The integrated Earth system model (iESM) has been developed as a new tool for projecting the joint human/climate system. The iESM is based upon coupling an integrated assessment model (IAM) and an Earth system model (ESM) into a common modeling infrastructure. IAMs are the primary tool for describing the human–Earth system, including the sources of global greenhouse gases (GHGs) and short-lived species (SLS), land use and land cover change (LULCC), and other resource-related drivers of anthropogenic climate change. ESMs are the primary scientific tools for examining the physical, chemical, and biogeochemical impacts of human-induced changes to the climate system. The iESM project integrates the economic and human-dimension modeling of an IAM and a fully coupled ESM within a single simulation system while maintaining the separability of each model if needed. Both IAM and ESM codes are developed and used by large communities and have been extensively applied in recent national and international climate assessments. By introducing heretofore-omitted feedbacks between natural and societal drivers, we can improve scientific understanding of the human–Earth system dynamics. Potential applications include studies of the interactions and feedbacks leading to the timing, scale, and geographic distribution of emissions trajectories and other human influences, corresponding climate effects, and the subsequent impacts of a changing climate on human and natural systems. This paper describes the formulation, requirements, implementation, testing, and resulting functionality of the first version of the iESM released to the global climate community.
As documented extensively in the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (AR5)
Integrated assessment models (IAMs) are the primary tools for describing the human components of the Earth system, the sources of greenhouse gases (GHGs) and short-lived species (SLS) emissions, and drivers of land use change. Earth system models (ESMs) are the primary tools for examining the climatic, biogeophysical, and biogeochemical impacts of changes to the radiative properties of the Earth's atmosphere. These two modeling paradigms developed largely independently of each other and their interactions have historically been relatively simplistic. Typically, projections of GHGs and SLS emissions have been produced by the human system components of IAMs, archived in databases, and used by ESMs to produce projections of climate and altered biogeophysical processes.
As IAMs have become more sophisticated, they have gradually expanded
to incorporate agriculture, land cover and land use change, and
representations of the terrestrial carbon cycle because processes in
those sectors affect anthropogenic emissions of GHGs and SLS in
important and unavoidable ways. Many studies
In conjunction with the World Climate Research Program (WCRP) Fifth
Climate Model Intercomparison Program (CMIP5) and the IPCC AR5, these two modeling communities have engaged in
an unprecedented degree of collaboration to ensure that the products
of the IAM community meet the needs of the climate and Earth system
modeling communities
But as sophisticated as this interaction has become, it is still
a one-way transfer of information from IAMs to ESMs
(Fig.
Illustration of the one-way coupling from the human to the climate system used in prior simulations of global environmental change.
It is therefore clear that future work must enable the processes in these sectors to interact with each other and the climate system rather than to remain as one-way transfers of information. If ESMs are to include better representations of the feedbacks of climate change on agriculture, land use, land cover, and terrestrial carbon cycle, as well as other human systems such as energy and the economy, then they will need the ability to incorporate the human system directly. Heretofore the tools have not existed for a fully consistent representation of the combined evolution of these two systems.
In order to advance beyond this paradigm, we have developed a new model framework, the integrated Earth system model (iESM). The goal is to create a first-generation integrated system to improve climate simulations and enhance scientific understanding of climate impacts on human systems and important feedbacks from human activities to the climate system. The first version of the iESM described in this paper is designed to address three major science questions: (1) is the present CMIP5 “parallel process” approach to climate assessment adequate? (2) Will human activities affect local and regional climate on scales that matter? (3) Will climate change itself affect human decision-making and its implications for biogeochemical and biogeophysical processes at global scales?
The iESM is a new configuration of models previously operated
separately. The iESM includes the human system components of an
IAM called the global change assessment model (GCAM)
Schematic of the integrated Earth system model (iESM) showing its major component models GCAM, CESM, and GLM as well as the two-way connections between these models.
The iESM includes both one-way and two-way communication of fluxes and
feedbacks among the components of the energy and land use systems
from GCAM, as well as the incorporation of their physical
consequences for both biogeochemical and physical fluxes in CESM. This
allows for the investigation of the degree to which this linkage may
change the evolution of the climate system over decades to a century.
We have used the iESM to investigate
the climate impacts on human systems and important feedbacks from
human activities to the climate system. The iESM results on impacts
and feedbacks are described in a series of earlier and companion
papers
This paper describes the scientific rationale for the construction of
the iESM (Sect.
Climate change can influence energy demand, supply, and production in
several major areas. Energy demand for adaptation and mitigation
measures may also increase under climate change
Climate change can have important impacts on building energy systems
through decreased heating and increased cooling. Previous studies are
limited in addressing the effect of a changing climate on building
energy demands while simultaneously considering other energy sectors
in the underlying human systems. In recent years, the impacts of
climate change on building energy use have been evaluated using IA
models by constructing estimates of heating and cooling degree days
from air temperature outputted from climate models
Renewable energy plays an important role in the energy system at the regional
and global levels, and it can be influenced by climate change to
a large extent. In current IA modeling efforts, the availability of
renewable energy (i.e., wind, solar energy, and hydropower) and its
economic potential are either modeled according to the historical
condition
Climate change also has important impacts on energy production,
especially thermal power plants, which are influenced by the
temperature of water used for cooling
The starting point for the team's development efforts was version 1.0
(now 1.1) of the CESM. CESM is a community
code and may be downloaded from the
The CESM uses a flexible coupler that couples the atmosphere, ocean, land, and ice component models. Components often use different grids, and the coupler performs the necessary interpolation of fluxes and state variables. The CESM system comprises the Parallel Ocean Program, version 2 (POP), the CLM, version 4.0 (CLM 4.0), the Los Alamos sea ice model (CICE), the Community Atmosphere Model, version 5 (CAM), and the Community Ice Sheet Model (CISM). POP and CICE are finite volume codes with semi-implicit and explicit time integration and are implemented on logically Cartesian meshes that are stretched to embed polar singularities in land regions and thereby remove these singularities from computation.
The CAM model has flexible formulations for atmospheric dynamics, and it has recently transitioned to the spectral finite element method coupled to an extensive suite of sub-grid physical parameterizations in its standard configuration. CAM runs on unstructured quadrilateral grids. The CLM contains a suite of column process parameterizations running at each grid point with no communication between grid points. CLM 4.0 represents surface and subsurface water, energy, carbon, and nitrogen dynamics with a nested hierarchical sub-grid treatment that allows glaciers, lakes, urban areas, agricultural fields, forest, grassland, and shrubland to share space on each grid cell. Incident radiation is intercepted in a two-layer canopy, with vegetation, soil, snow aging, and black carbon impacts on albedo. Subsurface processes include vertically resolved biogeochemistry, options for carbon and nutrient cycle parameterization, and recently improved treatment of wetlands and permafrost dynamics. CISM is based upon the Glimmer model, an open-source (GNU Public License, GPL), three-dimensional, thermomechanical ice sheet model designed to be interfaced to a range of global climate models.
In the fully coupled configuration, the CICE and POP component models
run with a nominal displaced-pole grid spacing of 1
A mechanistic representation for the influence of land use and land cover
change (LULCC) on carbon, nitrogen, water, and energy cycles was
developed and implemented for the CMIP5 land use harmonization
GCAM is a dynamic-recursive model with technology-rich representations
of the economy, energy sector, land use, and water linked to
a reduced-form climate model that can be used to explore climate-change mitigation policies, including carbon taxes, carbon trading,
regulations and accelerated deployment of energy technology
Regional population and labor productivity growth assumptions drive the energy and land use systems employing numerous technology options to produce, transform, and provide energy services as well as to produce agriculture and forest products and to determine land use and land cover. The GCAM model takes population, gross domestic product (GDP), technology efficiencies and costs, and certain policies as external boundary conditions and determines regional energy, land use, and emissions distributions as a result. GCAM, like all IAMs, is calibrated to a base year (e.g., 2005) to reflect differences in resource endowments, technology history, and consumer tastes across regions.
Using a run period extending from 1990 to 2100 at
5-year intervals, GCAM has been used to explore the potential
role of emerging energy supply technologies and the GHG
consequences of specific policy measures or energy technology adoption
including CO
GCAM is a Representative Concentration Pathway (RCP)-class model. This
means it can produce the emissions and land use outputs necessary to
force a full Atmosphere-Ocean General Circulation Model (AOGCM) or ESM as in the CMIP5 process
For iESM, the time step of GCAM was reduced from 15-year to
a 5-year standard with flexible time step capability. This
capability is important for scale consistency and compatibility with
CESM code. In addition, the land component, which simulates supply of
land products (food, energy, fiber), was completely reformulated to
follow functional forms that define productivity as a function of
geographic location, climatic conditions, and inputs, and thus made
more consistent with physical earth system parameters
The GLM is a tool for computing annual,
gridded, fractional land use states and all underlying land use
transitions, including the age, area and biomass of secondary
(recovering) lands and the spatial patterns of wood harvest and
shifting cultivation, in a format designed for inclusion in Earth
system models
GLM was selected as the primary tool to provide harmonized land use
data sets
For use in iESM, GLM was modified to use GCAM data on a 5-year time step and to accept data partitioned by GCAM's 151 agri-ecological zones (AEZs) instead of GCAM's 14 socioeconomic regions. In addition, GLM was altered to use the forest area data from GCAM and to spatially rearrange agricultural area within each AEZ to match potential forest area changes from GCAM.
To ensure that the iESM is reliable, flexible, and extensible, its technical implementation follows from an extensive set of requirements that are detailed below.
The primary goal of the development is to implement the iESM as an extension of the CESM to include a human-dimension component. This requirement implies that the IAM is treated as a new component in CESM and the protocols applied to the five existing components are adopted for the human component as well. To conform with these protocols, the human-dimension component has been integrated into CESM's software environment, including CESM's configure and build procedures, execution protocols, input and output conventions, and regression testing procedures. The execution protocols include CESM's procedures for synchronizing the coupling and time stepping of its various components and for exchange of fields among these components that conform with the conservation laws (e.g., conservation of mass) governing the dynamic evolution of the whole system.
The developers have also ensured that the iESM conforms to CESM's standards for repeatable experiments, including exact restarts and use of machine-independent representations for the initial, boundary, and restart data sets. CESM has adopted the Network Common Data Format (NetCDF) for these data sets to utilize its features for representation of numerical fields that can be transparently exchanged across computational platforms. This is complemented by the requirement that iESM conform to CESM's standards for hardware and software portability. This requirement helps ensure that experiments with iESM are, in principle, strictly repeatable assuming that the underlying software and hardware configuration has been validated by the CESM project. In practice, a precise description of the boundary and initial conditions, together with a detailed description of the model and its functionality, are needed to attain experimental reproducibility. To address this need, it follows that the functionality of the human-dimensions component should be clearly and comprehensively documented. The documentation should encompass individual pieces such as GCAM, GLM, the land use translator (LUT) code, as well as the pre-/post-processing code which operates on the data exchanged within the human-dimensions component.
The second principal goal is to incorporate and extend CESM's flexible modes of execution to iESM. The flexibility has two main dimensions: first, the trade-off between the physical completeness and complexity of the model and its execution speed; second, the equivalence between two-way communication between components with the introduction of feedbacks through their joint interaction. The first type of flexibility is realized by incorporating several versions of each critical component that range from very simple to very complete representations of the component dynamics, with a corresponding range from inexpensive to intensive computational resource demands. The second type of flexibility is implemented by introducing versions of each component that either produce the same output state (e.g., a climatology read from data file) regardless of the input state, or that compute an output state based on the input state combined with its evolution equations. The omission or inclusion of two-way communication corresponds to the omission or inclusion of feedbacks between the given component and the rest of the model system.
Both types of flexibility are realized by incorporating three basic versions of each component known as “stub”, “data”, and “active” versions. The “stub” version is used primarily for automated testing of the system integration and performs some very rudimentary housekeeping functions in response to commands from the integration layer of the whole CESM. The “data” version produces a time-evolving state through spatial and/or temporal interpolation applied to a fixed time-dependent input read from data files. The same state is reproduced regardless of the evolution and dynamics of the remainder of the coupled system. This version is computationally inexpensive but does not include the two-way feedbacks between the given component and the rest of the system present in the real world. The “active” version produces a time-evolving state governed by its initial conditions, a representation of the fundamental dynamical equations that pertain to that component, and the boundary conditions supplied by the rest of the coupled model system. This version is computationally intensive but includes the two-way feedbacks present in the real world.
To conform with this protocol, the iESM includes the stub and data version of the human-dimensions component, as well as the fully interactive assessment model GCAM. The stub and data versions are automatically tested to ensure that they are integrated and operating correctly using the same general test procedures applied to the existing components of CESM.
CESM utilizes a set of standard protocols to implement bilateral exchange among components of the coupled system, and these protocols have been adopted for internal communications within the human component as well as including GCAM, GLM, the LUT code that prepares GLM output for input into CLM, and the associated interfaces. These protocols ensure that the modes of interaction and exchange among components are visible, reproducible, flexible, and extensible.
The visibility follows from the requirement that all fields are exchanged through a single, top-level, standardized communication mechanism. This mechanism is capable of recording all information input to and output from all model components, together with the operations performed by the coupling layer to enable the exchanges. This capability also ensures that the interactions are strictly reproducible, since all exchanges are managed and recorded by one standardized communication mechanism.
This mechanism can be configured at run time to add arbitrary numbers of fields to be exchanged among any given pair of components. This ensures that the communication protocol can support increasingly complete and complex interactions among components using the same well-tested framework, without the need for lengthy modifications to the underlying software.
iESM has adopted these conventions for exchanging information to integrate the functional parts within the human-dimensions component and, ultimately, to couple the human-dimensions component to other components in CESM. In the first implementation, the data passed between the human-dimensions components and the rest of CESM are exchanged using data files to minimize the modifications to the existing CESM components. However, these data exchanges can be readily upgraded to the standard top-level interfaces, internal memory, and message passage adopted by the rest of CESM.
This solution automatically includes provisions for exchanging additional data, in particular the exchange of more or all of the forcing agents covered by the RCP handshake protocol (tntcat.iiasa.ac.at:8787/RcpDb/). The information exchanged at the interfaces between the human component and the rest of CESM depend on the precise experimental configuration. However, the interfaces themselves are invariant under changes in configuration to guarantee that a single set of communication software can be thoroughly and repeatedly tested and validated.
The IAM solves for the evolution of the human system using a fundamental assumption of market quasi-equilibrium, namely, that the inputs and outputs into energy generation, food production, and land resources are balanced on sufficiently large spatial and temporal scales. The length and timescales required for the market equilibrium assumption to hold are orders of magnitude larger than the corresponding scales used to solve the equations of motion for physical, chemical, and biogeochemical processes in the Earth system.
This disparity introduces a requirement on the design of the iESM, namely, to implement a flexible and extensible mechanism to handle differences in spatial and temporal resolutions between the human and physical components. To meet this requirement, iESM should include capabilities in temporal interpolation or accumulation (e.g., time averaging, or other operations) in all the interfaces depend on the ratio of time steps between the transmitting and receiving components linked by the interface. Similarly, spatial interpolation or accumulation should be included with the recognition that some preprocessing may be needed to prepare input data files to manage regridding.
These capabilities are consolidated into the interfaces among the human component and the rest of the CESM system to avoid “hard wiring” any assumptions about gradations in resolutions into the components themselves. The efficient exchange of data across different spatial grids is highly contingent on efficient communication between components and within a single component on highly distributed and massively parallel supercomputers. The interfaces are therefore based upon a common foundation of communication infrastructure that has been optimized to maximize computational throughput. In turn, the exchange of data between components operating on very different time steps introduces a strong dependency on the time management procedures for the whole coupled system. This dependency has been satisfied by completely prescribing the sequence of component execution, the interlaced calls to the interfaces, and the interpolation–accumulation operations in each interface call.
While CESM is designed for hybrid execution in any combination of serial and/or parallel execution of its various components, in the initial version of iESM the human component is run in serial mode. This mode of operation is necessitated by the multi-year time step of GCAM. Since the version GCAM used in iESM runs as a single-threaded application while the rest of the CESM is inherently multi-threaded, the processor elements devoted to the non-human components are idle while GCAM is run for a single time step. This introduces the risk that iESM utilizes computational resources much less efficiently than the parent CESM. It is therefore necessary to evaluate the relative cost of the human-dimensions component to ensure it is not a performance or memory bottleneck and re-factor or parallelize code as needed. Fortunately, the overall CESM performance is only marginally impacted by the introduction of this serial code.
GCAM and GLM, like the other components of CESM, are research codes
and are therefore under continual development and extension by their
primary developers and by the wider GCAM and land use communities.
Recent developments include significant new capabilities directly
relevant to studies of human–Earth system interactions, for example
the introduction of supply and demand for water resources
This design introduces several subsidiary requirements for the input to and output from GCAM and GLM. First, GCAM's and GLM's native input and output procedures must be extended as needed to perform file I/O in stand-alone mode to exchange data that are compatible with CESM. This in turn requires introducing input and output interfaces into GCAM and GLM that generalize the methods for information exchange to include message passing. As a result, the results from GCAM and GLM are indistinguishable whether using files or inline communication techniques to exchange data with the rest of iESM.
One of the challenges in constructing iESM is the complexity of the historical land use and land cover data required for the downscaling operations performed by GLM. In order to create a much simpler and more robust run-time environment for the GLM component, several important modifications are necessary. These include collating and converting the numerous input and output data sets into a much small number of NetCDF files. It was also helpful to standardize GLM's control interface to provide a simple and robust way to manage GLM functionality. To reduce the considerable demands for memory from GLM, it was necessary to re-factor GLM's data and control structures as needed to reduce its large in-memory footprint. Because CESM must meet a requirement for exact (bit-for-bit) restarts, it was necessary to extend GLM's functionality to add a restart capability.
To the extent feasible, it would be advantageous to have the coupled iESM reproduce the offline-coupled implementation using separate models. To meet this requirement, it is necessary to construct tests ensuring that the data exported by each interface agrees with the corresponding information exchanges in the offline-coupled implementation to the precision of the stand-alone implementation. In turn, these tests are based upon and therefore require a core level of state output and diagnostics to allow iESM to be validated against relevant observations and documented CESM/GCAM/GLM control runs.
The first phase of iESM code development was designed to update and codify the experimental protocol from CMIP5 to incorporate land use change and emissions of GHGs and SLS from GCAM into CESM, such that the models exchanged information at each time step rather than as a single, full-century pass at the start of the model future period (2005). The software development team acquired the GCAM and GLM model codes and incorporated them into the land node of CESM through a new component, the integrated assessment component (IAC). The IAC is currently visible only to the land model when run in iESM mode and drives prognostic land use change. Because the functionality of GCAM–GLM is encapsulated within a CESM component, it can also be replaced by a data model, enabling testing with a range of IAMs.
Code modifications were made to GCAM such that the model looks to CESM for instruction on when to begin each new time step. Thus, the first version of the coupled model operates by GCAM projecting land use, then CESM projecting climate and ecosystem change and returning productivity information to GCAM, which then incorporates that information into the land use decisions for the next time step.
The code has been tested and is running on leadership-class computing facilities at ORNL (Oak Ridge National Laboratory) (the Titan Cray XK7) and NERSC (National Energy Research Supercomputing Center) (the Hopper Cray XE6 and Edison Cray XC30) and has also been tested and configured on the DOE IARP (Integrated Assessment Research Program) cluster at PNNL/UMD (Pacific Northwest National Laboratory/University of Maryland) (Evergreen). The iESM code has also kept pace with current CESM versions, and was most recently updated (in summer 2013) to run with CESM 1.1.2, the most recent CESM release with a full carbon cycle spin-up available.
Scientific challenges were encountered in the design of the coupling between the CESM and the IAC component, specifically with regard to faithfully representing CESM's land productivity passed into the IAC as well as capturing the land use change being returned. These challenges were identified and solved through a series of soft-coupled runs where the project team ran each model one time step forward at a time and passed model output between them, as well as a series of offline, CLM-only runs with the IAC enabled. In this fashion, the coupling steps were refined while the software development was under way. This first development phase focused on the land use change components of the models. In parallel, algorithms to downscale GCAM GHG and SLS emissions have been developed and tested, and the code has been transferred to the development team.
The IAC is implemented like a standard component of CESM. The IAC component has stub, data, and active versions called SIAC, DIAC, and GIAC, respectively, that provide a range of functionalities and capabilities for the IAC component. The active IAC version (GIAC) is fully prognostic and runs the full suite of IAC subcomponents to produce dynamically varying land use/change data using carbon feedback scalars from CLM. The data component (DIAC) replaces the active GCAM/GLM sub-models with data derived from an offline IAM/GLM control run. It is currently used for testing and model spin-up, but in principle it could be used to force CESM with an arbitrary scenario, for example one of the three CMIP5 RCPs generated by an IA model other than GCAM. The stub component (SIAC) serves the same purpose as a CESM stub model, namely, to serve as a placeholder to satisfy interface requirements when the active or data component is not being run. The stub IAC is the default mode for CLM, which makes it 100 % bit-for-bit backward compatible with the current CLM.
Like other CESM components, IAC has routines to initialize its state, execute by evolving forward in time, and complete its operations by communicating its new state and generating history and restart (check-point) files. While these routines do not satisfy all aspects of the current CESM interface standard, they could be readily modified to do so in the future. The checkpoint/restart mechanism built for the IAC meets the CESM requirements for exact restarts to facilitate long integrations of the model system. Following the template of other CESM components, the IAC has a built-in clock, a top-level interface that mimics a CESM component, a centralized collection of control information implemented via a standard Fortran namelist, and a set of clock, grid, control and field parameters defined in a shared module for query by and exchange with other parts of the model system. All the coupling within the IAC is done via internal memory.
While the IAC was initially implemented as a separate component in CESM, we have placed the IAC component beneath the land model, since the all the coupling in the initial version of iESM would involve the CESM land component. Because we are using a mature coupling strategy, we can easily reposition the IAC component as needed in the future. But for the moment, the IAC is implemented as an option in CLM, and therefore the IAC model resides in its own subdirectory within the main code base for CLM. The stub, active, or data mode of IAC is set via the CESM configuration files. When this mode is set to stub, the results from iESM are identical at the bit-for-bit level with the corresponding version of the conventional CESM. All the input data sets and namelist parameters for the IAC are set by enhanced versions of the namelist generation procedures for CLM.
In the current iESM, the IAC is built as part of the compilation of the CLM code. The procedure that builds CLM calls scripts that build the IAC model. The IAC compilation is done for the stub, data, or active version of the IAC model depending on the mode specified by the user. Most of the IAC code is written in Fortran 90 or C, and leverages the CESM makefile. When the active IAC model is specified, GCAM is built via GCAM's build scripts that have been modified slightly to support coupling while retaining support for GCAM's implementation in C++. Currently, coupling between the IAC and CLM components is done via data files to leverage current CLM capabilities and to minimize changes to CLM. The IAC reads data from CLM history files at the start of a time step and writes data to a time-varying surface data set at the end of the IAC time step. Both sets of data evolve in time as the coupled system advances.
The IAC component consists of five subcomponents, including the models
GCAM and GLM and the interfaces IAC2GCAM, GCAM2GLM, and GLM2IAC
between these models and the rest of the IAC component
(Fig.
Schematic of the iESM interfaces among GCAM, GLM, and the CLM component of CESM. Several of these interfaces are unused in the initial implementation of the iESM.
Sequence of operations and information exchanged during the time stepping of the iESM. Years are denoted in red, spatial interpolation with the grid patterns, time interpolation and time stepping with clocks, and model components with boxes.
The IAC2GCAM interface translates and re-maps gridded information from
CLM on its terrestrial carbon state into regional scaling factors
(scalars) for crop yields and ecosystem carbon densities used by the
agriculture and land use module internal to GCAM
Fields input by IAC2GCAM.
The GCAM model produces worldwide land use projections incorporating
information about demographics, economics, resources, energy
production, and consumption
(Sect.
The GCAM2GLM interface serves to allocate GCAM output from 151 land
units to the 0.5
In terms of its interactions with the rest of the current IAC components,
the GLM model converts the annualized fractional land use states
output by GCAM2GLM into gridded data sets suitable for input into CLM,
while also computing the spatial pattern of wood-harvest area and the
area of natural vegetation occupied by both primary and secondary
vegetation. GLM converts the GCAM2GLM output
data into a variety of fields on its native half-degree grid, nine of
which are currently utilized in iESM including five wood-harvest
categories (Table
Integration into CESM has required extensive modifications to GLM, including the redesign of data structures to reduce memory requirements and to accommodate control by CESM of its temporal evolution. Other modifications include the addition of restart functionality, the introduction of a control interface, the conversion of all boundary data into NetCDF, and the provision for routing all input and output through the calling interface.
Fields output by GLM.
The GLM2IAC interface is tasked with converting the harmonized outputs
of GLM to time-varying data sets for land cover and wood-harvest area
in CLM's native input format. The translation of GLM state and
harvest variables to CLM land cover is based on code
Although CLM and the rest of CESM require minimal modifications to incorporate the IAC component, CLM was modified to permit updates to its time-varying input surface data sets after its initialization phase. This modification required introducing some changes in order to reread the time axis of the dynamic surface data set during the execution phase of CLM.
The IAC component advances in 1-year time steps and is called at the start of each calendar year. During this call, every sub-subcomponent in the IAC component is executed in order to prognose the time-varying CLM land surface data sets starting from the current CESM time step and ending 1 year into the future. To accomplish this, the IAC calculates the land surface for the time step 1 year in advance, then CLM interpolates between the current and future land surface at its native 30 min time step. In between the yearly IAC time steps, the IAC component is called monthly from CLM to create an annual average of CLM NPP (community land model net primary productivity) and HR (heterotrophic respiration) values. The GCAM subcomponent can be integrated using either one or three sequential 5-year time steps. The default is to use a 5-year time step and interpolate the yearly data needed for the rest of the IAC subcomponents. Prior to each GCAM call, the IAC computes the carbon scalars that constitute the feedback between CESM and GCAM.
Several technical requirements and protocols specific to large climate codes and CESM had to be introduced with the IAC component. The IAC component is bit-for-bit reproducible when rerun, and it restarts exactly from check-point files generated by previous runs. The IAC component is included in the CESM code repository and is tagged regularly in order to track code versions. A specific numerical experiment using the IAC in CESM can be described by specifying the model tag, the compset (which determines the model components), the grid, and a set of plain-text files that specify the features and input setting for the CESM components. The CESM configuration scripts have been augmented for iESM to include new compsets and new XML environment variables that specify items like the IAC mode (stub, data, active). The scripts have been further enhanced to incorporate several new libraries required by GCAM to support the open-source Berkeley DB XML (Oracle) database package with XQuery Access. These libraries include Berkeley DB XML, Berkeley DB, XQilla, and Xerces C++, which must be installed before the active IAC component can be run within CESM.
To facilitate running the IAC with different CLM grids, many of the IAC settings are specified via namelist or read from files specified at run time. All the output data are written in NetCDF to ensure portability across computing platforms and to exploit the self-documenting features provided by this format. All variables are given explicit types, real variables are assigned to a type of double precision wherever possible, and the Fortran code complies with the CESM coding standard and is written in Fortran 90. Because the GLM and GCAM are written in C and C++, Fortran–C interfaces have been implemented in several parts of the IAC component.
One of the core requirements of the iESM design is to reproduce simulations conducted with the offline-coupled version of the same codes. Satisfaction of this requirement implies that the online-coupled simulations with iESM would be statistically indistinguishable from the offline-coupled simulations. Since the offline-coupled experiments have been configured to emulate the large number of simulations conducted using the same suite of codes for the CMIP5, successful reproduction of the offline-coupled runs would mean that the iESM user community could employ the large literature analyzing the CMIP5 runs to understand the baseline (or control) climatology and climate dynamics of iESM. Since iESM includes a variety of bug fixes and enhancements relative to the offline-coupled model configurations, the emulation will be only approximate.
The tests to verify the degree to which iESM reproduces the
offline-coupled model have been conducted in three stages. First,
with the exception of GLM, each component in iESM has been checked
separately to show that, given the same input, the output of that
component matches that of the corresponding component in the
offline-coupled system to within the limits of machine precision
(Sect.
These tests consist of comparing offline runs of each subcomponent of the offline-coupled implementation and online runs of iESM using the same forcing. To facilitate these tests, the iESM designers have allowed each of the components (GCAM, GLM, LUT, GCAM2GLM, etc.) to continue writing the state and diagnostic files that were output in the original offline models. Additionally, the data flowing between each of the component models were captured and written out in double precision NetCDF format. The ease of tracking the data flowing between each component as well as the ability of the component developers to continue using trusted analysis tools have allowed the iESM team to verify that the results produced by the offline and online versions of each subcomponent are, in general, identical to within the machine roundoff precision of the underlying calculations.
Once the individual pieces were validated, the team forced the IAC with prescribed CLM history output and compared the offline-coupled runs to the online runs with identical forcing. These simulations were designed to test that the feedbacks from the ESM to IA sub-systems of the iESM are as identical as possible between the offline-coupled and online versions. Both the offline-coupled and online IAC systems were subjected to the same external forcing from CLM, and the resulting dynamic surface data sets from both IAC versions were then compared. The team was able to verify that the results were identical to single-precision roundoff.
Finally, this test has been repeated with consistent and uniform time synchronization between CLM and IAC. Since the original test (described above) was forced with prescribed data, it did not ensure that the the temporal interactions between CLM and IAC were correctly reproduced in the online version relative to the implementation of the same interactions in the offline-coupled version. The team enhanced iESM to guarantee the same temporal interaction between CLM and the IAC in the two versions and also provided an alternative, reduced length GCAM time step of 5-year duration. The iESM also passed this more realistic test of its normal mode of operation, one in which there is cyclic two-way interaction between CLM and IAC coordinated by the master timing mechanism of the whole online model system.
In order to test whether simulations from the offline-coupled and
online iESM are statistically indistinguishable, we conducted a pair
of integrations with these two versions of iESM based upon the RCP4.5
scenario. In these simulations, the copies of GCAM in both the
offline-coupled and online versions are subjected to the same
exogenous drivers and policy specifications that were used to create
the original RCP4.5 scenario used in CMIP5. The two runs produce
nearly identical future trajectories for global-mean surface air
temperature. To formally evaluate this, we projected the time- and
space-varying surface air temperature trajectories from these two
simulations onto the spatial warming fingerprint
Projection coefficients for the online-coupled and offline-coupled
model implementations for a pair of equivalent scenarios based on RCP4.5. The
coefficients are derived by projecting the spatial pattern of annual mean
surface air temperature onto the “fingerprint” of the surface
air warming trend derived from the RCP8.5 ensemble mean. The fingerprint is
taken to be the first empirical orthogonal function of the 96-year time
series of RCP8.5 annual mean surface temperatures, scaled so that its mean
value is 1
Several extensions to the iESM are already under development. First,
capabilities have been developed for energy-sector components of the
model to respond to climate change. These capabilities include
developing the building sector so that demands for energy for heating
and cooling are sensitive to temperature change
The first version of the iESM, however, already provides a significant new capability to the climate community. iESM represents the first coupled treatment of the human–climate system based on an IAM and ESM that both contributed to the most recent IPCC and US National Assessments and that support international communities of developers and investigators in integrated assessment and climate science. While iESM is designed to exploit the full capabilities of its parent models, it can be readily simplified and expanded due to its flexible and extensible architecture. The simplifications include inclusion or exclusion of human components, as well as potentially drastic reductions in the complexity and computational burden of the Earth system components by use of CESM's data modes. This capacity for faster execution helps ensure that iESM can be used to explore a large range of future scenarios of climate adaptation and mitigation in both a thorough yet economical manner. The possible expansions include inclusion of other IAMs that conform to the RCP handshake protocol, incorporation of additional forcing agents from the human system that can alter the climate system, and extension to simulate the supply and demand of other major resources, e.g., water, that interact strongly with natural and societal processes. This capacity for extensibility helps ensure that the iESM can and will continue to evolve with the state of integrated assessment and climate science.
Copies of the code for the iESM version 1 are readily available upon request from the corresponding author.
This research was supported in part by the Director, Office of Science, Office of Biological and Environmental Research of the US Department of Energy under contract no. DE-AC02-05CH11231 to the Lawrence Berkeley National Laboratory as part of their Earth system modeling program. The authors used resources of the National Energy Research Scientific Computing Center (NERSC), also supported by the Office of Science of the US Department of Energy, under contract no. DE-AC02-05CH11231. The CESM project is supported by the National Science Foundation and the Office of Science (BER) of the US Department of Energy. Computing resources were provided by the Climate Simulation Laboratory at NCAR's Computational and Information Systems Laboratory (CISL), sponsored by the National Science Foundation and other agencies. NCAR is sponsored by the National Science Foundation. The authors are also grateful for research support provided by the Integrated Assessment Research Program in the Office of Science of the US Department of Energy (DOE SC-IARP). This research used Evergreen computing resources at the Pacific Northwest National Laboratory's Joint Global Change Research Institute at the University of Maryland in College Park, which is supported by DOE SC-IARP. Pacific Northwest National Laboratory is operated by Battelle for the US Department of Energy under contract DE-AC05-76RL01830. The research was supported in part by support from the Office of Biological and Environmental Research of the US Department of Energy extended to the Oak Ridge National Laboratory. Oak Ridge National Laboratory is managed by UT-BATTELLE for the DOE under Contract DE-AC05-00OR22725. This research also used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the US Department of Energy under contract no. DE-AC05-00OR22725. The views and opinions expressed in this paper are those of the authors alone. Edited by: G. A. Folberth