Modelled LAI is evaluated against the same three satellite-derived LAI products used in the companion paper : (i) LAI3g, a combined dataset based on LAI from collection 5 of the Moderate Resolution Imaging Spectroradiometer (MODISc5) and LAI derived from an Artificial Neural Network, trained on MODISc5, and utilising the GIMMS NDVI data from NOAA-AVHRR sensors, available from 1982 to 2011 ; (ii) CGLS, the Copernicus Global Land Service LAI version 2 product based on spectral data from the SPOT-VGT and PROBA-V sensors, available from 1999 to 2020 ; and (iii) MODISc6, collection 6 of the MODIS LAI product (MOD15A2H), available from 2000 to 2023, .

The CGLS satellite product is used as a reference in the following sections to facilitate the comparison between the satellite and the models. This choice is justified by the good spatial and temporal consistency shown by CGLS . Note that the following sections also provide results from the comparison between the three satellite products.

Finally, the land cover distribution from ESA CCI is used to derive a common Plant Functional Type (PFT) mask to evaluate the differences among LSMs and satellite LAI products at the biome scale.

The seven LSMs utilised in the CRESCENDO project are evaluated in this study. A summary of their main features is provided below and listed in Table . Further details on the LSMs' phenology schemes are provided in .

The Community Land Model (CLM) is the terrestrial component of the Community Earth System Model (CESM, http://www.cesm.ucar.edu/, last access: 17 November 2024) and, in its version 4.5 CLM4.5, and biogeochemical configuration i.e. BGC compset,, it is the land component of the CMCC coupled model version 2 CMCC-CM2, and Earth System Model version 2 CMCC-ESM2,. CLM4.5-BGC features fifteen Plant Functional Types (PFTs), without specific treatment for crop areas. CLM4.5-BGC explicitly resolves carbon-nitrogen biogeochemical cycles , including plant phenology, which is described employing three specific parameterizations: (1) evergreen plant phenology; (2) seasonal-deciduous plant phenology; (3) stress-deciduous plant phenology .

CLM version 5.0 (CLM5.0) is the terrestrial component of the Community Earth System Model version 2 CESM2, http://www.cesm.ucar.edu, last access: 17 November 2024, and of the Norwegian Earth System Model NorESM2,. Compared to the previous version of CLM (i.e. CLM4.5), CLM5 introduces dynamic land units, updated hydrological processes (including revised groundwater scheme, canopy interception and new plant hydraulics functions), revised nitrogen cycling, an improved crop module and various major changes in soil and vegetation parameterization (see Lawrence et al., 2019). Although the phenology scheme of CLM5 is similar to CLM4.5, other model changes (particularly, updated stomatal physiology, nitrogen cycle and plant hydraulics) would indirectly affect the simulated LAI and phenology in CLM5 see.

JULES-ES is the Earth System configuration of the Joint UK Land Environment Simulator (JULES) and it is the terrestrial component of the UK community Earth System Model UKESM1,. JULES-ES implements 13 PFTs in a dynamical vegetation configuration and accounts for a full carbon and nitrogen cycle . The Leaf Area Index (LAI) varies based on the carbon status and extent of the underlying vegetation . Phenology operates based on an accumulated thermal time model.

JSBACH3.2 is the land component of MPI-ESM1.2 . JSBACH3.2 implements 12 PFTs, and the LoGro-P model for phenology . The LoGro-P model uses a logistic equation for the temporal development of the LAI targeting a prescribed PFT-specific physiological limit, independent of the carbon state of the vegetation. JSBACH3.2 distinguishes five phenology types, namely evergreen, summergreen, raingreen, grasses, and tropical and extratropical crops, which is a higher amount of phenology schemes compared to the other land surface models.

LPJ-GUESS is the terrestrial biosphere component of the European Community Earth System Model EC-Earth-Veg,. It simulates biogeochemistry cycles, vegetation dynamics, and land use featuring 25 PFTs. Similar to the two CLM models, LPJ-GUESS uses three phenology schemes: (1) evergreen plant phenology; (2) seasonal-deciduous plant phenology; (3) stress-deciduous plant phenology.

ORCHIDEE is the land component of the IPSL (Institut Pierre Simone Laplace) Earth System Model used in the CMIP6 effort . ORCHIDEE features 15 PFTs that vary based on the LUH2 forcing . The phenology module describes leaf onset and senescence based on temperature and soil moisture .

ISBA-CTRIP is the land component of CNRM-ESM2-1 and it works within the SURFEX version 8 modelling platform. It accounts for 16 vegetation types alongside desert, rocks and permanent snow . Differently from the other land surface models, ISBA-CTRIP computes the leaf phenology based on the daily carbon balance of the leaves as described in .

All LSMs were forced by near-surface atmospheric variables (2 m air temperature, precipitation, wind, surface pressure, shortwave radiation, longwave radiation, and air humidity) from the CRUNCEP version 7 reanalysis dataset , following the TRENDY protocol , and land cover values from the Land Use Harmonization version 2 LUH2,. Despite each LSM implementing the LUH2 data differently (e.g. different number of PFTs), the same vegetated areas evolution forces them to leave differences in plant growth, biodiversity, and seasonality among them.

LSM simulations cover 1850–2014, following the historical period, as defined in CMIP6 . Consequently, the comparison between models and satellite data covers the shared period from 2000 (the starting year of MODIS data) to 2011 (the last available year of LAI3g), as done in the companion paper .

To facilitate intercomparison across models and satellite data, modelled LAI on Plant Functional Types (PFT) were weighted averaged by PFT fraction for each grid box to produce an estimate of grid box mean LAI for all models. As each LSM was run on a different grid resolution (Table ), to enable cross-analyses the model outputs and satellite data used in this study were expressed as monthly means and regridded to a regular 0.5°×0.5° grid using the Climate Data Operators (CDO) toolset first order conservative remapping scheme .

Global variations in the monthly mean quantity, timing, and amplitude of annual LAI peaks and troughs (2000–2011 mean) derived from the satellite products (Sect. ) are compared with the LSM estimates (Sect. ) of the same variables. Peaks are identified as the month with the highest LAI value, representing the apex of the vegetation life cycle. On the contrary, the troughs are identified as the months with the lowest LAI value, depicting the plant's dormancy season. For a situation where the same minimum or maximum LAI values are recorded in multiple months per year, the first month in the year with that value is retained as the peak or trough. The values of peak and trough are computed for each of the satellite observation datasets and land surface models. Results from the land surface models are also aggregated and evaluated as a multi-model ensemble mean (MME). Finally, the agreement between LSMs and satellite products refers to differences of 0 months in peak and trough (i.e. both LSM and satellite product produce peak and trough occurring in the same month) and of 0.25 m2m-2 in LAI amplitude.

At the biome scale, the LAI peak timings estimated by the LSMs are generally delayed compared to the satellite estimates, particularly across regions dominated by both needleleaf and broadleaf trees (Fig. b–e, i, and j). An exception is for broadleaf evergreen trees (BET) in both hemispheres (Fig. d and i). In the regions dominated by BET (about 11 % of the vegetated regions), the simulated LAI peak values fall in the 25/75 percentile distribution of the satellite estimates (Fig. S5). The peak of broadleaf deciduous shrubs (BDS) in the northern hemisphere is also delayed compared to CGLS, MODIS, and LAI3g (Fig. f), which is not the case in the southern hemisphere for the MME (Fig. k). However, the LSMs exhibit large variability in the BDS biome (Fig. S5k). Similar hemispheric behaviour is observed in grass-dominated areas (Fig. g and l), while the crop biome is reasonably well captured (Fig. h and m).

The LAI trough timings estimated by the LSM are generally delayed compared to the satellite estimates in regions dominated by needleleaf trees (NDT, NET), broadleaf deciduous trees (BDT), Crop, and northern hemisphere Grass (Fig. b, c, e, g, h, j, and m). The trough timings of Broadleaf Evergreen Trees (BET, Fig. d and i), especially in the northern hemisphere (Fig. d), and Broadleaf Deciduous Shrubs (BDS, Fig. f and k), and southern hemisphere Grass (Fig. l) show values in agreement with satellite values (see also Fig. S6).

In general, the growing seasons simulated by the LSM show delays in their peaks compared to the satellite estimates, especially in the northern hemisphere. Moreover, LSMs sharing similar phenology parameterisation schemes, such as CLM4.5, CLM5.0, and LPJ-GUESS (Table ), display discrepancies in phenophases estimates, such as in the southern hemisphere BET biome, where CLM5.0 differs from CLM4.5 and LPJ-GUESS by approximately 6 months (Figs. S5i and S6i). This behaviour highlights the influence of models' features beyond the specific phenology schemes in representing the growing season cycle.

Focusing on seasonal amplitude, LSMs tend to simulate smaller seasonal amplitude in the areas dominated by needleleaf trees (Fig. b and c). This behaviour is mainly driven by ORCHIDEE (Fig. S7b and c). Reduced differences between LAI maximum and minimum are also simulated in the Broadleaf Deciduous Tree (BDT) biome (Fig. e and j), mainly due to JULES-ES (Fig. S7e and j). The seasonal amplitude is reasonably well captured in the Grass- and Crop-dominated areas (Fig. g, h, l, and m). The Broadleaf Deciduous Shrub (BDS) biome exhibits seasonal amplitude in line (Northern Hemisphere, Fig. f) or slightly longer (Southern Hemisphere, Fig. k) compared to satellite products. Finally, the regions dominated by Broadleaf Evergreen Trees (BET) display simulated seasonal amplitude values in agreement with CGLS and MODIS but smaller than LAI3g (Figs. d, i and S7d, i).

Several LSMs represent LAI values based on the values of specific leaf area and the amount of leaf carbon or biomass content (i.e. CLM4.5, CLM5.0, LPJ-GUESS, and ISBA-CTRIP, Table ). The implementation of similar parameterisation reflects on reduced differences between LSMs (Fig. S7), except for southern hemisphere BDS-dominated areas (Fig. S7k), where LPJ-GUESS substantially overestimates the LAI seasonal amplitude compared to CLM4.5, CLM5.0 and ISBA-CTRIP. On the other hand, LSMs primarily driven by temperature, such as JULES-ES and ORCHIDEE, tend to underestimate the LAI seasonal amplitude (Fig. S7), which is not the case when also leaf features are considered, as done in JSBACH (Table ). This comparison, then, underscores the need to incorporate leaf features and leaf carbon content in LAI computation within LSMs.

The LAI simulated by LSM shows differences with satellite products in both the timings (i.e. peak and trough) and the quantities (i.e. seasonal amplitude). Since these two metrics (i.e. timings and quantities) assess different vegetation features, there is no clear consistency between them. For example, JULES shows a general underestimation of seasonal amplitude compared to CGLS (Fig. d) with peak and trough timings dominated by delayed values (Figs. d and d). Similar delayed peak and trough timings are simulated by CLM5, ORCHIDEE and partially JSBACH (Figs. c, e, g and c, e, g) yet for these models an underestimation in seasonal amplitude is only noted for ORCHIDEE (Fig. g), CLM5 and JSBACH show a mixture of positive and negative differences in seasonal amplitude (Fig. c and e). On the other hand, LPJ-GUESS shows peak and trough timings that are reasonably close to the satellite estimates (Figs. f and f), while the seasonal amplitude for this model is broadly overestimated (Fig. f).

This behaviour emphasises the detachment between LSM ability in representing the vegetation timings (i.e.phenology) and quantity, as previously presented for CLM4.5 in even if the same variable, i.e. LAI, can be applied in assessing both metrics. These differences derive from the different forcings for vegetation timings, usually determined by temperature, soil moisture, solar radiation, and quantity, which is co-determined by climate forcing, photosynthesis and carbon and nitrogen allocations. Consequently, both metrics need to be considered when evaluating LSMs, as done here.

LSMs show a delayed peak in forested areas compared to satellite estimates combined with a delayed start of the growing season, as shown in Fig. 7 by , suggesting a possible modelled too-slow leaf production. LSMs also exhibit a delay in trough timings , suggesting a general shift in the growing season Fig.  and proposed in the companion paper. The assessment of the start, peak, and end of the growing season timings points out a temporal shift of the vegetative active season in LSMs compared to satellite records. Nevertheless, only in limited areas between 17 % and 35 % of vegetated areas, Table 3 in LSMs correctly reproduced growing season length.

The combination of the results from the present study and its companion highlights LSMs' limitations in correctly capturing timings, amplitude, and length of the vegetative active season pointing at the need for further development of vegetation phenology, distribution, and mass representation in the LSMs. In general, LSMs should improve their ability to capture vegetation heterogeneity in both plant traits, such as a higher number of PFTs, and phenology features, for example, by increasing the variety of phenology schemes to enhance their ability to represent vegetation responses to various stresses.

The LSMs involved in this study and its companion use state-of-the-art boundary conditions (i.e. atmospheric and land use forcings). The LUH2 land-use evolution employed in the CMIP6 effort is used in this study. It is noteworthy that each LSM implements the same land-use boundary conditions differently due to dissimilarities in the original resolution, number of PFTs, and land-use scheme implemented in each model (Table ). Nonetheless, the implementation of a common land-use dataset allows all LSM to reproduce the same vegetated areas evolution, leaving only differences in plant growth, biodiversity, and seasonality among them. Besides, the same set of atmospheric variables has been used to force all the LSMs following the TRENDY protocol . However, the atmospheric conditions (e.g. temperature and water availability) strongly influence the vegetation growth, and a different source of atmospheric forcing may partially affect the biases obtained in this study. For example, the ISBA-CTRIP model forced by WFDEI forcing exhibits smaller biases compared to the present results .

Despite using the same boundary conditions (atmospheric and land use forcings), the variability between the LSM may derive from various sources, such as differences in vegetation parameterization, crop and plant functional type population, soil characterization, and initial spatial resolution, as already noted in . In particular, the discrepancies in model grid resolution and a relatively coarse initial spatial resolution (between about 2° and 0.5°, Table ) induce differences in the simulated grid vegetation mixture, which may explain the mismatch between LSMs, especially in regions characterised by high biodiversity and areas with evergreen forests. The availability of data at the PFT level would reduce the resolution impact and refine the investigation of differences between LSMs as requested for the next phase of the Coupled Model Intercomparison Project CMIP7,. In general, the results of this study highlight the relevance for LSMs to capture the high vegetation heterogeneity on both PFTs and CFTs populations, that is the case for LPJ-GUESS, and phenology schemes, as for JSBACH.

Moreover, vegetation parameterizations used in the LSM are based on data from localised areas, typically located in the northern hemisphere e.g., which may lead to a possible misrepresentation of south hemisphere features. Consequently, the parameters used in LSMs need to be calibrated against more recent and widespread observations.

The results presented in this study highlight a prevalent delay in plant active season compared to observations, despite each LSM implementing different parametrization and processes, emphasising the need for further investigation of the representation of the processes involved in the start of the growing season within the LSMs.

In addition, the case of the Community Land Model provides an example of the impact of model structure versus parameterization on simulated LAI. In the present study, CLM is evaluated in two versions, namely CLM4.5 and CLM5.0. The latter contains various changes in the representation of soil, plant hydrology and carbon and nitrogen cycles i.e. model structure,, influencing the simulated vegetation quantity (see differences in seasonal amplitude in Fig. ). On the other hand, CLM5.0 applies limited modifications to the phenology parameterization, resulting in minimal differences in the biases of growing season timings compared to its earlier version (Figs. b and c and b and c), as also shown in . The comparison between CLM versions, then, stresses the separate influence of model structure and phenology parameterization on the simulated LAI features. Moreover, both CLM4.5 and CLM5.0 implement the same LUH2-derived PFT distribution, avoiding the influence of mismatched vegetation type in this version comparison. In general, it highlights the complexity of modelling land surface and vegetation processes and the need for further model development and evaluation.

Finally, a detailed comparison between phenology parameterizations requires data at the PFT level and a mapping between PFTs and phenology schemes as done by and requested in the next Coupled Model Intercomparison Project phase CMIP7,. The availability of that information will provide the possibility to improve our knowledge of the limitations and abilities of each phenology scheme.

Despite the satellite datasets agreeing in about half of the vegetated regions (Table ), differences emerge between them (Table ), even with peaks of up to 3 months in timings and above 1 m2m-2 in magnitude in limited areas, but smaller than the differences between LSMs and satellite records. The discrepancies between the three products may be caused by differences in their satellite types and orbits, spectral sensors, LAI estimation approaches, and other technical differences monthly averaging, gap filling, reflectance saturation, and spatial interpolation techniques, e.g..

The differences among satellite products occur in the tropics and high-latitude regions, which are often challenged by frequent clouds, snow, and polar nights. Moreover, each satellite product has a different approach for data reconstruction in the winter season above 55° N. For example, MODIS does not provide data in regions above 55° N during December and January, while CGLS uses information from climatology namely GEOCLIM, to fill the missing values in the winter northern high-latitude regions. Similar to MODIS, LAI3g does not provide data during the winter season in the northern hemisphere latitude but with a different latitudinal threshold (about 65° N).

Finally, LAI satellite datasets derive from empirical or statistical relationships with canopy reflectance or vegetation indices (Fang et al., 2019), making the LAI satellite records model-derived products (not direct observations) characterized by assumptions and uncertainties as emphasized by previous literature works that stress differences, caveats, and uncertainties of satellite products e.g.. For this reason, three separate satellite LAI products obtained from different acquisition sensors (namely AVHRR for LAI3g, MODIS for MODIS LAI, and SPOT/PROBA VEGETATION for CGLS) have been used in this study.