Model Documentation: REMIND

REMIND (REgional Model of Investment and Development) is a multi-regional Integrated Assessment Model (IAM) that allows investigating cost-efficient transformation pathways with respect to global or regional climate targets under various scenario assumptions. The model broadly consists of three main parts: a linear energy supply system, sector-specific energy demand representations and a macroeconomic growth model (see red, blue and yellow parts in Figure 1). These systems are all linked via a non-linear intertemporal optimization of economic welfare, which provides the model with a high degree of endogeneity across all sectors. 

In this project, the model is run in a setup of 21 world regions, which includes a more detailed regional representation of the European Union and Germany as a separate region. The model optimizes in five-year time steps from 2005 until 2150 where the first time steps up to 2020 are calibrated to historical data. The regional optimization in REMIND is run over several iterations of the model where specific parameters are adjusted across iterations to reach climate targets or other convergence goals.

REMIND covers all relevant greenhouse gas emitting sectors as well as options for carbon dioxide removal (Strefler et al., 2018). The model includes a comprehensive representation of the energy supply system (extraction of primary energy, conversion to secondary energy, provision of final energy) as well as price-elastic and increasingly detailed sector-specific modeling on the energy demand-side (buildings, industry, transport). Land-use, agricultural emissions, bioenergy supply and other land-based mitigation options are represented through reduced-form emulators derived from the detailed land-use and agricultural model MAgPIE (Model of Agricultural Production and its Impact on the Environment). It therefore has a broad scope for investigating climate change mitigation options across sectors and regions. 

The scenario generation with REMIND is a two-stage process. First, REMIND is calibrated to meet trajectories of GDP and population data based on the shared-socioeconomic pathways (SSPs) as well as final energy and energy service trajectories from detailed sector-specific energy demand projection models (EDGE-Buildings, EDGE-Industry, EDGE-Transport) for a reference scenario that describes a certain continuation of current trends. Second, based on this reference scenario a climate policy scenario is run from 2025 onwards which aims to meet specific climate targets at global or regional level (e.g. climate neutrality in Germany by 2045). Target compliance is implemented by adjusting a carbon price trajectory over iterations until the desired emission levels have been reached.

Figure 1: Flow chart of the model REMIND

Energy supply

The energy supply system in REMIND represents the conversion of primary energy carriers into secondary energy carriers and their transport and distribution to end-use sectors with a broad range of more than 50 technologies. In addition to all relevant power and heat generation technologies, the energy production and conversion technologies include, among others, different routes for the conversion of biomass as well as technologies for the production of hydrogen and synthetic fuels (e.g. Power-to-X).

The energy system accounts for system inertias and path dependencies induced by existing capital stocks, e.g. in power plant infrastructure and endogenous learning-by-doing. It furthermore takes challenges related to rapid upscaling of new technologies via cost-markups into account that are assumed to increase with the square of year-to-year capacity additions (adjustment cost). The REMIND model represents the endowments of exhaustible primary energy resources as well as renewable energy potentials based on bottom-up estimates. Several of the key technologies of the energy transition such as solar PV, wind, batteries, electrolysis and direct air capture are subject to cost reductions via endogenous learning-by-doing. Technological progress of other technologies is parameterized via exogenous assumptions.

The macroeconomy is represented in REMIND by a Ramsey-type growth model in which the aggregate macroeconomic good (GDP) can either be consumed to generate welfare or reinvested into the economy in all time steps.The macroeconomic good is the output of a nested constant elasticity of substitution (CES) production function that distinguishes between capital, labor and energy inputs at the highest level and buildings, industry and transport in the energy sector. The model includes endogenous trade of primary energy goods via a formulation of nash-markets (Leimbach et al., 2017). Regions can import or export to a global pool and world market prices are adjusted over the iterations of the model until a trade balance is reached such that the sum of all regional exports and imports to the pool are sufficiently close.

Energy demand

Industry sector

REMIND models the production of the subsectors cement, chemicals, and steel, as well as an aggregated subsector “other industry”, which are linked to the rest of the model via physical (cement and steel) or monetary (chemicals and other industry) production quantities in the CES production function (Pehl et al., in prep.). Steel production distinguishes between primary and secondary steel, which are produced either with fuels (coal, gas, hydrogen) in blast furnaces or with electricity from scrap. The final energy demand of industry in baseline scenarios is driven by exogenous paths for GDP, material efficiency (production volume per unit of GDP), and energy efficiency (energy input per unit of production), which were determined according to external scenario assumptions. In mitigation scenarios, policy instruments such as the CO2 price affect final energy prices and the model reacts by adjusting production volumes, the use of other final energy sources, and adjusted investments in the capital stock of an abstract technology to increase energy efficiency. The subsectors cement, chemical, and steel have marginal abatement cost curves for emission abatement through CCS.

Buildings sector

The main drivers for energy demand in the buildings sector are modeled in the EDGE-Buildings simulation model and projected into the future (Levesque et al., 2018). EDGE-Buildings distinguishes between energy for heating purposes and appliances. An increase in GDP per capita (income in this case) leads to an increase in living and usable space per capita, depending on behavior. In contrast, an increase in population density has a dampening effect.

The efficiency of building envelopes also improves with rising income and leads to lower specific heating and cooling requirements. Together with projections of heating and cooling degree days, which are also behavior-dependent (room temperature), the future useful energy demand for air conditioning in the buildings sector follows. In addition, the hot water, cooking and household electricity requirements are also projected into the future, which increase with income depending on the behavioral scenario.

We make assumptions about the future distribution of energy sources in the various end uses and the improvements in the technical efficiency of the corresponding technologies. All end uses together describe a baseline scenario for the energy demand of the building sector.

In REMIND, the building sector is represented by a branch of the CES production function whose efficiencies are determined from EDGE-Buildings during calibration to the baseline scenario. Deviations from the baseline scenario, e.g. due to a higher CO2 price, can lead both to a shift to low-emission energy sources and to a general decline in energy demand in the buildings sector. Mark-up costs can also be changed for certain production factors. For example, we lower the markup on heating with heat pumps in electrification scenarios to reflect an optimistic technological development and corresponding incentive and subsidy programs.

Transport sector

REMIND is coupled to the detailed transport model EDGE-T that simulates consumer choices between different transport modes and technologies via a logit function approach (Rottoli et al, 2021a; 2021b). Mobility is divided into passenger and freight demands, each broken down by trip length into long-distance and short-medium-distance components. REMIND provides EDGE-T with aggregate energy service demand (in passenger kilometers and freight ton kilometers) from the macro system and with energy prices from the energy supply system. EDGE-T then simulates the choice between transport modes (e.g. LDV versus bus transport), vehicle types (e.g. large versus compact vehicles) and powertrains (e.g. battery-electric, fuel-cell or internal combustion engines). EDGE-T provides REMIND with energy demand per carrier and capital cost of the transport fleet. REMIND and EDGE-T are iteratively coupled to converge such that the feedback of the transport system on the full-system optimization is integrated. The decision between transport modes and technologies involves monetary (fuel cost, vehicle ownership cost, value of time) as well as non-monetary components, or inconvenience costs (e.g. range anxiety, risk aversion, model availability). The latter are implicitly represented for all transport modes except for LDVs. For non-LDVs, their historical value is derived from past trends and their future development is driven by a set of technology scenario assumptions. In the case of LDVs, inconvenience cost components are explicitly modeled for each powertrain: the powertrain adoption results from an endogenous market where inconvenience cost components vary as a function of the powertrain market share.

Flexibility (storage, DSM, grids)

The REMIND model captures the challenges and options related to the temporal and spatial variability of wind and solar power (Ueckerdt et al., 2017; Pietzcker et al., 2014). In addition to flexible demand response, also inter-regional pooling as well as short-term storage (diurnal time-scales, mostly via batteries) and long-term storage (up to seasonal time-scales) play a key role for facilitating variable renewable energy (VRE) integration.

REMIND parameterizes corresponding technology and region-specific VRE storage and grid expansion requirements as well as curtailment rates (i.e., unused surplus share of VRE electricity generation). The parameterization is derived based on scenarios from two detailed hourly power system models. Moreover, the model features a parameterization of the flexible operation of electrolysis that leads this technology to run at lower full load hours and lower-than-average electricity prices calibrated to data from detailed hourly power system scenarios.

Policy instruments and measures

REMIND resolves a number of policy instruments and measures. Primarily, it adjusts carbon prices to reach the desired emission reduction targets (see above). However, there are also other policy levers that can induce specific changes. First, energy taxes are included in REMIND on final energy level across different energy carriers and end-use sectors, which can be modified to explore effects of different tax incentives across energy carriers in future scenarios. Similarly, mark-up costs on energy demand-side can be changed to simulate technology push scenarios, for instance, with respect to an accelerated uptake of heat pumps.
Different scenarios with respect to technological and behavioral changes in the transport sector can be chosen that are regarded as consequences of policy interventions (e.g. subsidies for electric vehicles, redesign of urban infrastructure to push public transport).

Finally, REMIND can be calibrated to different energy and service demand trajectories in the reference scenario, which assume changes to the current trends that also need to be justified by policy interventions (e.g. higher material efficiency via a circular economy).

Methods and model framework

REMIND solves for an inter-temporal Pareto optimum in economic and energy investments in each model region accounting for inter-regional trade in goods and the compliance with climate targets. The model is written in GAMS (General Algebraic Modeling Language) and formulates a non-linear optimization problem for every region in every iteration of the model, which is handed over to the solver CONOPT. The solver process happens in parallel for all regions during a model iteration to save runtime, while adjustments between model iterations (e.g. carbon prices) ensure that overall convergence targets (e.g. climate targets, trade balances) are met. The model concuts an intertemporal optimization of economic welfare across 21 world regions.

While the core of the model is written in GAMS, there are several R libraries that support REMIND, for instance, by the generation of the input and output data of REMIND or establishing (optional) couplings to other models (EDGE-T, Magpie). In particular, the mrremind library is used to create the input data for REMIND, while the remind2 library is used to create the reporting file with variables based on the IAMC standard from the GDX output file provided by GAMS.

Overview of exogenous assumptions and endogenous outputs

Exogenous assumptions
VariableSourceComment
PopulationSSP2 Scenario 
Macroeconomics / GDP (in reference run)SSP2 Ariadne ScenarioExogenous in the reference run with weak climate policy. Endogenous in the run with climate targets.
Final Energy / Energy Service Demand (in reference run)Sector models: EDGE-Transport, EDGE-Buildings, EDGE-Industry, as well as Ariadne-specific assumptions (FORECAST) for steel and cement production.Exogenous in the reference run with weak climate policy. Endogenous in the run with climate targets. Steel and cement production are currently always exogenous according to common assumptions in Ariadne. 
Secondary Energy Trade (e.g. Hydrogen, E-Fuels)Ariadne-specific assumptions on imports of hydrogen and e-fuels. 
Technology costs (partly)multiple, or partially dependent on the scenario narrativeAll technologies for which no endogenous learning is built in (e.g. thermal power plants, Fischer-Tropsch synthesis, pipelines) have exogenous technology costs.
Certain GHG emissions sources outside of the energy sector: LULUCF, Agriculture, Waste, F-GasesEDGAR, Scenarios from PBL and MAgPIE 
Energy taxes (other than the CO2 tax)Global tax data set from IIASAWe plan to update German energy taxes here soon. Future development constant in the standard case; otherwise depending on the scenario narrative.
Biomass potentialAriadne-specific assumptions for Germany (DBFZ study)Globally endogenous or parameterization on scenarios of the MAgPIE model.
   
Endogenous outputs:
VariableModel representationComment
Macroeconomics / GDP (in the climate policy run)Ramsey- growth model with CES production function 
Final Energy / Energy Service Demand (in the climate policy run)  Buildings / Industry: CES production function   Transport: Coupling with EDGE-transport 
Energy generation and distribution (PE -> SE -> FE transformations)Linear model of the energy system 
Primary energy trade (Biomass and fossils)Global pool-trading via nash-marketsAriadne-specific assumptions on low biomass imports
Primary energy production (fossil)Region-specific parameterized supply curves 
Emissions (Energy and industrial processes)Specific emission factors for energy flows / material flows 
Technology costs (partly)Endogenous non-linear learning curves that depend on the global cumulative installed capacityFor specific technologies that are not yet fully mature: e.g. solar PV, wind, electrolysis, DAC.
Investments into the energy systemResults from specific technology costs and capacity expansion 
Carbon Management / CDRCDR and CCU technologiesCDR: BECCS, DACCS, enhanced weathering, biochar   CCU: production of synthetic carbon-based liquid and gaseous energy sources (e-fuels)
CO2 priceIterative adjustment of the CO2 price until the respective emission targets are reachedAssumption: linear CO2 price between the target years
Energy pricesShadow prices of the balance equations of the energy system model 

The source code of the model is available at https://github.com/remindmodel/remind and further documented at https://rse.pik-potsdam.de/doc/remind/3.2.0/