Land biosphere processes are of central importance to the climate system. Specifically, ecosystems interact with the atmosphere through a variety of feedback loops that modulate energy, water, and
The land surface is of central importance in the climate system, as feedbacks between the land biosphere and the atmosphere impact climate across a wide range of temporal and spatial scales
The climate and the biosphere are also coupled biogeochemically through the carbon cycle
Of particular importance is the added complexity arising from land use and land cover change. Conversion of forests into cropland or grassland increases surface albedo, which may promote surface cooling at temperate latitudes
Incorporating DGVMs into ESMs allows the interactions between the biosphere and the rest of the climate system to be studied on the long timescales of vegetation dynamics and biogeochemical and biogeographical responses
DGVMs are frequently integrated into ESMs through an intermediary land surface model (LSM), which facilitates the sub-daily energy, water, and gas exchange calculations
Coupling LPJ-GUESS with an atmospheric model requires it to be able to calculate diurnal energy and water exchange rates between plant canopies and the atmosphere. To achieve this, we introduced several major modifications to LPJ-GUESS v4.0, namely (a) energy balance closure on a sub-daily time step and (b) a new radiative transfer scheme, capable of calculating upwelling short-wave radiation dynamically on a sub-daily time step as well as accounting for direct and diffuse solar radiation separately, and (c) an improved representation of heat and water transport in the soil. Section
LPJ-GUESS
LPJ-GUESS can represent managed land (croplands, pastures/rangelands, and managed forest) and land use change
Figure
Flowchart of the main daily simulation loop in standard LPJ-GUESS (red branch) and the modified version (LPJ-GUESS/LSM, blue branch). The shaded area indicates the sub-daily loop in the modified version. The dashed line encloses coupled iterative calculations.
Radiative transfer in standard LPJ-GUESS is based on Beer's law
In the LSM implementation, radiative transfer, energy balance, assimilation, and soil heat and water transport are all solved on a sub-daily basis. Elaborating on
We replaced the original two-layer soil column with a new profile consisting of nine layers. The top four layers have thicknesses of 7, 10, 13, and 20
The sunlit and shaded leaves of each cohort have different assimilation rates and stomatal conductances. The temperatures of sunlit and shaded leaves are different but common to all the cohorts in the patch. The vertical layering of the canopy is kept in the radiation calculations, but the new scheme distinguishes direct and diffuse radiation and two separate wavebands (visible and near infrared). Infrared radiation does not contribute to photosynthetic assimilation but needs to be accounted for in the energy balance calculations. A separate treatment of diffuse and direct radiation allows us to resolve sunlit and shaded leaves. This approach has been shown to lead to predictions of fluxes of energy, water, and
Each cohort exchanges sensible and latent heat with a common canopy air space, which in turn exchanges sensible and latent heat with the atmosphere (Fig.
Networks of sensible (red) and latent (blue) heat exchange between the ground surface, the canopy, and the atmosphere in the patch. Light green indicates the sunlit fraction of the cohorts, dark green the shaded fraction.
The energy balance of the patch canopy is described by the following equations
The sensible heat flux from the sunlit part of the canopy to the canopy air space is formulated as
The latent heat flux from the sunlit part of the canopy to the canopy air is
The energy balance equation for the ground surface is
The heat conducted into the ground is calculated as
The following two equations express conservation of latent and sensible heat:
Equations (
We adapted the two-big-leaf model of
The direct-beam radiation absorbed in a canopy layer
The short-wave radiation absorbed by the ground surface is calculated as the difference between the downward and upward beams at
The short-wave radiation reflected back at the atmosphere is obtained by evaluating the upward beams at
The optical elements in the canopy have different properties in the visible and near-infrared wavebands, so the equations above are applied separately to these two parts of the spectrum, and the contributions are summed to calculate total absorption. In order to keep the model development process tractable, we set the optical properties of the canopy to the following values, regardless of PFT:
The PAR absorbed by the sunlit leaves of a cohort
The long-wave radiation emitted by the sunlit part of the canopy is
The long-wave radiation emitted by the ground surface is
The bulk long-wave radiation emitted by the land surface toward the atmosphere is
In what follows, variables that are updated daily are denoted with the subscript “day”. Daytime averages are denoted with the subscript “dt”. All the other variables are computed on a sub-daily basis. Photosynthetic assimilation is now calculated within the sub-daily energy balance routine (Fig.
For a given cohort
The daytime averaged leaf temperature,
Separating the contributions to daily absorbed PAR from sunlit and shaded leaves, maximum carboxylation rates for the sunlit and shaded parts of the cohort are estimated as
The water stress factor
The CLM-type water uptake response function is formulated in terms of matric potential
The SSiB type water uptake response function is
Figure
Stomatal conductance and photosynthetic rate are related through a semi-empirical model. The photosynthesis rate depends on the
As noted above, we implemented two selectable stomatal conductance models. In the Ball–Berry model
Stomatal conductance as a function of water vapour deficit at the leaf surface.
Factor limiting plant water uptake as a function of volumetric soil water content. The dashed vertical lines represent, from left to right, the volumetric soil water content at a wilting matric potential of
In standard LPJ-GUESS, soil temperature is used in calculations related to ecosystem respiration and nitrogen cycling, while soil water content influences plant water uptake and evapotranspiration. Both quantities affect soil organic matter decomposition rates.
Soil temperature
Vertical water transport in the soil column is described by the Richards equation
Soil temperature, water content, ecosystem respiration, plant water uptake and evapotranspiration are calculated in the sub-daily loop. Equations (
The revised model was verified by performing energy and water conservation tests. At any given time step, the energy conservation error per unit time and per unit patch area,
The histogram shows the energy conservation error, as a percentage of the energy input, incurred at every time step. The symbols indicate the mean absolute error corresponding to each bin. The error bars indicate
The water conservation error is computed as
We therefore conclude that the magnitude of the errors in energy balance closure and water conservation are negligible.
We evaluated the revised model by comparing hourly and monthly simulated fluxes of sensible and latent heat and annual
List of plant functional types in the standard configuration of LPJ-GUESS (only PFTs predicted by the simulations in this study are listed).
Brief description of selected sites. The land cover classification was taken from the FLUXNET site description web pages. The reference level height is taken as the height of the measuring sensors above the canopy. A dash indicates that we were not able to find an observed LAI value for the site.
FLUXNET sites selected for model evaluation. Different symbols indicate different land cover types. The sites are labelled according to their site code (Table
For each site, we ran eight simulations, covering all possible configurations of the water uptake response functions and stomatal conductance schemes described in Sect.
Summary of the LPJ-GUESS/LSM simulations carried out. Simulations with different stomatal conductance schemes are arranged in columns: Ball–Berry (BB) and Medlyn (Med). Simulations with different water uptake response function types are arranged in rows: NOAH, CLM, modified CLM, and SSiB.
All natural PFTs were allowed to establish in forest and savanna sites. Since the focus of the model evaluation was placed on the turbulent fluxes, we restricted the simulated PFTs to grassy types at sites classified as grasslands, which limits modelled surface roughness. This was also done for Spain-Amoladeras and Congo-Tchizalamou. Amoladeras is classified as an open shrubland on the FLUXNET reference, but the vegetation is short and the most abundant species is
The simulations were spun up from a bare ground state following a standard procedure that combines 500 simulation years with a semi-analytic calculation of the equilibrium size of the soil organic matter pools (see Supplement), to bring C and N soil and vegetation pools to near equilibrium with the climate. During the spin-up phase, the site climate spanning the whole measurement period was repeated cyclically, with interannual trends in air temperature removed, and the atmospheric
Half-hourly measured fluxes were converted to hourly averages for direct comparison with model outputs. Sub-daily FLUXNET data are classified into four quality categories: 0 (measured), 1 (good-quality gap fill), 2 (poor-quality gap fill), and 3 (downscale from ERA reanalysis data). In our analysis, we only used sub-daily fluxes with a quality flag of 0 or 1. For monthly and annual fluxes, the quality flag varies between 0 and 1 and indicates the fraction of the sub-daily values in that month/year whose quality is either 0 or 1. We only used monthly and annual fluxes with a quality flag equal to or greater than 0.75. Following
LAI values for the spin-up period at three selected sites: PA-SPs
To evaluate the agreement between measured and simulated turbulent heat fluxes at each site for all different model configurations we used standard statistical metrics: correlation coefficient (
Panels
Foliar projective cover, averaged over the whole simulated period, of the plant functional types predicted for each site, given as a percentage. The LSM simulations use the CLM-type water uptake response factor. The dominant PFT for each site is highlighted in bold font.
The emerging ecosystem composition in both LSM runs is similar to the standard LPJ-GUESS prediction over forests and grasslands, but it is sensitive to the choice of stomatal conductance scheme at some savanna and woody savanna sites and at ZM-Mon (Table
Model predictions for the rest of the selected variables are shown in Table
Comparison of selected variables related to simulated ecosystem structure and function between standard LPJ-GUESS and the LSM version at the selected sites. The LSM values are from the CLM/BB and CLM/Med simulations. Gross primary production (GPP), autotrophic respiration (Ra), net primary production (NPP), heterotrophic respiration (Rh), and net ecosystem exchange (NEE) are given in
At PA-SPn, both NPP and
Differences in simulated carbon fluxes between standard LPJ-GUESS and the CLM/BB and CLM/Med runs for the remaining land cover types are summarized in Fig.
The above-described discrepancies between standard LPJ-GUESS and the LSM versions stem from the different physical environments simulated in the models. Calculating assimilation at the newly simulated canopy temperature, rather than the air temperature, can lead to either higher or lower productivity, depending on the optimal photosynthetic temperature ranges of each PFT and the impact of temperature on nitrogen limitation (Sect.
The large relative changes in NEE between simulations result from small discrepancies in magnitude. Figure
Comparison between observed and modelled annual NEE. The symbols indicate averages over sites with the same land cover type. Red triangles correspond to flux tower
Figure
Observed and simulated annual cycles of sensible (
Monthly-averaged diurnal cycle of sensible and latent heat flux at the AU-DaP
Monthly-averaged diurnal cycle of sensible and latent heat flux at the BR-Sa1
At the AU-Gin site, the shapes of the annual cycles of latent and sensible heat are well-reproduced in the simulations (Fig.
At the AU-DaS site (Fig.
Monthly averages of sensible and latent heat at the BR-Sa1 tropical rainforest site show little variability throughout the year (Fig.
At the GF-Guy site, another tropical rainforest, monthly averages of sensible heat are overestimated by
Performance of the CLM/BB
Table
Model performance statistics for simulated hourly (left) and monthly (right) sensible heat fluxes for the CLM/BB and CLM/Med simulations. Bold fonts indicate the model configuration that performed better. The mean and standard deviation of the observed fluxes (
The model tends to overestimate average sensible heat. The hourly and monthly mean biases are non-negative at all sites (except at CG-Tch for monthly fluxes, where it is slightly negative), but normalized RMSE and mean bias are smaller for the CLM/Med run at most sites. The simulations seem to perform comparatively better in grasslands; for the Med simulation, RMSE is between 0.4 and 0.9 of the sample average (hourly fluxes), whereas it is in the 0.6–1.6 range at savanna sites and in the 0.5–2.5 range at forest sites.
The variability of sensible heat flux is also overestimated by the model. In this case, the CLM/Med run performs better than the CLM/BB run for hourly fluxes, but the situation is the reversed for monthly average fluxes. Again, the simulations show better performance at grassland sites; for hourly fluxes, the Med simulation predicts
Model performance statistics for latent heat fluxes are presented in Table
Model performance statistics for simulated hourly (left) and monthly (right) latent heat fluxes for the CLM/BB and CLM/Med simulations. Bold fonts indicate the model configuration that performed better. The mean and standard deviation of the observed fluxes (
Performance of the CLM/BB
Latent heat fluxes tend to be underestimated in forest and savanna sites and overestimated over grasslands. The CLM/BB configuration seems to perform better at grassland sites, while the CLM/Med configuration performs slightly better at forest sites. Results for savanna sites are mixed in terms of RMSE, but the CLM/Med scheme yields somewhat smaller biases.
The variability of simulated latent heat fluxes is larger in the CLM/Med run than in the CLM/BB run. Over
To evaluate the overall performance of the different model configurations, we considered the cross-site averaged statistics of each simulation (Table
Cross-site averaged model performance statistics for simulated hourly and monthly sensible and latent heat fluxes. RMSE and bias are given in
Figure
Average performance statistics for each model configuration, obtained from modelled and measured hourly
Simulated sensible heat fluxes display similar correlation with observations in all runs. The correlation coefficient is very high (
Sensible heat is overestimated in all model configurations; the average bias is always positive, but the Med simulations perform better in this respect. In the case of hourly averages, BB runs show an average bias of
The model also generally overestimates the variability of sensible heat. For hourly fluxes, the standard deviation of the sample is, on average,
Correlations between modelled and measured latent heat fluxes are lower than for sensible heat:
Monthly averages of latent heat simulated by the non-LSM version of LPJ-GUESS show a slightly worse correlation with measurements than the LSM version of the model. The average bias is
In this work we described a number of modifications to the LPJ-GUESS DGVM aimed at making the model suitable for direct coupling with an atmospheric model. The newly incorporated energy balance module resolves the diurnal cycle of energy and water fluxes between the canopy and the atmosphere, as opposed to LPJ-GUESS's daily calculations. Calculating these fluxes on a sub-daily basis is necessary to match the shorter time step at which atmospheric models operate (typically 1 h or shorter, depending on resolution). The original daily PAR absorption calculations were replaced with a more sophisticated radiative transfer scheme by adapting the models of Sellers (1985) and Dai et al. (2004) to LPJ-GUESS's multi-cohort, multi-layer canopy (some differences in PAR absorption calculated by both schemes are shown in the Supplement). This enables the model to simulate the upwelling short-wave radiation flux on sub-daily timescales. Direct and diffuse radiations are treated separately, which allows us to resolve sunlit and shaded leaves in the canopy. This approach offers a reasonable compromise between accuracy of the modelled fluxes and computational efficiency (Wang and Leuning, 1998). The representation of soil physical processes was modified in two ways. Firstly, the original 1.5
The new physical schemes introduced in this work lead to discrepancies between the LSM and standard LPJ-GUESS, which stem from the different physical environments simulated in the models. Calculating assimilation at the newly simulated canopy temperature, rather than the air temperature, can lead to either higher or lower productivity, depending on the optimal photosynthetic temperature ranges of each PFT and the effect of temperature on N limitation (see Sect. 4.2). Canopy temperature also affects autotrophic respiration, while differences in the simulated soil humidity and temperature impact organic matter decomposition rates and heterotrophic respiration. The combination of these effects results in differences in the simulated equilibrium carbon and nitrogen pools (see the Supplement) and ecosystem–atmosphere carbon fluxes.
The new model was evaluated by comparing simulated fluxes of sensible and latent heat with flux tower measurements at 21 FLUXNET sites.
Average correlation and RMSE between observed and simulated sensible and latent heat fluxes for LPJ-GUESS/LSM, CLM 3.0, and CLM 3.5. RMSE values (between brackets) are given in
Effect of temperature (
Daily average climate measured at the four
Sensible heat is generally overestimated by the LSM. Poor performance in sensible heat flux estimation is a common issue of many land surface models
Observed and simulated LAI at the four
One issue in our simulations is the marked underestimation of latent heat flux during extremely dry periods, when the rooting zone is nearly depleted of water available for plant uptake. This, in turn, causes a strong spike in sensible heat (Fig.
We implemented two different stomatal conductance schemes: the Ball–Berry model
The results presented in Sect.
We found that the main reason for this behaviour is the occurrence of higher photosynthetic rates in the LSM simulations due to the mitigation of biochemical N limitation at higher leaf temperatures. Standard LPJ-GUESS uses a daily average of the forcing air temperature as a proxy for leaf temperature in the
At AU-Stp, all three simulations predict much lower productivities. In this case, water availability is the limiting factor. This site receives, on average, considerably less rainwater than the other three
As pointed out in Sect.
The developments presented in this paper will enable us to study feedbacks between the climate and the biosphere using the state-of-the-art DGVM LPJ-GUESS directly coupled to an atmospheric model. Work is in progress regarding the development of a flexible interface to enable such coupling and extending the model's ability to simulate cold-climate ecosystems. More work is also needed to characterize and fully understand the model's response to the switch from using air temperature as a proxy for leaf temperature to simulating leaf temperature explicitly, particularly as these concern the productivity of
During a given time step
The evaporation rates in the above equations can be expressed as follows:
Ecosystem structure.
Energy balance.
Radiative transfer.
Assimilation and stomatal conductance.
Soil physics.
LPJ-GUESS is a worldwide developed and refined DGVM. The model code is managed and maintained by the Department of Physical Geography and Ecosystem Science, Lund University, Sweden. The source code can be made available with a collaboration agreement under the acceptance of certain conditions. For this reason, a DOI for the model code cannot be provided. The code with the augmentations developed for this paper is available to the editor and reviewers via a restricted link, on the condition that the code is used only for review purposes and is deleted after the review process. Additional details and information can be found at the LPJ-GUESS website (
The supplement related to this article is available online at:
AA had the original idea and motivated the development. DMB designed the model augmentations described in this work with input from all the coauthors. DMB implemented the code, ran the experiments, and performed the model evaluation analysis. All the coauthors provided input regarding the analysis of the results and the discussion and helped shape the final form of the manuscript.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research has been supported by the Deutsche Forschungsgemeinschaft (grant no. SCHM 2376/2-1) and the Helmholtz Gemeinschaft (grant no. W3 – Impuls- und Vernetzungsfond). David Wårlind, Stefan Olin, Jing Tang, and Benjamin Smith were financially supported by the strategic research area “Modeling the Regional and Global Earth System” (MERGE). David Wårlind, Stefan Olin, and Benjamin Smith were financially supported by the Swedish national strategic e-science research programme eSSENCE. The article processing charges for this open-access publication were covered by the Karlsruhe Institute of Technology (KIT).
This paper was edited by David Lawrence and reviewed by two anonymous referees.