iMAPLE is an interactive model developed to investigate the interplay between air pollution and terrestrial ecosystems (Yue et al., 2024). The model evolved from version 1.0 of the Yale Interactive terrestrial Biosphere (YIBs) model, with an expanded focus on explicitly representing the interaction between atmospheric chemistry and land ecosystems (Yue and Unger, 2015). Compared to YIBs, iMAPLE incorporates several new processes, including dynamic fire emission (Pechony and Shindell, 2009; Li et al., 2012), wetland methane emissions (Walter et al., 2001; Zhu et al., 2014), and photosynthetic limitation under environmental stress (Arora et al., 2009). Furthermore, iMAPLE integrates the process-based hydrological scheme Noah-MP, including precipitation partitioning, infiltration, soil moisture redistribution, evapotranspiration, runoff generation, and groundwater-related processes. This enables dynamical simulation of soil temperature and moisture and facilitates tight coupling between the carbon (simulated by YIBs) and water (simulated by Noah-MP) cycles (Niu et al., 2011). These improvements substantially enhance the model's capacity to represent ecosystem–atmosphere interactions. Below, we briefly describe the parameterizations of vegetation biophysics and hydrological processes. A comprehensive model description is provided in Yue et al. (2024) and Niu et al. (2011).

ECHAM6 is the sixth generation of the ECHAM atmospheric general circulation model, developed by the Max Planck Institute for Meteorology (MPI-M) (Giorgetta et al., 2013; Stevens et al., 2013). Its dynamical core employs a hybrid numerical framework that combines spectral and finite-difference methods to solve the primitive equations (Arakawa, 1966; Bourke et al., 1977). Horizontally, ECHAM6 utilizes a truncated spherical harmonic expansion to calculate the nonlinear and parameterized terms of dynamical fields on a Gaussian grid. Vertically, the model employs a hybrid sigma-pressure coordinate system implemented on a Lorenz grid (Phillips, 1957; Simmons and Burridge, 1981). Turbulent mixing in the atmosphere is represented by a turbulent kinetic energy (TKE) scheme, and surface fluxes are computed using bulk transfer coefficients with Louis-type stability functions based on the bulk Richardson number (Louis, 1979; Brinkop and Roeckner, 1995). Land roughness lengths are provided by JSBACH and evolve with the surface state.

Compared to previous ECHAM versions, ECHAM6 incorporates substantial improvements in key physical processes, including a more sophisticated representation of land surface processes, updated radiation parameterizations, refined surface albedo calculations, and revised convection triggering criteria (Stier et al., 2005; Crueger et al., 2018; Tegen et al., 2019). In the default ECHAM6 configuration, land-surface processes are handled by JSBACH. After coupling with iMAPLE, JSBACH remains responsible for most land-surface calculations in ECHAM6, whereas iMAPLE is used for vegetation processes. Details of the exchanged variables are given in Sect. 2.3. The model supports multiple configurations, with horizontal resolutions spanning from T31 (3.75°) to T255 (0.7°) and 47 or 95 vertical layers (Stevens et al., 2013). In this study, we adopt the T63L47 configuration to enable coupling with the iMAPLE model. The T63 spectral grid is characterized by an approximate grid spacing of 1.875° in longitude and latitude. The L47 vertical configuration comprises 47 layers extending from the surface to 0.01 hPa.

ECHAM6 employs JSBACH version 3.10 to simulate land vegetation processes (Goll et al., 2017). Compared to the earlier versions, JSBACH v3.10 incorporates a soil water diffusion scheme that provides a more physically consistent representation of terrestrial water budgets. In addition, modules for the dynamical simulation of LAI, canopy radiative transfer, and leaf-level photosynthesis have been integrated into the current versions. Here, we provide a brief description of JSBACH v3.10 to facilitate the comparisons of its physical processes with those of iMAPLE v1.0.

In this study, iMAPLE was coupled to the ECHAM6 global circulation model (Fig. 1) and configured to run in parallel with JSBACH, enabling a systematic comparison of their process representations and model performance. Although both JSBACH and iMAPLE participated in the multi-model intercomparison project for Global Carbon Budget, iMAPLE demonstrated superior performance relative to JSBACH in simulating key carbon (e.g., GPP) and water (e.g., ET) fluxes (Friedlingstein et al., 2025). In addition, iMAPLE incorporates a more comprehensive representation of biogeochemical processes, providing greater potential for extending land-atmosphere coupling to atmospheric chemistry in future applications.

ECHAM6 integrates atmospheric dynamic processes at an interval of 7.5 min, which is much shorter than the 60-minute interval used in iMAPLE. As a result, simulated climatic variables from ECHAM6 are passed to iMAPLE every 8 ECHAM6 time steps. To ensure seamless information exchange, iMAPLE adopts the same core functional modules as ECHAM6 for input/output (I/O) operations, memory management, parallel computation, and calendar handling. At each coupling interval, ECHAM6 provides hourly meteorological variables to iMAPLE, including precipitation, surface air temperature, wind speed, humidity, surface pressure, CO₂ concentrations, direct and total radiation. In return, iMAPLE feeds back simulated soil temperature, soil moisture, and ET to ECHAM6. These variables further change the surface energy and water balance, thereby altering heat and moisture fluxes that define the atmospheric lower boundary conditions and regulate near-surface climate. Land-atmosphere energy exchange is governed by the surface energy balance, which constrains the transfer of energy between the land surface and the atmosphere: 15C∂Tsurf∂t=Rnet+Hsensible+Hlatent+G where Tsurf is the surface temperature, C is the heat capacity of topmost soil layer, Rnet denotes the net radiation, Hsensible is the sensible heat flux, Hlatent is the latent heat flux, G represents the ground heat flux, which is obtained from the soil temperature profile: 16G=-λs∂T∂z|Zsurf where λs is the soil thermal conductivity and z is the depth. Replacing soil temperature therefore revises the near-surface thermal gradient and G, which in turn shifts the tendency of Tsurf and the partitioning of surface energy into sensible heat. The latent heat flux is linked to ET as follows: 17Hlatent=L⋅ET where L is the latent heat of vaporization. Replacing ET revises Hlatent and affects the surface energy balance. The replacement of near-surface soil moisture updates the land hydrological state, thereby modifying available soil water and regulating subsequent land-atmosphere moisture exchange. The resulting surface heat and moisture fluxes provide the lower boundary conditions for vertical turbulent transport of dry static energy and specific humidity in ECHAM6. Specifically, an implicit surface-atmosphere coupling scheme uses surface dry static energy and saturated surface humidity, constrained by the surface energy balance, as lower boundary conditions, after which the atmospheric column is updated upward from the lowest model level. Consequently, changes in land surface states, including soil temperature, soil moisture, and ET, propagate to near-surface air temperature and humidity, and further influence the atmospheric column through turbulent diffusion. Extensive numerical tests were conducted to ensure integration stability and computational efficiency of the coupled modeling framework.

The land cover dataset used in iMAPLE was developed by integrating satellite-based observations from Moderate Resolution Imaging Spectroradiometer (MODIS) and Advanced Very High Resolution Radiometer (AVHRR) (Defries et al., 2000; Hansen et al., 2003), which include nine PFTs representing major global terrestrial ecosystems. In contrast, JSBACH employs a flexible land-cover library that allows users to specify the number and properties of PFTs. To reduce uncertainties arising from differences in boundary conditions, the iMAPLE land cover dataset was harmonized and mapped onto the JSBACH PFT categories (Table 1). Comparisons of simulated GPP and LAI from ECHAM6-JSBACH showed improved performance when using the updated PFT derived from iMAPLE (Table 2). Therefore, this updated land cover configuration of ECHAM6-JSBACH is adopted for subsequent comparison with ECHAM6-iMAPLE. The experiments were driven by observed sea surface temperature and sea ice fraction without atmospheric nudging. The interactive HAM/MOZ chemistry and aerosol modules were disabled. Greenhouse gas concentrations followed the CMIP6 historical pathway. Both ECHAM6-iMAPLE and ECHAM6-JSBACH simulations used identical atmospheric settings at T63L47 resolution for the period of 2000–2014. The first five years were used for model spin-up, and the remaining 10 years were averaged for analysis and performance assessment.

Observational and reanalysis datasets from 2005 to 2014 were used to evaluate and compare the performance of the ECHAM6-iMAPLE model and ECHAM6-JSBACH models. Specifically, Global LAnd Surface Satellite (GLASS) products were used to assess LAI. For GPP and ET, we used the benchmark product of FLUXCOM datasets (Tramontana et al., 2016; Jung et al., 2020), which generate global gridded flux estimates using machine learning algorithms trained on flux measurements from FLUXNET eddy covariance sites. We also used the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) product (Gelaro et al., 2017) for the evaluations of soil temperature, soil moisture, surface air temperature, and precipitation. Near-surface soil moisture was derived from the MERRA-2 surface wetness variable GWETTOP, which represents relative saturation in the topmost soil layer. GWETTOP was converted to volumetric water content (m³ m⁻³) by multiplying it by the grid-cell soil porosity. All datasets were interpolated to the T63 spectral resolution for consistent comparisons, corresponding to a horizontal resolution of approximately 1.875° in both latitude and longitude.

We evaluated the performance of ECHAM6-iMAPLE and ECHAM6-JSBACH in simulating terrestrial carbon fluxes (Fig. 2). Both models reasonably capture the spatial distribution of GPP consistent with FLUXCOM observations, characterized by high values in tropical regions, moderate values in boreal forests, and low values in arid areas. ECHAM6-iMAPLE simulates a global annual total GPP of 126.9 Pg C yr⁻¹, slightly higher than that of 124.1 Pg C yr⁻¹ estimated by the FLUXCOM benchmark. In contrast, ECHAM6-JSBACH produces a much higher global GPP of 134.2 Pg C yr⁻¹, with pronounced positive biases mainly over tropical forest regions, including the Indian subcontinent, the Amazon Basin, and Central Africa (Fig. 2c). Relative to ECHAM6-JSBACH, ECHAM6-iMAPLE exhibits improved performance (Fig. 2b), reflected by a higher correlation coefficient (R=0.75, p<0.01) and a lower root mean square error (RMSE = 1.51 g C m⁻² d⁻¹). This improvement can be primarily attributed to the calibration of vegetation photosynthetic parameters in iMAPLE based on site-level datasets (Yue and Unger, 2015), as well as the constraint of plant phenology through satellite-based observations (Yue et al., 2015).

Consistent with the GPP pattern, simulated LAI from ECHAM6-iMAPLE (Fig. 2e) shows a higher R of 0.86 (p<0.01) and a lower RMSE of 0.73 m² m⁻² against GLASS retrievals (Fig. 2d) compared to ECHAM6-JSBACH (Fig. 2f). Regionally, ECHAM6-iMAPLE successfully reproduces the elevated LAI in tropical rainforests, in good agreement with observational patterns. Globally, ECHAM6-iMAPLE underestimates LAI by about 9 %, primarily due to notable underestimation across high-latitude regions of the Northern Hemisphere (Fig. 2e). In contrast, ECHAM6-JSBACH shows a higher RMSE of 0.79 relative to GLASS, with the largest negative biases occurring in the central Africa and the Southeast Asia region (Fig. 2f).

We evaluated the simulated water fluxes from ECHAM6-iMAPLE and ECHAM6-JSBACH (Fig. 3). Both models successfully reproduce the observed ET patterns, characterized by hotspots over tropical rainforest and low values in arid and/or cold regions. Simulated ET from ECHAM6-iMAPLE (Fig. 3b) exhibits a higher spatial correlation with the FLUXCOM benchmark (Fig. 3a) (R=0.86, p<0.01) and a lower RMSE (15.2 mm per month) compared to ECHAM6-JSBACH (Fig. 3c). Globally, the area-weighted mean ET simulated by ECHAM6-iMAPLE is 45.5 mm per month, closely matching the observed value of 45.9 mm per month, whereas ECHAM6-JSBACH underestimates ET at 42.1 mm per month. The poorer performance of ECHAM6-JSBACH is mainly attributed to substantial underestimations in arid and semi-arid regions, particularly over the western United States and Australia (Fig. 3c).

For water use efficiency (WUE), defined as the ratio of GPP to ET, ECHAM6-iMAPLE exhibits high values in the northern middle-to-high latitudes and moderate values in the tropics (Fig. 3e). It achieves a spatial R of 0.52 (p<0.01) and an RMSE of 0.77 g C kg⁻¹ H₂O against the FLUXCOM benchmark (Fig. 3d). In contrast, ECHAM6-JSBACH exhibits substantially poorer performance, with a lower spatial R of 0.20 and a higher RMSE of 0.95 g C kg⁻¹ H₂O (Fig. 3f), which is primarily associated with underestimated ET across mid-to-high northern latitudes and Central Africa. At the global scale, the area-weighted mean WUE simulated by ECHAM6-iMAPLE is 1.43 g C kg⁻¹ H₂O, closer to the FLUXCOM estimate of 1.55 g C kg⁻¹ H₂O and representing a clear improvement over the 1.35 g C kg⁻¹ H₂O simulated by ECHAM6-JSBACH (Fig. 3f).

In addition to the spatial pattern, ECHAM6-iMAPLE effectively captures the seasonal cycles of terrestrial carbon and water fluxes (Fig. 4). Consistent with observations, simulated GPP and ET both reach their annual maxima during the vegetation growing season, with the most pronounced peak occurring in July (Fig. 4a and c). Although the LAI simulated by ECHAM6-iMAPLE exhibits a one-month lag relative to the observations, the temporal correlation remains high (R=0.96), indicating a good representation of seasonal dynamics (Fig. 4b). In contrast, ECHAM6-JSBACH predicts the annual peaks of terrestrial carbon and water fluxes in June, one month earlier than observed peaks. Moreover, the corresponding R between ECHAM6-JSBACH simulations and observations are consistently lower than those derived from ECHAM6-iMAPLE (Fig. 4). The divergence in peak timing and the systematically lower R values collectively demonstrate that ECHAM6-iMAPLE provides a more accurate representation of the seasonal dynamics of terrestrial carbon and water cycle processes compared to ECHAM6-JSBACH.

Soil temperature and moisture play critical roles in regulating vegetation growth (Davidson and Janssens, 2006; Green et al., 2019). Both ECHAM6-iMAPLE (Fig. 5b) and ECHAM6-JSBACH (Fig. 5c) predict high soil temperatures in tropical regions, especially in northern Africa and the Indian Peninsula. The two models exhibit extremely high spatial correlations with MERRA-2 reanalysis data (R=0.99 and 0.98, respectively). However, ECHAM6-iMAPLE more accurately reproduces the global mean soil temperature and shows a lower RMSE of 2.16 K, whereas ECHAM6-JSBACH shows a slight warm bias with a higher RMSE of 2.49 K. For soil moisture, ECHAM6-JSBACH shows substantial overestimation in tropical regions and the high-latitude Northern Hemisphere, leading to a 39.1 % overestimation of the global mean value compared to MERRA-2 (Fig. 5f). In contrast, ECHAM6-iMAPLE closely matches the global mean soil moisture from MERRA-2, with a lower RMSE of 0.07 and a moderate spatial R of 0.76 (Fig. 5e).

Changes in soil temperature and moisture influence surface meteorology through land-atmosphere coupling processes (Seneviratne et al., 2006). We therefore compared surface air temperature and precipitation simulated by ECHAM6-iMAPLE and ECHAM6-JSBACH (Fig. 6). Both models reproduce the large-scale spatial distribution of surface air temperature, albeit with overall warm biases (Fig. 6b and c). However, ECHAM6-iMAPLE performs better than ECHAM6-JSBACH, showing a smaller global mean warm bias (0.34 °C vs. 0.76 °C) and a lower RMSE (3.33 °C vs. 3.47 °C). This improvement is likely associated with the more accurate simulation of soil temperature in ECHAM6-iMAPLE (Fig. 5b), highlighting the role of land surface processes in constraining near-surface climate. For precipitation, however, both models substantially overestimate the global mean amount with regional overestimation in Asia but underestimation in central Africa (Fig. 6e and f). Although coupling with iMAPLE improves the simulation of ET (Fig. 3b) and soil moisture (Fig. 5e), the complex and nonlinear feedbacks within the hydrological cycle may instead magnify biases in precipitation estimates.