Process refinement contributed more than parameter optimization to improve the CoLM's performance in simulating the carbon and water fluxes in a grassland
Introduction
Carbon and water exchanges between land surface and atmosphere play an important role in modifying local, regional or global climate (Baldocchi et al., 2001). Land surface models (LSMs) have been widely used to estimate the carbon and water fluxes at various land surface types and different temporal and spatial scales (Dirmeyer et al., 2018; Slevin et al., 2017; Williams et al., 2009). Though substantial progress has been made to improve the accuracy and reduce the uncertainties of LSMs in predicting carbon and water fluxes from land surface during the past few decades, the deficiency and large uncertainties are still challenging for modeling communities, particularly when LSMs were applied in dryland ecosystems or under conditions of water stress (Baker et al., 2008; Li et al., 2014; Li et al., 2012; Saleska et al., 2003). Model defects and uncertainties mainly result from the model structure or process, parameters and input data (Bastrikov et al., 2018; Prentice et al., 2015; Raoult et al., 2016; Williams et al., 2009; Zhao et al., 2012; van Oijen, 2017).
Updating advanced knowledge of plant ecophysiology into LSMs is an effective way to gradually improve the performance and reduce the uncertainty of the model. For radiative transfer within the canopy, the “two-leaf” scheme was introduced into LSMs and substantially enhanced the model's performance to simulate the partitioning of received solar radiation between sunlit and shaded leaves (Dai et al., 2004; Leuning, 1995). For coupling of water and carbon fluxes, several representative stomatal conductance models were developed (Ball et al., 1987; Leuning, 1995; Medlyn et al., 2011; Sperry et al., 2017), including different algorithms to account for water stress, and integrated into LSMs to calculate plant-gas exchange. For the root water uptake function (RWUF), many efforts have been made. In tropical and subtropical forests, the model's performance was largely improved for several LSMs (Baker et al., 2008; Li et al., 2012; Zhu et al., 2017) when hydraulic redistribution (Amenu and Kumar, 2008; Caldwell et al., 1998; Oliveira et al., 2005) was integrated. Many different forms of root water uptake functions have been proposed for different LSMs (Feddes et al., 1978; Lai and Katul, 2000; Li et al., 2006; Zheng and Wang, 2007). For dryland desert shrubs, a non-linear root water uptake function, instead of a linear one, was introduced into the Common Land Model (CoLM) and the problem of underestimation in evapotranspiration was successfully resolved (Li et al., 2013b). Therefore, the description of the RWUF process in a land surface model can better describe the dynamic response of plants to different environmental conditions.
Because of the complexity of LSMs, a model may consist of dozens or hundreds of parameters. The values of these parameters are crucially important to simulate results (Prihodko et al., 2008). Calibrating model parameters against observations is no doubt a productive idea, as the majority of parameters are difficult or impossible to derive from experiments (Larsbo and Jarvis, 2006). Therefore, parameter optimization is used to improve the model's performance. Applying either Kalman filter, Markov Chain Monte Carlo, various Bayesian approaches, variations assimilation, or particle swarm optimization to LSMs is proven to be useful for improving the accuracy of estimation of carbon and water fluxes from different ecosystems (Bateni et al., 2013; Gill et al., 2006; Harrison et al., 2012; Huang et al., 2016; Lakshmi, 2000; Yang et al., 2017). The main applications of those technologies are focused on the optimization of model process parameters, primarily against observations of the carbon and water fluxes (Braswell et al., 2005; Fer et al., 2018). Compared with other algorithms, the advantage of PSO is that it is easy to implement with a set of non-linear equations to find the optimal solution, and the objective function can be easily defined.
The input data errors in both meteorological and land surface states are generally uncertain in LSMs (Post et al., 2018). Temporal-spatial resolution of meteorological forcing provided by tower station and different reanalysis data are main factors to control the uncertainty at site and regional scales (Ménard et al., 2015; Palma et al., 2018; Zhao et al., 2012). High temporal-spatial resolution can better express surface heterogeneity and regional weather changes, to quantify the response of ecosystems to water and carbon fluxes and their feedback to the climate (Medvigy et al., 2010; Paschalis et al., 2015). The imperfect specification of plant functional types (PFTs) is a large source of uncertainty in LSMs that is linked to uncertainties in carbon fluxes at regional scales (Hartley et al., 2017; Quaife et al., 2008). Various land cover maps have been offered to LSMs, but they often perform poorly on the regional or per-pixel scale (Giri et al., 2005). Compared with meteorological forcing, the ancillary data (site-specific measurements of LAI, soil texture, snow-free albedo) have little effects on model results (Ménard et al., 2015).
As a key link of carbon and water fluxes cycles, water use efficiency (WUE) is the ratio of gross primary productivity (GPP) to evapotranspiration (ET) and reflects the characteristic of ecosystem function at leaf, canopy and ecosystem scales (Bonan et al., 2014; Niu et al., 2011). The uncertainties of carbon and water fluxes contribute to the uncertainty of WUE. A number of researchers have found that the uncertainty of WUE is found in the different response of plants to water limitation, model may perform poorly and underestimate WUE in dry conditions (Farooq et al., 2009; Nelson et al., 2018). Therefore, the parameters related to plant water regulation lead to uncertainty in prediction of net primary productivity (Dietze et al., 2014). WUE is closely linked to stomatal conductance. Improvement in the form of stomatal conductance and its parameterization would result in better WUE simulations (De Kauwe et al., 2013; Knauer et al., 2015).
Using better observational techniques and equipment, the uncertainty of the measured input data may be partly decreased, but the input data uncertainties could still propagate into model results (van Oijen, 2017; Dietze, 2017). Compared to input data, model structure and parameter uncertainties have more potential for improvement (Abramowitz et al., 2007; Spadavecchia et al., 2011; Ukkola et al., 2016). Thus, this study focuses on the inappropriateness of the model structure and model parameter optimization strategy. Here, taking Xinjiang grassland ecosystem as the case, the model improvement is quantified by comparing the simulated carbon and water fluxes by CoLM with observed data. In this work we first performed a sensitivity study to compare how the sensitivity of GPP and LE to parameter change when changing the model structure of the RWUF of CoLM. Then, based on the sensitivity results, we utilize PSO algorithm to simultaneously optimize the sensitive parameters of carbon and water fluxes. Finally, we attempt to compare the improvement of model performance by modifying RWUF process and parameter optimization of PSO algorithm. Accordingly, the study will aim at improving the applicability of LSMs and reducing the simulation deviation in arid grasslands.
Section snippets
Model and Parameters
The CoLM was evolved from the National Center for Atmospheric Research (NCAR) land surface model (Bonan et al., 2002), the biosphere-atmosphere transfer scheme (BATS) (Dickinson, 1993) and the Chinese Academy of Sciences Institute of Atmospheric Physics LSM (IAP94) (Dai and Zeng, 1997). Taking account various physical, ecological and hydrological processes, the CoLM is designed to simulate the energy, carbon and water exchanges between land surface and the atmosphere (Dai et al., 2003). The
Parameter sensitivities of the CoLM
By applying the Morris screening method, identified sensitive parameters are different for different versions of the CoLM (V1 or V2) and different model output variables (GPP or LE). For the default CoLM (V1), 9 parameters (Vmax25, effcon, trop, hksati, porsl, gradm, bsw, hlti, psi0) are identified as sensitive ones for GPP (Fig. 3a) and 8 parameters (porsl, trop, Vmax25, effcon, psi0, gradm, hksati, binter) for LE (Fig. 3b). After replacing with a non-linear root water uptake function, the
Discussion
This study examines how the refining model process and applying parameter optimization can improve the performance of the CoLM in simulating carbon and water fluxes in a representative grassland ecosystem of Xinjiang, China. Better representation in model process with a non-linear RWUF is confirmed to be one of the most important mechanisms that governs plant evapotranspiration and photosynthesis in arid grassland ecosystem (Li et al., 2013b). Root water uptake is a dynamic process which is
Conclusions
The ultimate motivation of this research is to examine the relative contribution of refinement of model process or optimization of model parameters to the improvement of the performance of the CoLM in simulating carbon and water fluxes at a representative arid grassland ecosystem. Results indicate that either refinement of model process or parameter optimization significantly improved the model simulations. However, better representing model's root water uptake function in the CoLM contributed
Declaration of Competing Interest
Enclosed is a manuscript entitled “Process refinement contributed more than parameter optimization to improve the CoLM's performance in simulating the carbon and water fluxes in a grassland”, that we would like to submit as a Research Paper for publication in Agricultural and Forest Meteorology. No conflict of interest exits in the submission of this manuscript, and manuscript is approved by all authors for publication. I would like to declare on behalf of my co-authors that the work described
Acknowledgements
This work is supported by National Key R&D Program of China (2017YFA0603603) and the National Natural Science Foundation of China (Grant No. U1403382). We are grateful to professor Lawrence Band, Department of Environmental Science and Department of Engineering Systems and Environment, University of Virginia for his great help and valuable suggestions on this manuscript.
References (102)
Variational assimilation of land surface temperature and the estimation of surface energy balance components
J. Hydrol.
(2013)An effective screening design for sensitivity analysis of large models
Environmental Modelling & Software
(2007)Assimilating passive microwave remote sensing data into a land surface model to improve the estimation of snow depth
Remote Sens. Environ.
(2014)Different representations of canopy structure—A large source of uncertainty in global land surface modeling
Agric. For. Meteorol.
(2019)A comparative analysis of the Global Land Cover 2000 and MODIS land cover data sets
Remote Sens. Environ.
(2005)Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling
J. Hydrol.
(2009)Uncertainty in plant functional type distributions and its impact on land surface models
Remote Sens. Environ.
(2017)Partitioning of evapotranspiration and its controls in four grassland ecosystems: application of a two-source model
Agric. For. Meteorol.
(2009)Assimilating multi-source data into land surface model to simultaneously improve estimations of soil moisture, soil temperature, and surface turbulent fluxes in irrigated fields
Agric. For. Meteorol.
(2016)A simple empirical model of root water uptake
J. Hydrol.
(1989)
The dynamic role of root-water uptake in coupling potential to actual transpiration
Adv. Water Res.
Information content of measurements from tracer microlysimeter experiments designed for parameter identification in dual-permeability models
J. Hydrol.
Representing the root water uptake process in the Common Land Model for better simulating the energy and water vapour fluxes in a Central Asian desert ecosystem
J. Hydrol.
An efficient method for global parameter sensitivity analysis and its applications to the Australian community land surface model (CABLE)
Agric. For. Meteorol.
Optimization of ecosystem model parameters through assimilating eddy covariance flux data with an ensemble Kalman filter
Ecol. Modell.
Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes
Agric. For. Meteorol.
Using high-resolution simulated climate projections in forest process-based modelling
Agric. For. Meteorol.
Sensitivity, uncertainty and time dependence of parameters in a complex land surface model
Agric. For. Meteorol.
Uncertainty in model parameters and regional carbon fluxes: A model-data fusion approach
Agric. For. Meteorol.
Using the particle swarm optimization algorithm to calibrate the parameters relating to the turbulent flux in the surface layer in the source region of the Yellow River
Agric. For. Meteorol.
Systematic bias in land surface models
J. Hydrometeorol.
A model for hydraulic redistribution incorporating coupled soil-root moisture transport
Hydrol. Earth Syst. Sci.
. Seasonal drought stress in the Amazon: reconciling models and observations
J. Geophysical Res.: Biogeosci.
FLUXNET: a new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densitie
Bull. Am. Meteorol. Soc.
A model predicting stomatal conductance and its contribution to the control of photosynthesis under different environmental conditions
Progress in Photosynthesis Research
Land surface model parameter optimisation using in situ flux data: comparison of gradient-based versus random search algorithms (a case study using ORCHIDEE v1.9.5.2)
Geoscientific Model Development
Carbon-water flux coupling under progressive drought
Biogeosciences Discussions
The land surface climatology of the community land model coupled to the NCAR community climate model
Journal of Climate
Modeling stomatal conductance in the earth system: linking leaf water-use efficiency and water transport along the soil–plant–atmosphere continuum
Geoscientific Model Development
Estimating diurnal to annual ecosystem parameters by synthesis of a carbon flux model with eddy covariance net ecosystem exchange observations
Global Change Biol.
Snow temperature changes within a seasonal snowpack and their relationship to turbulent fluxes of sensible and latent heat
J. Hydrometeorol.
Hydraulic lift: consequences of water efflux from the roots of plants
Oecologia
Multiobjective calibration of land surface model evapotranspiration predictions using streamflow observations and spaceborne surface radiometric temperature retrievals
J. Geophys. Res.
A two-big-leaf model for canopy temperature, photosynthesis, and stomatal conductance
Journal of Climate
The Common Land Model
Bulletin of the American Meteorological Society
A land surface model (IAP94) for climate studies part I: formulation and validation in off-line experiments
Adv. Atmos. Sci.
Forest water use and water use efficiency at elevated CO2 : a model-data intercomparison at two contrasting temperate forest FACE sites
Global Change Biol.
Do land surface models need to include differential plant species responses to drought? Examining model predictions across a mesic-xeric gradient in Europe
Biogeosciences
Biosphere atmosphere transfer scheme (BATS) version 1e as coupled to the NCAR community climate model
Technical Report.
A quantitative assessment of a terrestrial biosphere model's data needs across North American biomes
J. Geophysical Res.: Biogeosci.
Prediction in ecology: a first-principles framework
Ecol. Appl.
Verification of land-atmosphere coupling in forecast models, reanalyses and land surface models using flux site observations
J. Hydrol.
Plant drought stress: effects, mechanisms and management
Agron. Sustainable Dev.
Simulation of field water use and crop yield
Soil Sci.
Linking big models to big data: efficient ecosystem model calibration through Bayesian model emulation
Biogeosciences
Multiobjective particle swarm optimization for parameter estimation in hydrology
Water Resour. Res.
Multi-objective parameter optimization of common land model using adaptive surrogate modeling
Hydrol. Earth Syst. Sci.
Multiobjective adaptive surrogate modeling-based optimization for parameter estimation of large, complex geophysical models
Water Resour. Res.
Quantifying the change in soil moisture modeling uncertainty from remote sensing observations using Bayesian inference techniques
Water Resour. Res.
Joint structural and physiological control on the interannual variation in productivity in a temperate grassland: a data-model comparison
Global Change Biol.
Cited by (8)
An optimal ensemble of the CoLM for simulating the carbon and water fluxes over typical forests in China
2024, Journal of Environmental ManagementCalibration and validation of phenological models for Biome-BGCMuSo in the grasslands of Tibetan Plateau using remote sensing data
2022, Agricultural and Forest MeteorologyCitation Excerpt :Based on SOS and EOS derived from snow-free NDVI, parameter optimization was employed to improve the phenological model's performance. Optimization of the PSO method improved the accuracy of carbon and water fluxes in grassland ecosystems for the CoLM land surface model (Li et al., 2020b). PSO offers the benefit of being easy to implement for optimization of non-smooth and non-linear functions, and its objective function can also be easily defined (Eberhart and Kennedy, 1995; Li et al., 2020b).
Interannual variations in GPP in forest ecosystems in Southwest China and regional differences in the climatic contributions
2022, Ecological InformaticsCitation Excerpt :All these models are capable of producing synthetic images with various advantages and disadvantages. Many researchers have evaluated the effects of individual sources (e.g., model structures, model parameters, or input data), which induce uncertainties in GPP modeling (Huang et al., 2021; Li et al., 2020b; Xiao et al., 2014). However, few studies have evaluated the impact of different climate zones and plant functional types (PFTs) on GPP uncertainty estimates under different models (Shao et al., 2014; Zhang et al., 2019a, 2019b).
Inter-comparisons of mean, trend and interannual variability of global terrestrial gross primary production retrieved from remote sensing approach
2022, Science of the Total EnvironmentCitation Excerpt :Using LSMs or LUE models, current estimates of mean annual GPP of the global terrestrial ecosystems has a wide range from 90 to 175 Pg C year−1 (Wang et al., 2012; Wang et al., 2021b; Zheng et al., 2020). The wide range of estimates are mainly resulted from model structure or process, i.e., which variables are considered to impact GPP and how their relationships are formulated in responding to a specific variable, and the variations in model parameters and input data (Li et al., 2020; Williams et al., 2009; Zhao et al., 2012). For a long term period, the trend in terrestrial GPP and its drivers are particularly important (Sitch et al., 2013), as this reflects the ability of terrestrial ecosystem in sequestrating CO2 from the atmosphere.
Performance of the ecosystem demography model (EDv2.2) in simulating gross primary production capacity and activity in a dryland study area
2021, Agricultural and Forest MeteorologyCitation Excerpt :Consequently, the higher contribution of model structure to the uncertainty is not surprising. Our finding is in agreement with a recent study which showed that in a dryland study area, model structure contributed more than the parameters to the improvement of DGVMs (Li et al., 2020). One of the key factors which is directly related to the EDv2.2 model structure is the lack of ecosystem heterogeneity represented in the model.