dynamics of land-atmosphere exchange simulated in the two experiments are driven by very different underlying canopy carbon dynamics, necessitating distinct sets of functional characteristics (i.e., parameters). In the BAU experiment, the strong seasonality in LAI necessitates a higher dependence on labile carbon inputs and greater carbon inflows and outflows through the canopy compared to ALT (Fig. 3). Sections 3.3 and 3.5, and Fig. 6 show how the biases from LAIEO propagate throughout the carbon cycle and lead to very different internal carbon dynamics, especially at the southern sites where the absolute seasonal amplitudes in the LAIEO signal were highest.
I think the accuracy of LAI should be evaluated in the context of the specific application. If the goal is to simulate carbon or water fluxes, LAI accuracy may not be critically important (see your Fig 3). For example, EO-derived LAI is known to have limitations, including saturation effects, differences between effective and true LAI (e.g., due to canopy clumping), and biases under snow cover. However, these issues may not strongly affect estimates of fAPAR, which are typically retrieved simultaneously from satellite reflectance and thus more directly reflect the canopy radiative transfer state. Since simulations of carbon fluxes primarily depend on APAR (APAR = fAPAR × PAR), uncertainties in LAI may not substantially propagate into flux estimates. Or it may depend on the upscaling scheme of the model itself (some models depend on using LAI to do the scaling, some using the fAPAR).
Again, we believe the reviewer is referring to Fig. 5. fAPAR was not the focus of this study. In this study we show how assimilating the seasonally biased EO-derived LAI impacts the resultant diagnostic assessment of the carbon cycle. We again point out Figs. 3, 4, and 6 that demonstrate how these seasonally biased LAI time series propagate to impact the seasonal canopy carbon dynamics, retrievals of ecosystem functional characteristics, and internal carbon cycling. Fig. 5 demonstrates how assessing model outputs solely against net and component biosphere-atmosphere fluxes risks falsely attributing confidence to a modelling framework. We show it is possible to get the right NEE for the wrong reason. We argue that understanding how carbon flows within a given ecosystem, not just into and out of an ecosystem, is indeed critically important.
As mentioned in the first paragraph of the introduction, LAI is considered an essential climate variable by the Global Climate Observing System and is widely used across a range of applications, including the benchmarking of dynamic global vegetation models (Sitch et al., 2024). In such benchmarking studies, it is not necessarily only the biosphere-atmosphere fluxes that are being assessed, but also the ability of the model representations of the underpinning processes (e.g. phenology, NPP allocation etc) to reproduce the observed LAI dynamics. Thus, for models or other studies incorporating and/or benchmarking against LAIEO, these biases may have profound impacts on our understanding of ENF carbon cycling.
Again, while fAPAR was not the focus of this study, we refer to the work done by Zhang et al (2025) who showed EO-derived fAPAR (fAPAREO) exhibits spurious seasonality in ENFs when compared against in situ and upscaled reference fAPAR maps. fAPAREO seasonality mirrored that of LAIEO shown here, with very low values outside of the summer period compared to the in situ and reference maps (see Figs. 3, 4, 7, 8 within).
Sitch, S., O’sullivan, M., Robertson, E., Friedlingstein, P., Albergel, C., Anthoni, P., ... & Zaehle, S. (2024). Trends and drivers of terrestrial sources and sinks of carbon dioxide: An overview of the TRENDY project. Global Biogeochemical Cycles, 38(7), e2024GB008102.
Zhang, Y., Fang, H., Hu, Z., Wang, Y., Li, S., & Wu, G. (2025). Validation of Global Moderate-Resolution FAPAR Products over Boreal Forests in North America Using Harmonized Landsat and Sentinel-2 Data. Remote Sensing, 17(15) 2658.
Instead, greater importance may lie in the biogeochemical properties of the leaves, such as light-harvesting pigment content and efficiency, Rubisco content and activity, and stomatal regulation. In Fig. 3, all simulated NEE are similar across experiments but differ from the eddy covariance (EC) observations. This suggests that improving the representation of biogeochemical processes in the model may be more important than refining LAI seasonality alone.
Again, we believe the reviewer is referring to Fig. 5 not Fig. 3. At FI-Var and FI-Hyy diagnostic NEE (ALT, BAU) is very much in line with the EC observations (RMSE: 0.36 – 0.60 gC m-2day-1; diagnostic-observation overlap: 93 – 99%). At SE-Htm, EC NEE exhibited sudden shifts from week to week in the growing season. These sudden jumps became more pronounced in the latter two years of the study period, and both BAU and ALT were unable to replicate the high degree of temporal variability. This may indicate a missing process representation in the model. Regarding the biogeochemical properties of the leaves mentioned by the reviewer, we recognise the importance of these processes. The potential photosynthesis rate (PPR) parameter embeds much of this information into a single calibrated parameter that was shown to differ between ALT and BAU (Fig. 4). However, we again reiterate there are important differences between BAU and ALT (other than biosphere-atmosphere fluxes that arise due to LAIEO biases) in the internal carbon dynamics as described in section 3.5, shown in Fig. 6 and discussed in sections 4.2 and 4.3.
Detailed comments:
~ Line 15: “This investigation highlights the need for improved EO LAI estimates in northern evergreen forests…”. I agree that there are biases in EO-derived LAI seasonality in northern evergreen forests. However, the authors should be more specific about how such improvements could be achieved. If the bias primarily arises from snow obscuring satellite reflectance, then it may be fundamentally difficult to resolve within a purely satellite-based framework. On the other hand, if improvement requires incorporating more complex modeling or data assimilation approaches, this should be clearly stated, as it moves beyond a strictly observation-based (EO-only) methodology.
We make no suggestions as to how the EO community should go about improving these products or what the primary cause of the biases are as this would be outside the remit of our expertise. Rather we are pointing out that these biases continue to persist and have the potential to significantly impact northern ENF analyses/studies which naively incorporate whole-year LAIEO. Within a model-data fusion context, we leverage the information content of LAIEO in the summer period, which we assume to be unbiased (validation of LAIEO most often occur with upscaled field-based LAI estimates taken during the peak growing season), to provide constraint on annual maximum LAI. We additionally assimilate location-specific leaf trait data (leaf lifespan) to constrain turnover rates of canopy carbon, ultimately improving the seasonality of diagnostic LAI. This approach is clearly stated throughout the text of the manuscript.
~ Line 40: “LAI EO estimates are typically derived from surface or top-of-atmosphere reflectances measured by passive optical sensors onboard satellites.” This statement is somewhat confusing and depends on how “EO” is defined. If EO refers to satellite-based observations, then all satellite measurements originate from TOA reflectance. However, many LAI products (e.g., MODIS LAI) are actually derived from surface reflectance, which is obtained after atmospheric correction of TOA measurements through upstream processing. The authors should clarify this distinction and use more precise terminology.
We appreciate the reviewer pointing out the imprecise language and will make appropriate changes to ensure the text is more technically correct.
~ Line 50: The cited studies (e.g., Cohen et al., 2006; Tian et al., 2004) may be outdated. For example, MODIS LAI products have evolved substantially (currently at Collection 6), whereas these cited were based on much earlier versions (e.g., Collection 1 or 3). There might have been improvements between these versions. The authors should consider citing more recent literature.
The paragraph beginning on line 48 which includes references Cohen et al., 2006 and Tian et al., 2004 was meant to illustrate that seasonal biases in EO-derived LAI over ENFs have been identified and known for over 20 years, yet no improvements have been made in the subsequent years up to the present. As the reviewer has pointed out, MODIS LAI products have evolved over the previous decades, however, as shown in Fig.1 of this study and by Fang et al. (2021) on a global scale, the most recent version of MODIS LAI (Collection 6) continues to exhibit spurious seasonality.
Reference: Yan, K., Park, T., Yan, G., Liu, Z., Yang, B., Chen, C., Nemani, R., Knyazikhin, Y., & Myneni, R. (2016). Evaluation of MODIS LAI/FPAR Product Collection 6. Part 2: Validation and Intercomparison. Remote Sensing, 8(6), 460–26. https://doi.org/10.3390/rs8060460
Line 52-54: The Fang et al (2021) analysis is spurious. If you look at Yan et al 2016, their figure 4c shows the seasonality of ENF (their B7). It shows that the seasonal minimum is always above 1 LAI. It is perhaps that Fang et al (2021) did not strictly filtering low quality data based on the satellite QA.
We are unclear as to why the reviewer has asserted that the study by Fang et al. (2021) is spurious. As stated in section 4.4 of Fang et al. (2021), “the qualitative quality control layers were considered to select only the best quality data in the analysis. For the CI and LAI data, only those from the full and main retrieval algorithms, respectively, were considered…” . In the reviewer’s suggested reference, (Yan et al., 2016 (part1 not part2)) the Figs. 4a (MODIS Collection 5) and 4c (MODIS Collection 6) suggest MODIS LAI does indeed dip below 1 m2m-2 for ENF around DOY 25 and DOY 350. One year of MODIS data (2003) is used to generate these figures. The monthly average LAI for global ENF reported in Fang et al. (2021) covers a much longer time period (2003 – 2017) and so is not directly comparable to Yan et al. (2016).
~ Line 80: The manuscript compares two experiments (BAU and ALT), but it is unclear whether the results have been evaluated against other independent benchmarks. Have the authors compared their simulated GPP and ET with products such as MODIS GPP/ET, or with outputs from satellite-driven terrestrial biosphere models (such as BESS, BEPS)?
Eddy covariance-based estimates of GPP and Reco were not assimilated in either BAU or ALT experiments and instead used for evaluation as outlined in section 2.4.3 lines 257-258. Additionally, as stated in section 2.1 lines 110-111, “observations of LMA and foliage litter production were left out of the assimilation to be used for independent model evaluation.” We did not compare simulated GPP with the MODIS product or with other terrestrial biosphere models. The three sites in this study were chosen due to the richness of available data including eddy covariance estimates of biosphere-atmosphere fluxes, widely considered the gold standard measurement for CO2 fluxes between atmosphere and biosphere at the scale this analysis was conducted.
~ 90: For your hypothesis, is it possible to implement the leaf life span into large scale studies? It seems the leaf life span is based on some strong assumptions, and how to reflect the geographical differences and within species plasticity?
It is indeed possible to implement leaf lifespan into large scale studies. Across Finland, for example, there are published maps of the number of needle cohorts (a proxy for leaf lifespan) for the dominant species (Norway spruce, Scots pine) (Tupek et al., 2015) that could readily be integrated with our suggested approach.
It is unclear what is meant by the leaf lifespan being based on strong assumptions. Here, the leaf lifespan observations used in the ALT experiment are a simple weighted mean of species-specific leaf lifespans extracted for each site location, weighted by their contribution to total basal area at each site. A single effective leaf lifespan value is needed to align with the DALEC model which represents the entire canopy as one pool. As pointed out by the reviewer, within-species leaf lifespan does vary considerably across space driven by phenotypic plasticity (Reich et al., 1996, Reich et al., 2014). For example, in Finland alone leaf lifespan of Norway spruce ranges from approximately 5 years in the south to 13 years in the north (Tupek et al., 2015). As detailed in section 2.4.6, we account for this by extracting the location and species-specific values from needle cohort maps generated by Tupek et al. (2015) at the Finnish sites. At the Swedish SE-Htm site, leaf lifespan for Norway spruce was based on a needle cohort survey performed in 2023.
Reich, P. B., Oleksyn, J., Modrzynski, J., & Tjoelker, M. G. (1996). Evidence that longer needle retention of spruce and pine populations at high elevations and high latitudes is largely a phenotypic response. Tree Physiology, 16(7), 643-647.
Reich, P. B., Rich, R. L., Lu, X., Wang, Y. P., & Oleksyn, J. (2014). Biogeographic variation in evergreen conifer needle longevity and impacts on boreal forest carbon cycle projections. Proceedings of the National Academy of Sciences, 111(38), 13703-13708
Tupek, B., Mäkipää, R., Heikkinen, J., Peltoniemi, M., Ukonmaanaho, L., Hokkanen, T., ... & Lehtonen, A. (2015). Foliar turnover rates in Finland-comparing estimates from needle-cohort and litterfall-biomass methods. Boreal Environment Research, 20, 283–304
Fig. 1: Field LAI is derived from DHP measurements. How do the authors account for the leaf-to-wood ratio in this conversion? Is this ratio assumed to be constant throughout the year? If the DHP measurements represent plant area index (PAI), they may exhibit relatively weak seasonality, as suggested by Fig. 1. In that case, separating leaf area from woody components becomes critical, and the assumed leaf-to-wood ratio introduces additional uncertainty that should be discussed or quantified. From my experience, the SE-Htm panel shows many extremely low values and abrupt changes in MCD15A2H LAI. This is likely due to incorrect use of the QA flags, or possibly no QA filtering at all. You should exclude all back-up algorithm retrievals as that likely means cloud or snow presents and the reflectance is corrupted (also possible due to sun-sensor poor geometry, but it should be more like to happen during the winter seasons, while your spurious data spread all over the year).
In section 2.4.2 we explicitly mention the following with regards to leaf-to-wood ratios:
“The field-based LAI data used here are more accurately described as plant area index (PAI) due to the influence of woody components in estimation. Site-specific values of woody area index (WAI) are needed to convert PAI to LAI, which were unavailable at the sites in this analysis. WAI varies considerably both across and within biomes (Fang et al., 2019). Lacking this information, we neglected a PAI adjustment. However, the impact on seasonal trajectories is most likely minimal given the assumption that the surface area of woody components does not change considerably within a given year (Chen, 1996).”
As pointed out by Brown et al (2020), numerous EO product evaluation studies have lacked site-specific information on WAI and have used PAI as a proxy for LAI (e.g., Camacho et al., 2013; De Kauwe et al., 2011; Heiskanen et al., 2012; Yin et al., 2017).
MODIS LAI (MCD15A2H) was included only in Fig. 1 to illustrate that seasonal biases are not isolated to any one product. CGLS 300m LAI was assimilated in the BAU and ALT experiments. For Fig. 1, we filtered MCD15A2H based on QA flags, accepting only those values from the main algorithm and rejecting those from the backup algorithm. We thank the reviewer for pointing out that we did not specify the filtering and will add a statement in the caption of the figure. Despite the filtering, MCD15A2H still exhibits unrealistic jumps from one timestep to the next at the pixel level. For this reason, we state in section 2.4.2 “Product processing [of CGLS 300m] also includes a temporal smoothing procedure, providing estimates that are more conducive for site-level analyses, contrary to the noisy time series of the MODIS LAI product at pixel scale (Brown et al., 2020).”
Brown, L. A., Meier, C., Morris, H., Pastor-Guzman, J., Bai, G., Lerebourg, C., ... & Dash, J. (2020). Evaluation of global leaf area index and fraction of absorbed photosynthetically active radiation products over North America using Copernicus Ground Based Observations for Validation data. Remote Sensing of Environment, 247, 111935.
Camacho, F., Cernicharo, J., Lacaze, R., Baret, F., & Weiss, M. (2013). GEOV1: LAI, FAPAR essential climate variables and FCOVER global time series capitalizing over existing products. Part 2: Validation and intercomparison with reference products. Remote Sensing of Environment, 137, 310-329.
De Kauwe, M. G., Disney, M. I., Quaife, T., Lewis, P., & Williams, M. (2011). An assessment of the MODIS collection 5 leaf area index product for a region of mixed coniferous forest. Remote Sensing of Environment, 115(2), 767-780.
Heiskanen, J., Rautiainen, M., Stenberg, P., Mõttus, M., Vesanto, V. H., Korhonen, L., & Majasalmi, T. (2012). Seasonal variation in MODIS LAI for a boreal forest area in Finland. Remote Sensing of Environment, 126, 104-115.
Yin, G., Li, A., Jin, H., Zhao, W., Bian, J., Qu, Y., ... & Xu, B. (2017). Derivation of temporally continuous LAI reference maps through combining the LAINet observation system with CACAO. Agricultural and Forest Meteorology, 233, 209-221.
~Lines 123-125: (1) rooting depth is an additional uncertainty. (2) Is CUE temporally constant?
Maximum rooting depth is a calibrated parameter; therefore, its uncertainty is explicitly accounted for. CUE is also calibrated and is assumed to be time invariant.
~ Line 129: It appears that other variables (not LAI) with strong seasonal dynamics may be driving much of the variability in the simulated carbon fluxes, LAI and other related outputs, ultimately contributing to the improved seasonality of the results. Could the authors clarify which variables are primarily responsible for this seasonal control? For example, are these driven by meteorological forcings (e.g., radiation, temperature, VPD) or by internal model parameters/processes (stomata, quantum yield efficiency)?
The sites in this study are all northern Fennoscandian forests which are primarily temperature and light limited systems which mainly controls the seasonality of the biosphere-atmosphere fluxes. This is a key reason why different ecological strategies/canopy dynamics (i.e. BAU, ALT) can equally fit the seasonal cycle of NEE/GPP/Reco; however, only the evergreen representation (ALT) is consistent with our understanding of the overall system (functional characteristics and internal carbon cycling)
L315: should LMA be a dynamic variable, spatially and temporally?
LMA is calibrated at each site independently for each of the two experiments (BAU and ALT) and is therefore spatially dynamic. We agree real-word LMA for a forest stand may vary temporally; however, our model does not include temporal variation of LMA. We do not anticipate that such temporal variation would significantly impact our results.
Citation: https://doi.org/10.5194/egusphere-2026-2222-AC1
-
RC3: 'Reply on RC1 about the MODIS ENF LAI in January', Hongliang Fang, 15 May 2026
reply
-
RC2: 'Comment on egusphere-2026-2222', Anonymous Referee #2, 13 May 2026
reply
Leaf area index (LAI) products derived from Earth observation (LAI_EO) are known to show unrealistic seasonal amplitudes over northern evergreen needleleaf forests. Directly incorporating LAI_EO to diagnose ecological traits and carbon dynamics would lead to erroneous results. To correct these biases, this study explored the usage of leaf lifespan information in a data assimilation scheme. The study is finely designed and the manuscript well organized. I have a few small comments for authors to consider in their revision.
I feel the title is a bit eyepopping. Spurious seasonality of LAI_EO is already known. May consider “Assimilating leaf life span and summer Earth observation LAI for improved carbon cycle analysis across three northern evergreen needleleaf forests”.
Fig. 1. I’d suggest authors to show some field pictures of the three sites. It would be much helpful for readers to understand the area, especially about the understory condition.
L140-145. While the parameter vectors are listed in Table A1, what are the state vectors? Do you mean GPP, Reco, NEE, and LMA (L190)? Suggest to clarify them in their first appearance.
L140-147. I feel the “observations” here are somewhat confusing, because both parameter and state vectors can have observational values. Need to be more specific in the context.
L165. From Fig. 1, we can see that LAI_EO values are unrealistically lower than field measurements in the FI-Hyy and SE-Htm sites, but we don’t know about the FI-Var site. Strictly speaking, the “seasonal bias” is only clear for the first two sites.
L185-188. What’s the understory of the three sites? Do they all belong to the deciduous species? Moreover, the <10% of total basal area does not mean <10% of deciduous LAI in the total site LAI.
2.5.2 Instead of using the Hellinger distance, some other metrics, such as absolute bias, coefficient of determination, and RMSE are often used in other similar data assimilation studies. I wonder whether the selection of difference metrics would impact the final results. This might be briefly discussed.
Fig. 6 is not very clear. I have to zoom in by 150% to see the numbers.
Citation: https://doi.org/10.5194/egusphere-2026-2222-RC2 -
AC2: 'Reply on RC2', Timothy Green, 03 Jun 2026
reply
Leaf area index (LAI) products derived from Earth observation (LAI_EO) are known to show unrealistic seasonal amplitudes over northern evergreen needleleaf forests. Directly incorporating LAI_EO to diagnose ecological traits and carbon dynamics would lead to erroneous results. To correct these biases, this study explored the usage of leaf lifespan information in a data assimilation scheme. The study is finely designed and the manuscript well organized. I have a few small comments for authors to consider in their revision.
I feel the title is a bit eyepopping. Spurious seasonality of LAI_EO is already known. May consider “Assimilating leaf life span and summer Earth observation LAI for improved carbon cycle analysis across three northern evergreen needleleaf forests”.
We thank the reviewer for the title suggestion. The current title is intentionally provocative to draw attention to a long-standing issue that is often ignored in the land surface modelling community. Numerous studies continue to use LAIEO uncritically for benchmarking and evaluation (eg., Sitch et al., 2024), despite the known biases in northern evergreen needleleaf forests. We demonstrate this risks drawing flawed ecological interpretations and assessments of carbon cycling (BAU scenario) over this important domain that is experiencing rapid climate change. Hence, we prefer the title to focus on the problem (spurious LAIEO) rather than the proposed solution (assimilating leaf lifespan information) to highlight the need for a more careful consideration of LAIEO when used across a variety of applications.
Sitch, S., O’sullivan, M., Robertson, E., Friedlingstein, P., Albergel, C., Anthoni, P., ... & Zaehle, S. (2024). Trends and drivers of terrestrial sources and sinks of carbon dioxide: An overview of the TRENDY project. Global Biogeochemical Cycles, 38(7), e2024GB008102. https://doi.org/10.1029/2024GB008102
Fig. 1. I’d suggest authors to show some field pictures of the three sites. It would be much helpful for readers to understand the area, especially about the understory condition.
We do not possess photos of the sites; however, we will add references/links to publicly available photographs in the manuscript where they are available (e.g., https://meta.icos-cp.eu/labeling/ search “Hyltemossa” yields photos above and below the canopy surrounding the eddy covariance tower).
L140-145. While the parameter vectors are listed in Table A1, what are the state vectors? Do you mean GPP, Reco, NEE, and LMA (L190)? Suggest to clarify them in their first appearance.
We appreciate the use of the word “state” in the manuscript is somewhat confusing and thank the reviewer for bringing this to our attention. State variables refer to the temporally evolving simulated carbon and water pools (e.g., wood carbon pool, foliage carbon pool, litter carbon pool, surface soil moisture) and auxiliary variables such as LAI. Fluxes refer to the rate of transfer or flow of carbon/water between the pools or between the atmosphere and pools (e.g., GPP, heterotrophic respiration from the litter pool, foliar litter production). Parameters always refer to values being retrieved by CARDAMOM (Table A1), which constitute both process parameters and initial sizes of the carbon/water pools. We will make this distinction clearer in the revised manuscript.
L140-147. I feel the “observations” here are somewhat confusing, because both parameter and state vectors can have observational values. Need to be more specific in the context.
We will clarify this in the revised manuscript. For example, by extending the description of CARDAMOM (section 2.1.2) to “... CARDAMOM varies model parameters to minimise the mismatch between simulated model states, fluxes and/or diagnostic variables (e.g., leaf lifespan) and their concomitant observations, weighted by associated observational uncertainty.”
We will additionally add the following text beginning at line 150:There are therefore two possible routes through which observations may be used to constrain the calibration. The first is to compare the misfit of observations against their equivalent modelled properties. The second is as prior information on the model parameters. We note that the initial sizes of the carbon and water pools are treated as parameters in this context.
L165. From Fig. 1, we can see that LAI_EO values are unrealistically lower than field measurements in the FI-Hyy and SE-Htm sites, but we don’t know about the FI-Var site. Strictly speaking, the “seasonal bias” is only clear for the first two sites.
We acknowledge that we do not possess a timeseries of field-based LAI estimates at FI-Var to definitively show the LAIEO seasonal bias, rather one digital hemispheric photography-based value falling within the growing season of 2019. This is typical of most sites that contain any field-based LAI information. However, we do know the overstory at FI-Var is composed almost entirely of an evergreen needleleaf species (Scots pine) with an estimated leaf lifespan of over 5 years (derived from Tupek et al., 2015). Furthermore, according to the FI-Var ICOS station description document (https://meta.icos-cp.eu/files/2aYEi8pVB3x1ZqR50yMNuED4/FI-Var_Description.pdf), the understory is sparse and the ground/floor contains mosses, lichens, and low dwarf shrubs. Taken together, this implies CGLS 300m LAI values decrease to improbably low values in spring and autumn (approaching 0 m2m-2 in three of the four years of this study). This bias is further supported by our data assimilation experiment which indicates that inversion of the full seasonal CGLS LAI time series at this site results in a calibrated model with deciduous-like parameterised traits. We will add an explicit caveat to the manuscript outlining the above.
Tupek, B., Mäkipää, R., Heikkinen, J., Peltoniemi, M., Ukonmaanaho, L., Hokkanen, T., ... & Lehtonen, A. (2015). Foliar turnover rates in Finland-comparing estimates from needle-cohort and litterfall-biomass methods.
L185-188. What’s the understory of the three sites? Do they all belong to the deciduous species? Moreover, the <10% of total basal area does not mean <10% of deciduous LAI in the total site LAI.
At SE-Htm, the understory is sparse and the forest floor is mainly covered by Sphagnum moss (https://www.icos-cp.eu/sites/default/files/cmis/SE-Htm%20ICOS%20Ecosystem%20Station%20Labelling%20Report.pdf).
At FI-Var, the understory is sparse and the ground/floor contains mosses, lichens, and low dwarf shrubs (https://meta.icos-cp.eu/files/2aYEi8pVB3x1ZqR50yMNuED4/FI-Var_Description.pdf).
At FI-Hyy, the ground vegetation is composed of dwarf shrubs (Vaccinium myrtillus, Vaccinium vitis-idaea), feather moss (Pleurozium schreberi) and other bryophytes (https://meta.icos-cp.eu/objects/V1dIG1MK5wm_Hj5FUMmPD72T). As pointed out by Heiskanen et al., (2012), seasonal LAI variability of understory vegetation in the Hyytiälä study area is rather modest.
We will add summary understory information to Table 2 in the manuscript.
As the reviewer points out, it is true that <10% of overall basal area does not equate to <10% of deciduous LAI to total site LAI. Basal area was used as the weighting mechanism in the calculation of effective leaf lifespan at each site (Eq. 4). Lacking species-specific LAI information from the field datasets we chose basal area for the weighting mechanism, following studies that investigate community weighted mean traits in forest biomes (eg., Muscarella & Uruarte, 2016; Báez et al., 2022). If species-specific LAI was somehow available, the weights would vary over the course of the year with LAI seasonality. Basal area proportions are more stable and we assume carry lower uncertainty relative to other potential weighting mechanisms like foliage biomass derived from allometry as basal area is more closely tied to direct measurements (DBH measured in all trees of inventory plots).
Báez, S., Fadrique, B., Feeley, K., & Homeier, J. (2022). Changes in tree functional composition across topographic gradients and through time in a tropical montane forest. PLoS One, 17(4), e0263508.
Heiskanen, J., Rautiainen, M., Stenberg, P., Mõttus, M., Vesanto, V. H., Korhonen, L., & Majasalmi, T. (2012). Seasonal variation in MODIS LAI for a boreal forest area in Finland. Remote Sensing of Environment, 126, 104-115.
Muscarella, R., & Uriarte, M. (2016). Do community-weighted mean functional traits reflect optimal strategies?. Proceedings of the Royal Society B: Biological Sciences, 283(1827).
2.5.2 Instead of using the Hellinger distance, some other metrics, such as absolute bias, coefficient of determination, and RMSE are often used in other similar data assimilation studies. I wonder whether the selection of difference metrics would impact the final results. This might be briefly discussed.
The Hellinger distance was chosen as it is well-suited to the nature of CARDAMOM’s outputs, namely a 300-member ensemble for all parameters, diagnostic variables and simulated carbon pools and fluxes for each assimilation experiment (BAU vs ALT). This enables outputs to be expressed as probability distributions rather than single point estimates and thus explicitly accounts for uncertainty. Employing more traditional metrics such as RMSE, absolute bias or R2 would neglect these uncertainties thus limiting comparisons to single values (typically medians). We did, however, report RMSE of NEE/GPP/Reco from the two experiments by comparing quality-screened eddy covariance values against median ensemble members (see Fig. 5).
Fig. 6 is not very clear. I have to zoom in by 150% to see the numbers.
We thank the reviewer for pointing out the lack of clarity for Fig. 6. We will put the figure and Fig. A1 into a landscape orientation and increase the size of the labels to improve readability.
Citation: https://doi.org/10.5194/egusphere-2026-2222-AC2 -
RC5: 'Reply on AC2', Anonymous Referee #2, 04 Jun 2026
reply
I think my concerns were finely addressed.
Citation: https://doi.org/10.5194/egusphere-2026-2222-RC5
-
RC5: 'Reply on AC2', Anonymous Referee #2, 04 Jun 2026
reply
-
AC2: 'Reply on RC2', Timothy Green, 03 Jun 2026
reply
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 437 | 82 | 34 | 553 | 19 | 20 |
- HTML: 437
- PDF: 82
- XML: 34
- Total: 553
- BibTeX: 19
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
I find the overall motivation of the manuscript somewhat unclear. The authors argue that EO-derived LAI exhibits biased seasonality in evergreen needleleaf forests (ENFs). Based on this premise, I would expect the study to clearly demonstrate how such biases propagate into carbon fluxes, and how the proposed method improves those estimates.
However, based on Fig. 3, while LAI may show seasonal biases (though this itself depends on resolving issues such as leaf-wood separation in field measurements), the differences in key carbon fluxes (NEE, GPP, and Reco) between the BAU approach (direct use of satellite LAI) and the authors’ alternative method (ALT) appear minimal. This raises the question of whether LAI seasonal bias meaningfully affects the simulated fluxes in practice.
Moreover, the alternative approach is relatively complex and heavily parameterized, yet the resulting improvements in fluxes are not clearly demonstrated. As such, it remains unclear whether the added model complexity is justified by a substantial gain in predictive performance.
Major comments:
The abstract (Lines 10–16) is somewhat confusing. It is unclear what is meant by “seasonal canopy dynamics”. Does this refer to LAI, carbon fluxes, or other processes? While LAI seasonality is known to be uncertain in ENF biomes, the extent to which these biases propagate to canopy processes remains debatable. For example, based on Fig. 3, the differences in NEE, GPP, and Reco between the BAU (direct use of satellite LAI) and ALT (the proposed method) simulations appear relatively small. This raises the question of whether the impact of LAI seasonality bias on ecosystem fluxes is important.
I think the accuracy of LAI should be evaluated in the context of the specific application. If the goal is to simulate carbon or water fluxes, LAI accuracy may not be critically important (see your Fig 3). For example, EO-derived LAI is known to have limitations, including saturation effects, differences between effective and true LAI (e.g., due to canopy clumping), and biases under snow cover. However, these issues may not strongly affect estimates of fAPAR, which are typically retrieved simultaneously from satellite reflectance and thus more directly reflect the canopy radiative transfer state. Since simulations of carbon fluxes primarily depend on APAR (APAR = fAPAR × PAR), uncertainties in LAI may not substantially propagate into flux estimates. Or it may depend on the upscaling scheme of the model itself (some models depend on using LAI to do the scaling, some using the fAPAR).
Instead, greater importance may lie in the biogeochemical properties of the leaves, such as light-harvesting pigment content and efficiency, Rubisco content and activity, and stomatal regulation. In Fig. 3, all simulated NEE are similar across experiments but differ from the eddy covariance (EC) observations. This suggests that improving the representation of biogeochemical processes in the model may be more important than refining LAI seasonality alone.
Detailed comments:
~ Line 15: “This investigation highlights the need for improved EO LAI estimates in northern evergreen forests…”. I agree that there are biases in EO-derived LAI seasonality in northern evergreen forests. However, the authors should be more specific about how such improvements could be achieved. If the bias primarily arises from snow obscuring satellite reflectance, then it may be fundamentally difficult to resolve within a purely satellite-based framework. On the other hand, if improvement requires incorporating more complex modeling or data assimilation approaches, this should be clearly stated, as it moves beyond a strictly observation-based (EO-only) methodology.
~ Line 40: “LAI EO estimates are typically derived from surface or top-of-atmosphere reflectances measured by passive optical sensors onboard satellites.” This statement is somewhat confusing and depends on how “EO” is defined. If EO refers to satellite-based observations, then all satellite measurements originate from TOA reflectance. However, many LAI products (e.g., MODIS LAI) are actually derived from surface reflectance, which is obtained after atmospheric correction of TOA measurements through upstream processing. The authors should clarify this distinction and use more precise terminology.
~ Line 50: The cited studies (e.g., Cohen et al., 2006; Tian et al., 2004) may be outdated. For example, MODIS LAI products have evolved substantially (currently at Collection 6), whereas these cited were based on much earlier versions (e.g., Collection 1 or 3). There might have been improvements between these versions. The authors should consider citing more recent literature.
Reference: Yan, K., Park, T., Yan, G., Liu, Z., Yang, B., Chen, C., Nemani, R., Knyazikhin, Y., & Myneni, R. (2016). Evaluation of MODIS LAI/FPAR Product Collection 6. Part 2: Validation and Intercomparison. Remote Sensing, 8(6), 460–26. https://doi.org/10.3390/rs8060460
Line 52-54: The Fang et al (2021) analysis is spurious. If you look at Yan et al 2016, their figure 4c shows the seasonality of ENF (their B7). It shows that the seasonal minimum is always above 1 LAI. It is perhaps that Fang et al (2021) did not strictly filtering low quality data based on the satellite QA.
~ Line 80: The manuscript compares two experiments (BAU and ALT), but it is unclear whether the results have been evaluated against other independent benchmarks. Have the authors compared their simulated GPP and ET with products such as MODIS GPP/ET, or with outputs from satellite-driven terrestrial biosphere models (such as BESS, BEPS)?
~ 90: For your hypothesis, is it possible to implement the leaf life span into large scale studies? It seems the leaf life span is based on some strong assumptions, and how to reflect the geographical differences and within species plasticity?
Fig. 1: Field LAI is derived from DHP measurements. How do the authors account for the leaf-to-wood ratio in this conversion? Is this ratio assumed to be constant throughout the year? If the DHP measurements represent plant area index (PAI), they may exhibit relatively weak seasonality, as suggested by Fig. 1. In that case, separating leaf area from woody components becomes critical, and the assumed leaf-to-wood ratio introduces additional uncertainty that should be discussed or quantified. From my experience, the SE-Htm panel shows many extremely low values and abrupt changes in MCD15A2H LAI. This is likely due to incorrect use of the QA flags, or possibly no QA filtering at all. You should exclude all back-up algorithm retrievals as that likely means cloud or snow presents and the reflectance is corrupted (also possible due to sun-sensor poor geometry, but it should be more like to happen during the winter seasons, while your spurious data spread all over the year).
~Lines 123-125: (1) rooting depth is an additional uncertainty. (2) Is CUE temporally constant?
~ Line 129: It appears that other variables (not LAI) with strong seasonal dynamics may be driving much of the variability in the simulated carbon fluxes, LAI and other related outputs, ultimately contributing to the improved seasonality of the results. Could the authors clarify which variables are primarily responsible for this seasonal control? For example, are these driven by meteorological forcings (e.g., radiation, temperature, VPD) or by internal model parameters/processes (stomata, quantum yield efficiency)?
L315: should LMA be a dynamic variable, spatially and temporally?