Lecture 05 - Hands on with Climate IAMs

NOTE THE REQUIRED BEFORE-CLASS SOFTWARE INSTALL

Install Julia and the Mimi modeling framework on your laptop. To do so, first install Julia by following the installation steps documented in this powerpoint: Julia Installation Instructions. All of these steps will be done using the Terminal (Mac/Linux) or Command Prompt (Windows). This might feel awkward at first, but trust me, it’s a critical skill to get good at. Also note that the install might take a while, so please do this well before the lecture and be patient with it. Please do all these steps and confirm they are working before lecture on 2025-09-16.

Reading: Nordhaus 2017, Moore et al. 2022

Optional readings: Barrage and Nordhaus (2024), Waldhoff et al. (2014)

Slides as Powerpoint: Download here

Video link: On Youtube

Content

Introduction and Course Transition

This lecture marks a significant transition point in the Applied Earth Economy Modeling course. The first third of the course has focused on establishing foundational concepts and understanding the trajectory of where the field is headed. Now, the course shifts toward hands-on work with specific tools and applications. The distinction between understanding what earth economy modeling is, which has been covered in previous lectures, and actually applying specific tools in practice represents the core focus of this transition. A key goal of the course is to equip students with practical tools that enable independent research capabilities. This objective drives both the inclusion of a research project in the curriculum and the structured approach to practicing with these tools together before students embark on independent work.

The lecture begins with an essential discussion of where climate integrated assessment models (IAMs) fit within the broader ecosystem of modeling approaches. Following this contextual framework, the session examines two critical readings that represent both the foundation and evolution of climate IAMs. The first is a seminal article by Nobel Prize-winning economist William Nordhaus, who created the Dynamic Integrated Climate-Economy (DICE) model. The second reading, by Francis Moore, presents Green DICE, a modern extension of Nordhaus’s framework that incorporates ecosystem services and natural capital. While these theoretical discussions provide important context, the primary focus of the session is getting hands-on experience with Julia programming language and running the DICE 2016 model directly on students’ computers.

Technical Setup and Installation Verification

Before diving into the theoretical content, it’s important to verify that all students have successfully completed the technical setup required for the hands-on portion of the lecture. The installation process involves downloading Julia, a powerful scientific computing language, along with approximately 3 gigabytes of packages representing about half a million lines of code. This collection represents the collective knowledge and contributions of many developers in the scientific computing community. While not all of these lines of code will execute for the specific applications in this course, having this powerful computational framework installed provides students with a sophisticated calculator capable of complex mathematical operations and model simulations.

The ease of modern installation processes can make it easy to forget how impressive this capability is. In approximately fifteen minutes, students have access to tools that would have required extensive manual configuration and compilation just a few years ago. This accessibility democratizes advanced computational modeling, allowing students to engage with Nobel Prize-winning economic models on their personal computers.

Positioning Climate IAMs in the Modeling Landscape

Understanding where climate IAMs fit within the broader landscape of modeling approaches is crucial for appreciating their strengths and limitations. The conceptual framework used to organize different types of models considers two primary dimensions: economic detail and biophysical detail. Within this framework, integrated assessment models occupy a specific niche, though the category itself contains significant diversity.

The DICE model, which serves as the primary focus of this lecture, is positioned in the lower left of this conceptual space when considering both economic and biophysical detail. It represents a single-sector, single-region model, making it relatively undetailed in both spatial and sectoral dimensions. However, this simplicity is balanced by its inclusion of optimizing agents over time, which provides meaningful economic and temporal dynamics. This positioning reflects a deliberate design choice that prioritizes tractability and transparency over comprehensive detail.

Climate IAMs, as a category, typically emphasize the economic component more than physical detail. They generally include sophisticated climate modeling capabilities but maintain limited representation of landscape dynamics and spatial heterogeneity. This contrasts with other types of IAMs, such as those used in developing the Shared Socioeconomic Pathways (SSPs) and Land Use Harmonization (LUH2) scenarios, which often provide more detailed spatial and sectoral resolution but may sacrifice some economic sophistication.

William Nordhaus and the Origins of DICE

William Nordhaus represents a pivotal figure in the development of climate economics. His work on the DICE model fundamentally shaped how economists approach the question of optimal climate policy. Understanding Nordhaus’s contribution requires appreciating both the elegance of his approach and the controversies it has generated.

The context for understanding Nordhaus’s influence can be illuminated through the work of Bjorn Lomborg, a well-known environmental economist who frequently supports non-consensus opinions on environmental issues. Lomborg has earned a reputation as an intelligent climate skeptic who systematically examines environmental topics through economic analysis, often making counter-arguments against environmental protection based on cost-benefit calculations. His famous article “Recycling is Trash” exemplifies this approach, arguing that recycling can sometimes be worse for the planet when subjected to rigorous cost-benefit analysis.

Lomborg’s recent book attempts a comprehensive climate skepticism, delving into scenarios, scientific measurement accuracy, and economic analysis. However, his entire argument ultimately hinges on one crucial number: the cost of solving climate change. The source for this number, which underpins much of modern climate skepticism’s economic arguments, is Nordhaus’s 2017 paper and the DICE model it describes.

The Mathematical Framework of DICE

The DICE model’s mathematical structure embodies a utilitarian approach to climate policy optimization. Nordhaus aims to maximize social welfare, denoted as W, over time. This optimization is not myopic but extends from the present (T=1) to a terminal time period (T_max), capturing the long-term nature of climate policy decisions. The framework’s sophistication lies in its integration of economic production, climate dynamics, and policy choices within a single optimization problem.

The utility function at the heart of the model depends on consumption over time, indexed by T. Population dynamics matter significantly in this framework, so utility is multiplied by the number of people in each time period. The discount factor, R, emerges as perhaps the most crucial and controversial parameter in the entire model. This discount rate fundamentally determines how much climate mitigation the model considers optimal, as it governs the relative weight given to future versus present welfare.

The specific functional form chosen for utility is a constant relative risk aversion (CRRA) function: C^(1-alpha). This function is elegantly simple, monotonic in consumption, and homogeneous of degree one, making it mathematically tractable while capturing essential features of how utility changes with consumption levels. Abatement efforts are indexed over time, reflecting the temporal flexibility in climate policy where society can choose to abate emissions now or defer action to later periods.

Climate damages represent another critical component of the model, formulated as a quadratic function of average temperature at time t. The specific coefficients of this damage function, along with the discount rate, constitute the most heavily debated aspects of the DICE framework. Despite the complexity of the climate-economy system being modeled, the mathematical structure remains remarkably simple and grounded in standard economic theory. The model maximizes utilitarian value, produces consumption through a production function, and allows society to choose how much to invest in abatement. Abatement increases utility by reducing future climate damages, but this benefit must be weighed against the opportunity cost of reduced current consumption.

The production function in DICE allocates economic efforts to produce output (Y_T), which equals consumption minus the costs of abatement and mitigation. This structure creates a clear trade-off between current consumption and future climate protection, with the optimal balance determined by the model’s parameters, particularly the discount rate.

The Discount Rate Debate

The central question addressed by DICE is determining the optimal level of climate mitigation, and this fundamentally depends on how much society cares about the future relative to the present. If society places high value on future welfare, early mitigation becomes highly effective because benefits accrue over extended time periods and reduce cumulative damage. Conversely, if future welfare is heavily discounted, minimal mitigation appears optimal because the benefits seem too distant to justify current sacrifices.

Nordhaus advocates for what he terms a descriptive approach to setting the discount rate, arguing it should reflect actual economic outcomes observed in real markets. This contrasts with the prescriptive approach, used in other prominent analyses like the Stern Review, which bases the discount rate on normative goals such as achieving specific climate targets. Nordhaus contends that the descriptive approach is superior because it grounds the analysis in observable behavior rather than subjective judgments about what ought to be.

However, this argument faces significant philosophical challenges. Observing only current behavior introduces an inherent bias toward those who exist now, systematically excluding the preferences of future generations who cannot participate in current markets. If one believes future people matter morally, the descriptive approach fails to solve the underlying ethical problem of intergenerational equity. The debate between Nordhaus and Nicholas Stern, who used a much lower discount rate and consequently found much more mitigation to be optimal, characterizes much of modern climate economics.

This philosophical divide has practical implications. The Nordhaus model’s relatively high discount rate leads to conclusions favoring modest near-term climate action and relatively low estimates of the social cost of carbon. These results explain why the DICE model is frequently cited by climate skeptics to support arguments against aggressive climate policy.

Key Results from DICE 2016

The DICE 2016 model generates several important empirical results that illuminate the relationship between economic growth, emissions, and optimal climate policy. Industrial CO2 emissions in the model are directly tied to aggregate GDP, creating a fundamental link between economic activity and climate forcing. Nordhaus compares different model versions, showing that the original DICE 2013 projected higher emissions while the 2016 update shows somewhat lower trajectories.

The model also enables comparison with other modeling efforts through model intercomparison exercises. The RCP8.5 scenario represents the most pessimistic emissions pathway in the climate modeling community, and both DICE 2013 and 2016 generate baseline runs that approach this high-emissions scenario. Nordhaus analyzes how base parameters affect optimal mitigation levels and the social cost of carbon, finding values significantly lower than those advocated in the Stern Review.

These results have profound policy implications. The relatively low social cost of carbon generated by DICE supports arguments for modest climate action, while the model’s structure suggests that some degree of warming may be economically optimal when balancing mitigation costs against climate damages. This conclusion has made DICE both influential and controversial in climate policy debates.

Technical Implementation: From Theory to Practice

Moving from theoretical understanding to practical application requires engaging with the model’s computational implementation. The canonical DICE 2016 model was originally implemented in GAMS (General Algebraic Modeling System), a specialized software package for optimizing large systems of equations used extensively in integrated assessment modeling. However, GAMS presents significant barriers to accessibility, with licensing costs around $3,000 per year and additional expenses for advanced solvers.

The movement toward open-source alternatives represents a democratization of climate-economy modeling. Julia has emerged as a powerful challenger to GAMS’s traditional dominance in this space. Julia offers exceptional speed for numerical computation, particularly when combined with the JuMP library for mathematical programming. JuMP provides connections to various solvers, including some originally developed for GAMS, while also enabling the use of free, open-source alternatives.

This shift toward open-source tools has pedagogical importance beyond mere cost savings. The tools students learn in an academic setting will remain accessible after they leave the university environment and lose access to expensive institutional licenses. This continuity enables ongoing engagement with climate-economy modeling throughout their careers.

Green DICE: Extending the Framework

While the standard DICE model provides important insights, its treatment of natural capital and ecosystem services remains limited. Francis Moore and colleagues developed Green DICE to address this limitation, extending the DICE framework to explicitly incorporate environmental values beyond climate damages. This work, co-authored with Bastian Olvera, represents a significant conceptual and technical advance in integrated assessment modeling.

The standard DICE production function includes manufactured capital (K) and human capital represented as effective labor (A × L), where the A factor captures technological progress and human capital accumulation. This formulation follows standard economic theory where output Y generates social welfare through consumption. The optimization problem remains maximizing utility from consumption subject to production constraints.

A fundamental criticism of the standard model centers on its absence of environmental inputs to production. While manufactured capital and labor are explicitly represented, natural capital remains invisible in the production function. Green DICE addresses this gap by adding natural capital as an explicit factor of production. This natural capital can enter the production function directly alongside K and L, or it can be disaggregated into marketed values such as timber production and ecosystem services like sediment retention and water purification.

The distinction between marketed and non-marketed ecosystem services proves crucial. While marketed natural capital values enter the production function directly, ecosystem services often provide value that markets fail to capture. Green DICE focuses primarily on marketed values but emphasizes the importance of non-marketed ecosystem services for complete welfare accounting. The model also incorporates direct valuation of ecosystem services through existence value and bequest value, representing ways society values the environment beyond its contribution to economic production.

Quantitative Results from Green DICE

The Green DICE framework is not merely conceptual but fully quantitative, with every component corresponding to specific equations and variables that can be solved computationally. Running Green DICE generates results that can be directly compared with standard DICE outputs while highlighting the impact of incorporating ecosystem services.

The standard DICE model, represented by dotted lines in comparative analyses, shows optimal emissions increasing until approximately 2060 before beginning to decline. Adding ecosystem services fundamentally alters this trajectory. When only market values of ecosystem services are included, emissions begin falling earlier. Including use values strengthens this effect further, and the complete specification incorporating all ecosystem service values results in immediate emission reductions.

Perhaps most significantly, the complete Green DICE specification barely crosses the 1.5°C warming threshold established in the Paris Agreement, contrasting sharply with the 3.0°C warming by 2100 projected by DICE 2016 under optimal policy. This dramatic difference demonstrates how incorporating environmental values changes optimal climate policy prescriptions while maintaining the same fundamental economic framework and optimization approach.

Hands-On Implementation with Julia

The practical portion of the lecture focuses on implementing and running climate IAMs using Julia. The installation process begins with downloading the long-term support release of Julia for Windows, Mac, or Linux operating systems. Setting the path variable ensures the command prompt can locate Julia, enabling command-line access to the interpreter. This setup may seem archaic to those accustomed to graphical interfaces, but command-line proficiency remains central to scientific computing and becomes increasingly valuable with advanced usage.

Opening Julia from the command prompt displays ASCII art and loads the compiler. Once ready, the Julia prompt appears, providing access to a powerful calculator environment. While interactive computation is useful, copying and pasting commands becomes inefficient for complex analyses. Script files with the .jl extension provide a better workflow, allowing users to save and run sequences of commands. Julia scripts can be executed by calling Julia with the script filename from the command line.

Julia integrates seamlessly with Jupyter Notebooks, which combine formatted text, code, and outputs in a single document. This integration facilitates reproducible research and educational materials. The most advanced usage involves integrated development environments like VS Code, which provides syntax highlighting, debugging capabilities, and project management tools.

Package Management and Mimi Framework

Julia’s built-in package management system simplifies installing and maintaining scientific software. Entering package mode by typing the right bracket character at the Julia prompt provides access to package management commands. The Mimi modeling framework, developed by David Anthoff, provides specialized tools for running integrated assessment models in Julia. Mimi is well-documented and available on GitHub, with comprehensive tutorials covering installation and usage.

The installation process includes core packages like GDAL for geospatial data, CSV handling utilities, and statistical functions. Updating all packages ensures compatibility and access to the latest features. The course also requires installing MimiDICE2016.jl directly from GitHub, demonstrating how Julia can incorporate packages from various sources beyond the official registry.

After installation, loading packages makes their functionality available in the current session. The command “using Mimi” imports the Mimi framework, while importing MimiDICE2016 loads hundreds of equations defining the DICE model’s structure. These equations encode the optimization problem of maximizing social welfare subject to production constraints and climate damages.

Running and Exploring the DICE Model

Creating a model object with the command “M = MimiDICE2016.get_model()” instantiates a complete representation of the DICE model including data, equations, and solution methods. This object-oriented approach encapsulates all model functionality within a single entity that knows how to configure and solve itself. The analogy to an airplane object with a takeoff procedure illustrates how the model object contains both data and methods for its own execution.

Running the model with “run(m)” triggers the solution process using Julia and JuMP to maximize social welfare and compute all endogenous variables like production, emissions, and temperature over time. The solution process involves sophisticated numerical optimization techniques but remains accessible through simple function calls.

The Mimi framework includes powerful visualization capabilities through the Explore functionality. Executing “explore(m)” launches a user interface, likely built with Electron, providing interactive exploration of model results. Users can examine economic variables like GDP, climate variables like radiative forcing and ocean temperature, and the connections between biophysical systems, emissions, and damages. This exploration tool enables replication of results from Nordhaus’s 2017 paper while also facilitating sensitivity analysis and scenario exploration.

Version Control and Collaborative Development

As earth economy modeling advances toward more complex, interconnected systems, version control becomes essential. Git has emerged as the standard tool for code collaboration, used by major technology companies and the open-source community. Proficiency with Git has become essential for technical positions in both academia and industry.

The Earth Economy DevStack repository, developed at the University of Minnesota, provides a playground for linking multiple models. While DICE represents one self-contained model and Green DICE extends it with additional functionality, researchers often need to combine results from multiple models addressing different aspects of the earth-economy system. The DevStack facilitates these model coupling exercises.

Cloning a repository downloads a complete copy including all files and version history. The hidden .git folder stores this history and enables synchronization with remote repositories. This distributed version control model allows multiple researchers to collaborate on model development while maintaining a complete audit trail of changes.

Looking Forward: Assignment Structure and Research Applications

The upcoming Assignment 3 represents the first substantial technical assignment of the course, building on the setup completed in Assignment 2. Students will modify existing models to answer research questions like exploring the impact of different discount rates on optimal climate policy. Rather than accepting model results based on others’ assumptions, students will gain the capability to rerun models with alternative parameters and examine sensitivity to key assumptions.

This hands-on capability transforms students from passive consumers of model results to active participants in climate-economy research. The ability to access, modify, and run Nobel Prize-winning economic models on personal computers represents a democratization of scientific tools that would have been unimaginable just a generation ago. These skills provide a foundation for independent research that extends beyond the classroom into professional careers in climate policy, economic analysis, and integrated assessment modeling.

Conclusion

This lecture has bridged the gap between theoretical understanding of climate IAMs and practical implementation skills. Starting from the conceptual positioning of these models within the broader modeling landscape, through the mathematical foundations established by Nordhaus, to the extensions developed in Green DICE, and finally to hands-on implementation in Julia, students now possess both the theoretical knowledge and practical tools for engaging with climate-economy modeling. The combination of understanding the assumptions and debates underlying these models with the ability to modify and run them independently provides a powerful foundation for contributing to one of the most important policy challenges of our time.

Transcript

All right, everybody, let’s get started. Welcome to Lecture 5: Hands-On with Climate IAMs.

This marks a transition in the course. The first third focused on foundational concepts and where we’re headed. Now, we’re switching to hands-on work with specific tools. This is Applied Earth Economy Modeling. That means we need to learn what earth economy modeling is, which we’ve discussed, but now we’ll focus on applying specific tools.

One key goal of this course is to equip you with tools for independent research. That’s why we have a research project, but I don’t want to send you off without guidance. We’ll practice with these tools together.

Today, I’ll start with a discussion of where climate IAMs—integrated assessment models—fit within the broader modeling space. Then, we’ll jump into two readings: one is a well-known article by Nobel Prize-winning economist William Nordhaus, who created the DICE model. The second is by Francis Moore, who developed Green DICE, a modern extension of Nordhaus’s framework. We’ll spend some time on that, but the fun part is getting hands-on with Julia and running the DICE 2016 model on your computer.

First, did everyone get the install instructions and Julia running on your computers? Any challenges? Did all the packages install correctly? How long did it take? Fifteen minutes? That’s fast. It’s easy to forget how impressive it is—downloading about 3 gigabytes, probably half a million lines of code, representing the collective knowledge of many contributors. Most of those lines won’t run for this, but you’ve set up a powerful calculator.

We’ll get hands-on with that soon.

Let’s start with some context. We’ll keep returning to this as a roadmap for where our models fit. The other day, I circled integrated assessment models. I wish we’d split these into two boxes before publishing, but here’s my updated version. Even within each box, we organized them by economic detail, biophysical detail, or less detail. DICE is placed in the lower left—it’s not numeric, but it’s one sector and one region, so it’s undetailed in both respects. However, it includes optimizing agents over time, so it has economic and sectoral detail.

Climate IAMs focus more on the economic component than physical detail. They typically have detailed climate modeling but limited landscape representation. That’s what we’re plotting on this axis. In contrast, other IAMs, like those from the SSPs and LUH2, are more detailed in other areas.

Now, let’s talk about a key figure: Bill Nordhaus. I like reading books I both agree and disagree with. One of the hardest books I’ve read recently was by Bjorn Lomborg, a well-known environmental economist who often supports non-consensus opinions. He’s a smart climate skeptic. For example, he wrote a famous article, “Recycling is Trash,” arguing that recycling can be worse for the planet based on cost-benefit analysis. He systematically examines environmental topics, often making counter-arguments against protection using economics.

His recent book attempts climate skepticism, going into detail about scenarios, science, and measurement accuracy. But his argument hinges on one number: the cost of solving climate change. How do you get that number? The answer is Nordhaus 2017.

Nordhaus created the DICE model to answer what is the optimal amount of climate mitigation, recognizing both the benefits (less climate damage) and costs (spending on mitigation rather than productive sectors). We chose the 2017 paper because it preceded his Nobel Prize and is a seminal model. There’s also Barrage and Nordhaus 2023, the latest update, but the fundamentals remain similar.

Let’s discuss the math. Nordhaus aims to maximize social welfare—a utilitarian approach. We’ll call that W. He’s not myopic; he maximizes social welfare from now (T=1) to T_max. The utility function is based on consumption over time, indexed by T. Population also matters, so utility is multiplied by the number of people. The discount factor, R, is crucial and controversial. The discount rate determines how much climate mitigation is optimal.

The utility function is a constant relative risk aversion (CRRA) function: C^(1-alpha). It’s simple, monotonic, and homogeneous of degree one. Abatement is indexed over time, as we can choose to abate now or later. Damages are another critical component, modeled as a quadratic function of average temperature at time t, with coefficients. This is the damage function, which, along with the discount rate, is heavily debated.

This is remarkably simple—standard economics: maximize utilitarian value, produce consumption through a production function, and choose how much to invest in abatement. Abatement increases utility by reducing climate damages, offsetting the loss from reduced current consumption.

I won’t test you on this, but it’s elegant. The production function allocates efforts to produce output (Y_T), which equals consumption minus abatement and mitigation costs.

The big question is how much mitigation is optimal, which depends on how much we care about the future. If we care a lot, early mitigation is effective, as benefits accrue over time and reduce cumulative damage. If we don’t care much about the future, little mitigation makes sense, as benefits are too far off.

I won’t dive into the debate here; 8601 covers the mathematics of the discount rate in depth. What I want to highlight is the applied nature of the debate. Nordhaus claims the discount rate should reflect actual economic outcomes—the descriptive approach. The alternative, used in Reference 10 and elsewhere, is the prescriptive discount rate, based on normative goals like reducing climate change by a certain amount.

Nordhaus argues the descriptive approach is better because it’s based on observation, while the prescriptive rate is “making stuff up.” I’d argue against that. If you only observe current behavior, you’re biased toward those who exist now, leaving out future generations. Observing current agents ignores preferences of those not yet born. If you care about future people, the descriptive approach doesn’t solve the ethical problem. If you don’t care about future generations, the two approaches are consistent. But if you think they matter, you must consider their preferences.

Could we create enough capital and consumption goods for future generations to be happy, even without solving climate change? Maybe. The Nordhaus model, and its use in climate skepticism, hinges on the discount rate. The opposite perspective is from Nicholas Stern (Reference 10). The debate between Nordhaus and Stern characterizes much of modern economics. Stern used a lower discount rate and found much more mitigation is optimal.

Key results in Nordhaus: industrial CO2 emissions are tied to aggregate GDP. He compares different models—original DICE 2013 showed more emissions, the update shows less. He also compares model intercomparison groups. RCP8.5 is the most pessimistic scenario, with DICE 2013 and 2016 getting close for baseline runs. He analyzes how base parameters affect optimal mitigation and social cost of carbon. The Stern review argued for a much higher social cost of carbon than DICE. This is why DICE is favored by climate skeptics—it supports low mitigation and low social cost of carbon.

As applied economists, we won’t solve the DICE model by hand—it’s challenging and requires detailed knowledge of the production function and constraints. Instead, we’ll dive into model details. I went to the canonical DICE2016 page to download the model, but it wasn’t on GitHub and wasn’t well documented. The key question is: what happens if we rerun the model with a different discount rate?

GAMS is a software package for optimizing large systems of equations, used extensively in integrated assessment modeling. It’s expensive—about $3,000 per year, more for advanced solvers. I prefer open-source tools. Julia is now challenging GAMS’s monopoly in this space. Julia is fast, especially with the JuMP library for mathematical programming. JuMP links to solvers, including some from GAMS, and allows substitution of free solvers.

The tools you learn now will remain relevant when you lose access to expensive licenses. We’ll use Julia and JuMP when we dive into Nordhaus’s model. But I don’t want to leave you with just the DICE model’s low discount rate and social cost of carbon. Instead, let’s look at Green DICE. Francis Moore et al. 2022 extends DICE. Moore and her collaborator Bastian Olvera (a skilled programmer) co-authored the first Green DICE paper.

Let’s draw this out conceptually. Green DICE takes the basic DICE approach. In the standard model, the production function includes manufactured capital (K) and human capital (A × L). The A factor increases human capital, reflecting effectiveness, not just population. This is standard economics—output Y leads to social welfare. The optimization is to maximize utility from consumption, subject to the production function.

A criticism of the standard model is the absence of the environment—manufactured capital and labor, but no environmental input. Green DICE adds natural capital. Natural capital can be included directly in the production function, alongside K and L. It can also be separated into marketed value (like timber) and ecosystem services (like sediment retention). Ecosystem services may not be valued in the production function, but they’re important. Green DICE focuses on marketed values but emphasizes the importance of ecosystem services.

We also care about the direct value of ecosystem services—existence value, request value. These are direct ways we value the environment. Green DICE is not just a conceptual diagram; every line and box corresponds to equations and variables, making it quantitative and solvable.

When you run Green DICE, you get results similar to DICE but with added ecosystem service value. The standard DICE model (dotted line) shows optimal emissions increasing until 2060. Adding ecosystem services (market-only, use values, complete specification) results in more stringent mitigation, with emissions falling immediately. The complete specification barely crosses the 1.5°C threshold from the Paris Accord, much less than the 3.0°C by 2100 in DICE 2016.

This is persuasive—it uses the same language as DICE but adds details about ecosystem services. The article we read adds more feedback loops, including policy adoption affected by the socioeconomic system. Implementation of climate policy depends on public support, creating feedback loops that influence emissions.

The key output is that coupling social-economic feedback with Green DICE results in lower emissions than RCP7. Most runs are about half a degree less, but even aggressive scenarios barely miss the Paris target of 1.5°C. This reflects that many damages are already baked in, increasing the importance of adaptation as well as mitigation.

There’s more research linking general equilibrium models to explicit land use change components. If you’re interested, explore the referenced papers—great for research topics.

Any questions on the literature before we switch to code?

Let’s review what you’ve done so far. We installed Julia (long-term support release) on Windows or Mac. Any Linux users? I’m switching over—I love Linux. We set the path variable so the command prompt finds Julia. We validated it was running. Who hadn’t used a command prompt before? It looks archaic, but it remains central to computing. The better you get, the more you appreciate it.

On Windows, I use the Start menu to open Command Prompt. Typing “julia” brings up ASCII art and loads the compiler. Once ready, you get the Julia prompt. Julia is a powerful calculator, but typing everything out isn’t efficient. People copy and paste into documents, but script files (.jl) are better. You can run scripts by calling “julia” with the script file.

Julia works natively in Jupyter Notebooks, which combine formatted information and code outputs. R is moving toward Jupyter Notebooks too. The most advanced usage is within an integrated development environment—hence, installing VS Code.

For installing packages, Julia has a built-in package installer. Hit the right bracket to enter package mode. I had you install the Mimi modeling framework, which is a set of Julia scripts for running integrated assessment models, developed by David Anthoff. Mimi is well-documented and available on GitHub. The tutorial covers adding packages. I had you run a line that installs Julia, Mimi, and other packages like GDAL, CSVs, and statistics. We updated everything and imported Mimi.

We also added a package directly from GitHub—Alexandra Pavlov’s MimiDICE2016.jl. Everyone should have reached this point.

Let’s go further. If you’re in the package editor, hit backspace to return to the Julia prompt. Enter “using Mimi.” Does everyone have Julia running? Good.

Now, import MimiDICE2016. It loads hundreds of equations—maximizing social welfare subject to production and climate damages. Now, create the model object: M = MimiDICE2016.get_model(). This returns an object with data, equations, and functions. The get_model function returns the object M, which contains all functionality, including how to run itself.

Think of it like an airplane object with a takeoff procedure. The M object knows how to run. The command “run(m)” tells it to solve the model using Julia and JuMP, maximizing social welfare and assigning variables like production and temperature over time.

The Mimi framework includes an Explore functionality. “explore(m)” launches a user interface (likely Electron). You can explore the model, view the economy, GDP, and parameters over time. This is integrated assessment modeling—radiative forcing, climate dynamics, ocean temperature, and more. The model links biophysical systems to emissions and damages. The Explore tool lets you replicate Nordhaus’s 2017 paper.

For the upcoming assignment, review the Mimi documentation. You’ll want to modify an existing model—once you have DICE or Green DICE on your computer, you can answer questions like “What happens if the discount rate changes?” You won’t rely on someone else’s assumptions; you can rerun the model yourself.

Any questions on what we’ve covered?

One last skill: let’s talk about Git. As we move toward more Earth Economy models, we’ll use a package developed at the University of Minnesota called Earth Economy DevStack—a playground for linking models. DICE is one model; Green DICE links extra functionality. What if you want to pair results with a different model, like an ecosystem service or land use change model? That’s where we’re headed.

Let’s walk through cloning the Earth Economy DevStack repository. If you’re still in Julia, type “exit()” to return to the command prompt. Type “git” to see available commands. The most common is “git clone.” Who has Git experience? When hiring for technical jobs, Git proficiency is essential—it’s the language for code collaboration, used by Microsoft and others.

We’ll clone a repository. Navigate to github.com/jandrewjohnson/eartheconomydevstack. Copy the link. In your command prompt, use “cd” to change directories. “cd ..” goes up a level; “cd files” changes to the files directory. You can use relative or absolute paths. Now, type “git clone [URL].” If you’ve already cloned it, it will say so. Otherwise, it downloads the repository.

When you check your C drive, under Users > [your username] > Files, you’ll see the cloned repository. The hidden .git folder stores the history and enables collaboration. You don’t need to access it directly, but know it’s there.

We’ll use this repository and launch VS Code from it. Did anyone have trouble downloading via git clone? You can do this on MSI (Minnesota Supercomputing Institute), which often requires command line access. The typical workflow is to use Git to get your code, then run Julia or other tools.

Keep an eye out for Assignment 3—the first real assignment. Assignment 2 was mostly about getting tools set up. Any last questions?

Alright, have a good day, everyone!