the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
REMIND2.1: transformation and innovation dynamics of the energy-economic system within climate and sustainability limits
Lavinia Baumstark
Nico Bauer
Falk Benke
Christoph Bertram
Stephen Bi
Chen Chris Gong
Jan Philipp Dietrich
Alois Dirnaichner
Anastasis Giannousakis
Jérôme Hilaire
David Klein
Johannes Koch
Marian Leimbach
Antoine Levesque
Silvia Madeddu
Aman Malik
Anne Merfort
Leon Merfort
Adrian Odenweller
Michaja Pehl
Robert C. Pietzcker
Franziska Piontek
Sebastian Rauner
Renato Rodrigues
Marianna Rottoli
Felix Schreyer
Anselm Schultes
Bjoern Soergel
Dominika Soergel
Jessica Strefler
Falko Ueckerdt
Elmar Kriegler
Gunnar Luderer
Download
- Final revised paper (published on 28 Oct 2021)
- Preprint (discussion started on 06 Apr 2021)
Interactive discussion
Status: closed
-
RC1: 'Comment on gmd-2021-85', Anonymous Referee #1, 05 May 2021
The manuscript presents a detail description of the REMIND model. It is well written the content is mostly very clear. However, there are some comments that must be addressed before the manuscript could be considered for publication. The main concern is that it is not clear what new knowledge about REMIND this manuscript brings to the literature. There are several other research papers that describe several components of REMIND (cited in this manuscript as well) that should be better described here (not just refer the reader to other papers). Additionally, it is not clear what is new in this version of REMIND, compared to previous versions that have been published. The manuscript is not clear about this and does not present a comparison or a section that explicitly presents the improvements. Other minor/detail comments are presented next.
Comments Section 1:
Main comment: It is not clear from this section what is new in this version of REMIND to respect the older versions. This needs to be clearly stated. Other minor comments are described next:
- The use of SSPs helps to cover uncertainties regarding technological development for renewable or fossil fuel availability, but also social and behavioral development like population growth, dietary preferences, environmental awareness or international cooperation.
- It is not clear what is meat by “damages” in the following text: , the REMIND model represents some damages and can thus be used for cost-benefit analyses or least total cost analyses
- In the next sentence, what is meant by self-consistent? Also, it is understood that the energy sector is modeled, while the economic (macroeconomic) is input. Hence, it seems a bit contradictory the next sentence: REMIND enables the exploration of a wide range of plausible developments and of possible futures of the energy-economic system exploring self-consistent transformation pathways.
- It is mentioned several times (up to page 3) that REMIND is an NLP, however, no description or ideas of the non-linear components are described. What make the model nonlinear? Some initial hints would be beneficial .
- “REMIND calculates economic and energy investments”. What is meant by an “economic investment”?
- Introduce the concept of Pareto optimum. Pareto is normally use in the context of multi-objective optimization.
- “The optimization is subject to equilibrium constraints, such as energy balances, economic production functions or the budget constraint of the representative household.” It is not clear the mathematical structure of the REMIND model. Is it an NLP in the form of an optimization or equilibrium (MPEC) model? (MPEC: Mathematical program with equilibrium constraints).
- “REMIND is usually run in a decentralized mode where each model region is optimized separately, and clearing of global trade markets ensured via iterative solutions (see section 2.2).“ -> How the model ensures convergence? This needs to be clarified in the corresponding section.
- “CH4 emissions from fossil fuel extraction and residential energy use” What about CH4 from the agro-industrial (food, beef-lamb) sectors? If not considered, it must be clarified.
- “Historical data for the year 2005 is used to calibrate most of the free variables (e.g. primary 130 energy mixes in 2005, secondary energy mixes in 2005, standing capacities in 2005, trade in all traded goods for 2005).” Why there are not updates to the base calibration year? 2005 seems a bit old to account for new trends.
Comments Section 2:
- “(for more information about the modular structure see Dietrich et al., 2019 - Appendix A.” It would be good to introduce some of this information in this article, since it is such a critical piece of the model structure of REMIND.
- “This paper focuses on realizations which are active in default scenarios. More detail about all modules and their interlinkages can be found in the model documentation”. I still believe that this is information is relevant and should be somehow described and discussed in this manuscript.
- “By default REMIND calculates results for the 12 following world regions:” A table with regions and other important information would be better than just listing countries/regions.
- “By parallelizing the calculation of the individual regions in decentralized optimization mode (see section 2.2) the computation time increases only moderately with increasing spatial detail.” It would be interesting to have a general idea of the computational complexity of the model (minutes, hours, days?) depending on the type of scenarios.
- “Time represents a separate dimension” -> What is meant by this? Not clear at all.
- “In essence, the time dimension only increases the number of markets for which the algorithm has to find an equilibrium” I would be extremely careful about the use of the concept “equilibrium”. To this point, the model has been introduced as a NLP optimization problem, with some equilibrium constraints. But there is not clear mathematical structure to really understand what the model does. If it is a pure optimization model, what talk about equilibrium? Why Pareto optimal is mentioned earlier? Please be clear and consistent with the type of solution that is obtained. Also, it was mentioned that CONOPT is used to solve the NLP problem, hence, it is also questionable when authors refer to the “algorithm” use to solve REMIND, since it is in fact a solver who does this process and authors have not developed an algorithm. In the case that an algorithm is indeed implemented, then this has not been clearly stated and differentiated from the NLP-CONOPT process.
- Based on the previous comment, I found then that there is indeed a NASH mode in REMIND. This helps to understand the concept of equilibrium. However, there is still not clarity in terms of what the base structure of REMIND is, how different structures are solve, what type of solution is obtained, solution algorithm, etc. This needs to be further clarified.
Comments Section 3:
- “It is possible to compute the Pareto-optimal global equilibrium including inter-regional trade as the global social optimum using the Negishi method (Negishi, 1972), or the decentralized market solution among regions using the Nash concept (Leimbach et al., 2017)”. This is interesting but needs further clarification. In practice, a Nash solution is an equilibrium, that can be categorized in some conditions as Pareto optimal. In fact, it has been studied that Pareto optimal strategies are a subset of Nash Equilibrium strategies (see paper DOI: 1109/ICCCNT45670.2019.8944817)
- “REMIND considers the trade of coal, gas, oil, biomass, uranium, the composite good (aggregated output of the macroeconomic system), and emissions permits (in the case of emissions-trading-system (ETS) based climate policy).”. Are ETS global in REMIND? Or can be defined for particular regions? If global, how are allowances distributed?
- “To match 2005 values in the IEA statistics, REMIND adjusts the regional by-production coefficients of combined heat and power (CHP) technologies.” This refers to the calibration of REMIND in the Energy Sector? If so, it is still not clear how the full calibration process works since it will also depend on the macroeconomic results and other sectors that may not be linked to CHP plants (transport for instance?).
- “represent challenges and options related to the temporal and spatial variability of wind and solar power” Please elaborate on this. If the mode runs with 5 years’ time steps, how the temporal variability is considered? It is not inter-annual? Curtailment rates, which are mentioned later, will also depend on the increased levels of demand in future periods as well as the inclusion of other flexibility technologies, such as electrolyzer, which can transform excess electricity into a different energy carrier. Hence, it does not seem correct to consider curtailment rates.
- AC3: 'Reply on RC1', Lavinia Baumstark, 09 Jun 2021
-
RC2: 'Comment on gmd-2021-85', Anonymous Referee #2, 07 May 2021
Overall I found the paper to be well-written and to provide a reasonably comprehensive review of the different model components. I thought there was an appopriate level of technical descriptions, with links to references where individual topics are discussed in greater detail. At a high level, my biggest question pertains to the purpose of this paper in the peer reviewed literature; i.e., can the authors state in the text what the value added is of this paper (as opposed to the model)? As noted, there are already several published model documentation papers, as well as reasonably comprehensive online model documentation. Similarly, the results shown in this manuscript were a cursory review of SSP scenarios that were already published and documented in a number of papers four years ago. It doesn't seem appropriate to be re-publishing this scenario data as if it were new.
One way that the authors could differentiate this from the prior literature would be to run scenarios that illustrate the value of new features that have been added to the model since the last documentation in 2017. For example if there's more sophisticated representations of variable renewable energy, the paper could show energy curtailment by region and scenario, or other variables that are interesting but that weren't reported in the SSP inter-comparison exercise and perhaps weren't available at that time anyway.
In terms of reviewing the model, I was generally impressed by the number of features and key interactions captured, but noted two weaknesses that should be explained in the text. First, why is the model calibration year 2005 when it is currently 2021, and the necessary data to calibrate the model to more recent years has been available for a long time? I'd think that the calibration year should be 2010 at a minimum. And, second, why are there only 12 global regions? For policy modeling purposes it's often advantageous to have single-country regions, and 4 of the 12 are single-country which is good, but that leaves some very heterogeneous regions. Canada-Australia/NZ seems an especially interesting market region given that they're at opposite sides of the world. Is there any sub-regionalization in the renewable energy markets, or any other way to prevent windy regions of Canada from supplying electricity to buildings in New Zealand? Similarly, "Other Asia" presumably includes a very wide range of development levels, as South Korea is mixed in with a large number of low-income countries. Can the authors comment on the level of effort/difficulty with adding regions to the model? Perhaps several components already include enhanced detail?
The final thing I was wondering about the model pertains to the renewable energy supply curves, which appear to include uninhabited lands of Russia, Canada, Australia, etc. Am I correct in understanding that all land area is included in these supply curves, starting in the base year, and that there's no consideration of transmission line distance? While this would be a difficult thing to do well (chicken and egg issue with the transmission lines), that does seem a pretty major omission that would tend to make much more wind energy available for much cheaper that it should, for countries like those named above that have large tracts of uninhabited land, thousands of kilometers from any population centers.
What follows are some minor questions and requests for clarification:
* How is proprietary data masked or filtered and re-processed for distribution, given that the model is open source but (presumably) not all data used in its calibration is free?
* Line 34 - should be "example", not "exemplary".
* Fig 1 - I believe "labor efficiency" should be re-named "labor productivity" for consistency with the literature.
* Line 166 (and others): The China region is called "CHA" on line 166 and "CHN" on line 170. My preference would be to always use CHN, the official 3-digit ISO code, similar to the handling of the other single-country model regions (USA, IND, and JPN).
* Can the authors provide a country-to-model-region mapping list in an Appendix? A number of the boundaries are unclear from the descriptions (e.g., Latvia/Estonia/Lithuania, Turkey).
* Line 286 - should be "modes", not "models" (I think; please check)
* Line 615 (about hydropower potential): "The regional disaggregation is based on information from a background paper produced for this report (Horlacher, 2003)" I'm wondering if this is a typo, or perhaps copied from an older document? Otherwise I can't see how a paper published 18 years ago was produced for this report.
* Lines 700-715: for the more detailed version of the buildings module, can the authors comment on how this was calibrated? The disaggregation of energy consumption to the services is not something readily available in external data sources, and the paper cited is under review so a brief description here would help.- AC4: 'Reply on RC2', Lavinia Baumstark, 09 Jun 2021
-
CEC1: 'Comment on gmd-2021-85', Juan Antonio Añel, 17 May 2021
Dear authors,
We have checked your manuscript, and unfortunately, at the moment, it does not comply with our 'Code and Data Policy'. Currently, you archive the code of your model in Github. However, as we state in our policy and Github on its website, it is not a suitable repository for long-term archival.
Therefore, please, move your code to one of the suitable repositories that we list before the end of the Discussions period and make the necessary changes in the manuscript in potential reviewed versions.https://www.geoscientific-model-development.net/policies/code_and_data_policy.html#item3
Best regards,
Juan A. Añel
Geosc. Mod. Dev. Executive Editor
-
AC1: 'Reply on CEC1', Lavinia Baumstark, 17 May 2021
Dear Juan A. Añel,
Thank you for your comment.
I have some need for clarification: The release of the model which was used for this manuscript is archived via Zenodo (Luderer, Gunnar, et al. (2020, October 15). REMIND - REgional Model of INvestments and Development (Version 2.1.3). Zenodo. http://doi.org/10.5281/zenodo.4091409). Unfortunately, we are currently referring to the old release (REMIND2.1.2) in the section "data and code availability". We will correct this in the revised manuscript. The results shown in the manuscript are also archived at Zenodo (https://doi.org/10.5281/zenodo.4313156). Which additional information would you like to have archived in a suitable way? e.g. one file containing all GAMS code? Would this comply with your 'Code and Data Policy'?
Many thanks, Lavinia Baumstark (on behalf of all co-authors)
-
CEC2: 'Reply on AC1', Juan Antonio Añel, 17 May 2021
Dear Dr Baumstark,
Then the problem is that your 'Code and Data Availability' section is confusing at the moment. Please, remove the mention to GitHub in future versions of your manuscript. The code that is necessary to publish is all the code needed to reproduce your work.
Best regards,
Juan A. Añel
Geosc. Mod. Dev. Executive Editor
-
AC2: 'Reply on CEC2', Lavinia Baumstark, 18 May 2021
Dear Juan A. Añel,
Thank you for the quick response. We will clean up our 'Code and Data Availability' section and delete the part mentioning the GitHub-Link. In addition, we will archive GAMS-files for each scenario containing all code needed for this scenario to ZENODO.
Best regards, Lavinia Baumstark (on behalf of all co-authors)
-
AC2: 'Reply on CEC2', Lavinia Baumstark, 18 May 2021
-
CEC2: 'Reply on AC1', Juan Antonio Añel, 17 May 2021
-
AC1: 'Reply on CEC1', Lavinia Baumstark, 17 May 2021
-
EC1: 'Topical Editor comment on gmd-2021-85', Daniel Huppmann, 07 Jun 2021
Dear authors,
In addition to the comments by the reviewers and the executive editor, please also take into account the following considerations when preparing the submission of responses and an updated version of the manuscript:
- There are references to the code in the style "number plus module name" throughout the manuscript. It is not clear what the logic of the numbering is, so I'd suggest to provide an explanation and/or an overview figure or table with all (or at least the most relevant and referenced) modules.
- In the section on steady-stage and equilibrium, you should introduce the general-equilibrium concept early on (not in the last paragraph). Also, this section should cross-reference the "perfect-foresight" assumption of REMIND.
- Table 2 does not have a very complicated structure and could be replaced by a sentence or a list.
- The "additional tax of 50% of the current carbon price" on net-negative CO2 emissions (page 29, line 777f) seems to be a very arbitrary modelling choice. Please provide a rationale for this value.
- The term "internally consistent" may be more intuitive than "self-consistent".
- The phrase "investments turn out regrettable" (p13, line 337) and "capital is enlarged" (page 18, line 465) should be revised.
- The sentence "the marginal of the (variable of) taxed activities is impacted by the tax [..]" (page 19, line 499) is not clear.
- The phrase "these assets are then stranded" (page 23, line 573) should be revised.
- Subsection header 3.4 should be renamed, as this section also includes non-GHG emissions.
- "roughly a doubling" (page 33, line 850) should be revised.
- Section 5 is quite short and the section title "Discussion" is therefore not adequate.
- The programming language(s) should be clearly stated in the section "Code and data availability".
- Please use the year of the latest update of the model description when citing the IAMC wiki (currently, it does not have a year in the reference)
-
AC5: 'Reply on EC1', Lavinia Baumstark, 09 Jun 2021
Please find our response in the PDF supplement.
Best regards, Lavinia Baumstark (on behalf of all co-authors)
-
EC2: 'Reply on AC5', Daniel Huppmann, 13 Jul 2021
Dear authors, thank you for submitting a revised version of your manuscript and a detailed list of responses.
I would like to comment on the changes in response to my earlier notes.
1. Section 3.1.2, steady-state vs. equilbrium:
I was hoping for a more thorough rewrite of this section - I find it very hard to digest and the overlap with section 2.2 is adding to my confusion. In the last paragraph of Section 3.1.2, you introduce (for the first time) the welfare theorems by Arrow and Debreu, which state that "a competitive market equilibrium can be determined as a Pareto optimum". You then proceed to claim that this is exactly what the Negishi approach does. But in my (maybe wrong?) understanding, a Nash solution (under certain assumptions, which seem to be met by your statement on internalizing an externality) is also the result of a competitive market equilibrium - which (per Section 2.2) differs from the Negishi approach.
It could be that trying to explain the solution method before having discussed the underlying economic principles puts the reader onto a very challenging path to follow your exposition. Please reconsider at which point in the manuscript to introduce these concepts, and how to properly introduce the difference beteen equilibrium, general equilibrium and steady state.
2. The arbitrary choice of a 50% tax mark-up on net-negative emissions: I understand the rationale for disincentivizing net-negative emissions, but my question concerned the choice of that value. I do not think that "the 50% assumotion is the middle ground" is a very convincing argument.
-
AC6: 'Reply on EC2', Lavinia Baumstark, 13 Aug 2021
Dear topical editor,
Thank you for your helpful comments, shown in italics below. Find our replies directly below each comment.
1. Section 3.1.2, steady-state vs. equilbrium:
I was hoping for a more thorough rewrite of this section - I find it very hard to digest and the overlap with section 2.2 is adding to my confusion. In the last paragraph of Section 3.1.2, you introduce (for the first time) the welfare theorems by Arrow and Debreu, which state that "a competitive market equilibrium can be determined as a Pareto optimum". You then proceed to claim that this is exactly what the Negishi approach does. But in my (maybe wrong?) understanding, a Nash solution (under certain assumptions, which seem to be met by your statement on internalizing an externality) is also the result of a competitive market equilibrium - which (per Section 2.2) differs from the Negishi approach.
It could be that trying to explain the solution method before having discussed the underlying economic principles puts the reader onto a very challenging path to follow your exposition. Please reconsider at which point in the manuscript to introduce these concepts, and how to properly introduce the difference beteen equilibrium, general equilibrium and steady state.
We moved information from section 3.1.2 to section 2.2 for providing economic background when describing the solution methods. Section 3.1.2 is revised and now includes the paragraph “Arrow and Debreu (1954) introduced two welfare theorems with the general equilibrium theory. The so-called Second Welfare Theorem, in particular, states that the market equilibrium can be determined from a Pareto optimum solution. This finding provides the conceptual basis for the Negishi approach, and the market equilibrium is determined from the social planner’s solution. Manne and Rutherford (1994) first applied the Negishi approach in an intertemporal setting using a joint maximization algorithm (which is similar to the present algorithm).”. Section 3.1.2. is changed to “In economics, the long-term economic growth is called “steady state”, meaning the stability of the evolution problem (note: in contrast to physical sciences, “steady state” in the context of macro-economic growth theory means that key characteristics of the system, such as the savings rate, income share of labor, etc., remain constant, while the overall economy still grows). If an economic system is stable, a deviation from the steady state growth path leads to transition processes that close the gap to the steady state (or balanced growth path) asymptotically. During this process the markets are in equilibrium (i.e. prices equal demand and supply) in each time step. This ensures that basic accounting requests are met (i.e. no loss of commodities at the global level). The REMIND model is supposed to analyse transitions to a balanced growth path in response to policies while market equilibrium is ensured at each time (step). The general equilibrium concept on which REMIND is based is mathematically and numerically tractable and the fundamental theoretical framework of a majority of economic models. It aggregates the independent decisions of various economic agents so that production and consumption are consistent, with a balance between supply and demand, which leads to an efficient allocation of goods and services in the economy. Yet, this concept also has some limitations. On the one hand, there are strong assumptions like the perfect information for all agents. On the other hand, uniqueness and robustness of the equilibrium cannot be demonstrated for a very general set
of assumptions (Balasko, 2009). The ability of REMIND to model long-term growth dynamics and ensuing energy demands is hardly contained by limitations of the equilibrium concept. Application of this concept is contained to international trade interactions, while the dynamics of long-term growth is mainly driven by preferences, productivities, technological change, capital accumulation, population growth and endowments (e.g. fossil resources).”.
2. The arbitrary choice of a 50% tax mark-up on net-negative emissions: I understand the rationale for disincentivizing net-negative emissions, but my question concerned the choice of that value. I do not think that "the 50% assumotion is the middle ground" is a very convincing argument.
The value of 50% is a policy assumption like many other assumptions which are necessary to run scenarios with such a long-horizon model. We changed the sentence to “REMIND assumes the value of 50% to balance the likelihoods that net-negative emissions might be treated equally to emission reductions or not incentivized at all, i.e. a tax of 100% which would preclude any revenues.”.
-
AC6: 'Reply on EC2', Lavinia Baumstark, 13 Aug 2021
-
EC2: 'Reply on AC5', Daniel Huppmann, 13 Jul 2021
-
AC5: 'Reply on EC1', Lavinia Baumstark, 09 Jun 2021
Peer review completion






Can the world keep global warming below 2 °C?and, if so,
Under what socio-economic conditions and applying what technological options?, it is the goal of REMIND to explore consistent transformation pathways.