Lecture 13 - GTAP-InVEST

Reading: Johnson et al. 2023

Slides as Powerpoint: Download here

Video link: On Youtube

Content

Welcome and Course Overview

This lecture marks the final content-bearing session for this half-semester course, focusing on GTAP Invest and the culmination of earth economy modeling linkage that has been the central theme throughout the semester. The session will review all the different components of earth economy modeling, with particular attention to the area that still requires more detailed development: dependencies modeling. A detailed examination of the GTAP Invest model will follow, including specific results that demonstrate key findings and how to express results within this complex modeling space. Time has also been reserved to discuss future directions for this research area and possibilities for students to pursue these topics as research papers, potentially for dissertations or other academic projects.

Administrative Matters and Course Expectations

Regarding the schedule and deadlines, the outlines that have been submitted are looking excellent. The fundamental goal of this course, as articulated in the syllabus, is to provide students with hands-on skills for conducting research in this area. The final project is not designed to produce a polished manuscript complete with an extensive literature review, but rather to show evidence of good progress in using these tools and methods. Students who find their projects interesting and worthwhile can present their work on Thursday, the last day of class. These presentations can feature work that is still in progress, as long as they demonstrate genuine effort and meaningful advancement in understanding and applying the techniques covered in the course.

Extensions beyond the presentation date are available, even extending past the end of the grading period if necessary. Students who are making good progress will receive an A in this methods course. For those who wish to take additional time to develop their papers more fully, intermediate submissions can be graded so that students do not feel pressured to rush their work. Students are encouraged to reach out individually to discuss their specific situations and needs. Given that a substantial amount of content has been compressed into this half-semester format, maintaining flexibility in deadlines and expectations is particularly important. The bilateral email discussions regarding research ideas have been critical to the success of the course, and these ongoing conversations are greatly appreciated.

The Earth Economy Modeling Framework

Returning to the comprehensive picture of earth economy modeling, the course has repeatedly examined the diagram showing the relationship between GTAP and INVEST, which underlies many of the figures that have been presented throughout the semester. Earth economy modeling represents a large conceptual space that is inclusive of many different models, all of which can feature two-way linkages between the earth system and the economic system. GTAP Invest serves as one specific example within this broader framework, representing one particular way of plugging in specific models for each step in the process. The research community is actively expanding beyond this single implementation to experiment with different models at various stages of the linkage.

The focus for this session will be on the model that has been developed in-house, examining how GTAP functions and how land supply functions operate within the model. While extensive discussion has already covered SEALS and INVEST in previous lectures, today’s emphasis will be on the process of upscaling ecosystem service changes and expressing them as economic shocks within the economic model. This upscaling process represents how the dependencies of the economy on nature are modeled within the framework.

The PNAS Publication and Publishing Strategy

The assigned reading for this lecture is the 2023 PNAS article titled “Investing in Nature Can Improve Equity and Economic Returns.” When publishing in top-tier journals such as Nature, Science, and PNAS, the editorial focus is squarely on results rather than on methods or model development. These journals aim to reach a general scientific audience, which means that findings must be centered on compelling results that have broad implications. However, many significant results actually depend on new models or substantial methodological advances, creating a tension in how to present the work. The solution is to present the methodological details in supplemental information while focusing the main manuscript on the key result that will capture reader interest and demonstrate the work’s broader significance.

In the case of the official GTAP Invest publication, the methodological details were relegated to a lengthy appendix, allowing the main text to focus on the keynote result: that investing in nature can simultaneously improve both equity and economic returns. This represents a compelling finding that challenges conventional assumptions about trade-offs between economic development and environmental protection, making it suitable for a high-impact general science journal.

Model Development and Software Implementation

The process of publishing GTAP Invest involved creating linkages between GTAP and INVEST, beginning with what is known as a soft linkage and progressively moving toward a hard linkage. The initial versions of the model employed comparative static formulations, which involve comparing the equilibrium state of the economy before and after a particular shock or intervention. This comparative static approach offers the advantage of allowing results to be exchanged between experts in different models without requiring deep knowledge of each other’s modeling systems. Many collaborative modeling projects end at this stage of soft linkage, as it provides sufficient information exchange for many research questions.

However, for recursive dynamic formulations where the model must be run repeatedly over multiple time steps, the approach of manually emailing results back and forth for each time step quickly becomes impractical. To address this challenge, a comprehensive software stack was developed to automate the entire process of model linkage and iteration. The software development approach was borrowed from the Natural Capital Project (NATCAP), which has pioneered the creation of user-friendly interfaces, robust code implementation, and publicly accessible GitHub repositories that allow users to fork the code and adapt it for their own purposes.

Success in open science is measured by the ability to scale up research tools so that other researchers can run and adapt the code independently, without requiring constant assistance from the original developers. All components of GTAP Invest have been made open source with the exception of two critical elements: the GTAP database itself, which is expensive to license, and the model solution method, which relies on proprietary software called Gempak. The RunGTAP program utilizes a limited version of Gempak, which consists of a set of mathematical equation solvers that have been custom-built specifically for computable general equilibrium models.

While considerable effort has gone into making everything else in the modeling system open source, replacing Gempak has proven to be particularly challenging due to its specialized capabilities and long history of development. This is one of the primary reasons that recent work has begun using Julia programming language, specifically the JuMP library, which shows considerable promise for achieving fully open-source equation solving capabilities. There is a notable trend among economists to increasingly adopt Julia for modeling applications, often skipping Python entirely, because open-source development remains limited in traditional economic software platforms like Stata and MATLAB.

The Gempak System and Mathematical Implementation

Gempak implements what is known as the Tableau language, which is similar to GAMS (General Algebraic Modeling System), providing a framework for defining mathematical equations at scale and coupling them with efficient computational solvers. Beyond its core solving capabilities, Gempak also includes a suite of utilities for tasks such as aggregating the GTAP database to different regional or sectoral resolutions and conducting sensitivity analysis to test the robustness of results. In this sense, Gempak functions both as a programming language and as a comprehensive modeling environment, providing integrated tools for the entire modeling workflow.

Endogenizing Land Supply

One particularly important area for earth economy modeling is the process of endogenizing specific resource pathways, with land use change being a prime example. In existing economic models that treat land as a fixed endowment, there is no supply curve for land that can be shifted in response to changing conditions. If land is simply treated as a fixed quantity that cannot respond to economic signals, then it becomes impossible to meaningfully shock the land supply or to model how land use responds to changes in prices or policies. To address this limitation, modifications had to be made to the models to ensure that land is calculated endogenously, meaning that land supply can respond to economic conditions and can therefore be shocked in meaningful ways that reflect real-world dynamics.

Pollination as a Case Study

To make these abstract concepts more concrete and tangible, pollination serves as an excellent case study for illustrating the complete pathway from biophysical ecosystem service provision to macroeconomic impacts. The goal, as demonstrated in the PNAS article, is to use land use and land cover maps generated by SEALS, which are derived from endogenously determined land use change at the regional scale, as inputs to the INVEST ecosystem service models. Different configurations of land use on the landscape result in different levels of ecosystem service provision, such as varying degrees of pollination sufficiency across agricultural areas.

The pollination model in INVEST establishes a threshold of 0.3 for pollination sufficiency, above which crop yields remain unaffected by pollinator availability, but below which yields begin to decline due to insufficient pollination services. The calculation of pollination sufficiency employs a mathematical technique called convolution, which considers not only the presence of cropland but also the proximity and availability of natural habitat in the surrounding landscape that can support pollinator populations. Large, contiguous expanses of cropland that lack fragments of pollinator habitat interspersed within or adjacent to the agricultural land result in reduced yields for crops that depend on insect pollination.

Converting Biophysical Outputs to Economic Shocks

The biophysical output from INVEST, which provides spatially explicit estimates of pollination sufficiency, is not by itself sufficient for integration into an economic model. The pollination sufficiency values must be converted into specific, quantifiable changes in agricultural yields that can be expressed in economic terms. To accomplish this conversion, the research employed the EarthStat dataset, which provides gridded information on both crop yields and production quantities for 175 different crops at a spatial resolution of approximately 10 kilometers.

To determine the overall change in agricultural production resulting from changes in pollination services, several factors must be considered simultaneously for each location on the landscape. First, the pollination dependency of each crop must be taken into account, as different crops vary substantially in how much their yields depend on insect pollination. Second, the analysis must determine whether each crop is actually being grown in grid cells that are experiencing insufficient pollination sufficiency according to the INVEST model outputs. The availability of high-resolution spatial data on crop locations makes it possible to sum up the production of pollination-dependent crops across all affected grid cells and calculate the total dollar value of production that is at risk due to inadequate pollination services.

This aggregation produces an estimate of the change in dollar-level production, which can then be further aggregated to calculate a percentage change in agricultural productivity for each region in the GTAP model. For example, detailed maps have been created showing the spatial distribution of productivity shocks to the oilseed sector, which is particularly dependent on pollination services. In these maps, white areas indicate locations where there is no crop production occurring, and therefore pollination sufficiency is not estimated for these locations since the service is not relevant where crops are not being grown.

Distinguishing Service Supply from Service Demand

An important conceptual distinction must be made when interpreting these maps: they represent the service value to people, which is essentially the demand for ecosystem services, rather than showing where the ecosystem service itself is being supplied. The supply of pollination services may occur in one location, specifically in areas of natural habitat that support pollinator populations, but the impact of those services on human welfare may occur in a completely different location where the pollinated crops are actually being grown. This distinction between supply and demand locations is critical for understanding the spatial dynamics of ecosystem services and for designing policies that effectively protect the natural areas that generate these services.

Technical Implementation in GTAP

Once the percentage change in productivity has been calculated for each region and sector, the implementation in GTAP follows a relatively straightforward technical process. The calculated percentage changes are first exported as a comma-separated values (CSV) file, which is then converted into a HAR file format that is compatible with the GTAP modeling system. This HAR file containing the shock values is then plugged into the GTAP code, where it can interact with the model’s equations.

The Tableau language, which is used by Gempak, defines a comprehensive system of equations that govern price formation and economic activity levels for each region in the model. Within this system, there is a specific variable called A0 that represents total factor productivity for each sector and region. This A0 variable is adjusted according to the percentage changes defined in the shock file, and the model is then re-solved using standard computational methods to find the new equilibrium with the altered productivity levels. While the conceptual process may seem complex, the actual technical implementation is relatively straightforward once the correct variable has been identified and the appropriate file formats have been prepared.

The Frontier of Dependencies Research

The dependencies linkage represents the component of earth economy modeling that currently needs the most research and development effort. Other areas of the modeling framework, such as the economic model itself or the land use change model, have well-established disciplines with decades of methodological development behind them. However, ecosystem services research has historically focused primarily on valuation approaches that measure the value of ecosystem services based on individual preferences, as revealed through stated or revealed preference methods, rather than focusing on the productivity impacts of ecosystem services and their macroeconomic consequences.

The specific process that has been developed for pollination, while instructive as an example, will necessarily differ for every other ecosystem service that might be incorporated into the modeling framework, because each service has unique impact pathways that connect biophysical changes to economic outcomes. For example, carbon storage and sequestration has a relatively well-understood pathway that connects changes in carbon storage to atmospheric concentrations, global temperature change, economic damages, and ultimately to changes in gross domestic product. This pathway has been studied extensively and is incorporated into integrated assessment models of climate change.

However, other ecosystem services such as sediment retention, erosion control, water quality regulation, and habitat provision for species are considerably harder to model in terms of both their biophysical dynamics and their economic impacts. These services often have complex, indirect pathways of impact that may operate through multiple channels and may involve threshold effects or nonlinear relationships. There is significant potential for research contributions in developing methods to link changes in these various ecosystem services to macroeconomic impacts through well-founded causal pathways that can be defended both biophysically and economically.

Key Questions in Earth Economy Modeling

The discussion now turns to examining what can be learned from this body of work and how the results can inform policy decisions. The results from earth economy modeling research can be organized around three fundamental questions that guide the field. The first question asks how much nature actually benefits the economy in quantitative terms. The second question examines who loses or wins when natural capital is depleted, addressing issues of distributional equity. The third question explores what policies can be implemented to protect both the value that nature provides and the equitable distribution of that value across different populations.

Quantifying Nature’s Economic Value

Addressing the first question about the magnitude of nature’s economic contribution, collaborative work was conducted with the World Wildlife Fund as part of their Global Futures Project. This project involved developing comprehensive estimates using earth economy modeling techniques to evaluate different scenarios for future development pathways. The analysis compared a business-as-usual scenario, which assumed continued land use change and greenhouse gas emissions following historical trends, against a global conservation scenario that featured widespread habitat protection and stabilized land use patterns.

The main finding from this analysis was that the net present value of cumulative economic damages from ecosystem service degradation between 2011 and 2050 amounted to approximately 9.8 trillion dollars under the business-as-usual scenario. While this figure is substantial, it is notably less than the 44 trillion dollars per year that has been estimated by the World Economic Forum in their assessments of nature’s contribution to the economy. The difference can be attributed to the deliberately conservative approach taken in the earth economy modeling work, which considered only those ecosystem service linkages that have strong empirical support and well-established causal pathways.

Despite the conservative approach, a value of approximately 10 trillion dollars represents a significant economic magnitude that demonstrates unequivocally that nature matters for economic prosperity. Perhaps even more importantly than the absolute magnitude of the economic value, when the results were disaggregated to examine who benefits from investing in ecosystem services, a clear pattern emerged. The percentage change in GDP by the year 2050 under the business-as-usual scenario was most negative for low-income countries, meaning these countries would suffer the greatest proportional economic losses from continued environmental degradation. Conversely, under the global conservation scenario, the percentage increase in GDP was greatest for these same low-income countries. This finding shifted the research focus toward the equity question, examining more closely who loses or wins when nature is depleted and who benefits most from conservation efforts.

The Dasgupta Review and Distributional Impacts

The Dasgupta Review, which was commissioned by the UK Treasury to examine the economics of biodiversity, requested a specific assessment of the distributional impacts of ecosystem service collapses across different income groups and regions. The review presents a conceptual framework that defines economic demand as the product of human population and economic activity per capita, divided by production efficiency, while supply is defined as the biosphere’s regeneration rate. The role of earth economy modeling in this context was to provide applied, decision-relevant science by running approved versions of the models and analyzing the distributional consequences of environmental degradation.

The modeling results confirmed that low-income countries are hurt most severely by degraded natural capital, experiencing larger proportional losses in economic welfare than wealthier nations. The analysis also examined extreme scenarios involving ecosystem collapse, finding that such collapses could result in up to a 10 percent reduction in GDP, with annual damages reaching 2.7 trillion dollars. While these estimates remain deliberately conservative due to the focus on well-supported linkages, the key policy-relevant finding is clear: nature matters most for the economic wellbeing of low-income countries, which tend to be more directly dependent on ecosystem services for their livelihoods and have less capacity to substitute away from degraded natural capital.

Policy Analysis for Nature Protection

The third fundamental question addresses what can actually be done to protect nature’s value and ensure equitable distribution of benefits. Collaborative work with the World Bank on their report “The Economic Case for Nature” involved analyzing specific policies that could mitigate the economic losses resulting from ecosystem service degradation. The World Bank specifically requested analysis of actionable policies rather than abstract theoretical constructs or trade-off frontiers that might be intellectually interesting but difficult to implement in practice. Four main policy options were identified and analyzed using the earth economy modeling framework.

The first policy option involves repurposing agricultural subsidies that currently encourage environmentally harmful practices. Removing subsidies that damage the environment serves a dual purpose: it improves market efficiency by reducing distortions in resource allocation, and it reduces negative environmental externalities by removing incentives for practices like excessive fertilizer application or conversion of natural habitat to agriculture. However, removing subsidies typically faces strong political opposition from those who currently benefit from them. To address this opposition and make the policy more politically feasible, the analysis proposed reallocating the savings from removed subsidies as lump-sum payments directly to landowners, with the goal of achieving revenue neutrality while still eliminating the distortionary effects of production-linked subsidies.

The second policy option focuses on investing in agricultural research and development to increase productivity on existing agricultural land. Higher yields on land already in production reduce the economic pressure to expand agriculture into new areas, thereby reducing deforestation and habitat conversion. The modeling examined scenarios with increased government investment in agricultural R&D, with some scenarios using the savings from removed agricultural subsidies to fund this research investment, creating a policy package that is revenue-neutral overall.

The third policy option involves implementing payments for ecosystem services (PES) programs, which represent a market-based approach that directly compensates landowners for providing positive environmental outcomes. The analysis modeled the implementation of global forest payment schemes, which would involve financial transfers from high-income countries to low-income countries to create incentives for protecting forested land. The modeling also examined PES policies that operate without international transfers, reflecting the structure of most real-world PES programs that function at the national or regional level with domestic funding sources.

An important consideration for all payment-based programs is the degree of compliance and enforcement, which determines whether landowners actually deliver the environmental outcomes they are being paid to provide. Compliance has become increasingly measurable through remote sensing technologies that can monitor land cover change and detect violations of program agreements. PES programs have an advantage over some other policy instruments in terms of monitoring because they involve ongoing payments that create continuous incentives to comply, and non-compliance can result in immediate termination of payments.

Combined Policy Impacts and Equity Outcomes

When these various policies are combined into comprehensive policy packages, the modeling results demonstrate that protecting nature provides the greatest benefits to low-income countries. The combined policy scenarios led to the largest increases in regional welfare for low-income country groups, sometimes with these gains coming partially at the expense of high-income countries that must contribute funding for programs like international PES schemes. This finding reinforces the conclusion that policies designed to protect nature are especially beneficial for those populations that are most dependent on ecosystem services for their livelihoods and economic wellbeing.

Analyzing Policy Trade-offs

The research also examined policy trade-offs using the concept of a Pareto efficiency curve, which plots economic outcomes against environmental outcomes to identify the frontier of policy options that cannot be improved upon without making some stakeholder worse off. Specifically, the analysis plotted economic revenue on one axis against environmental quality, measured as mean hectares of natural habitat preserved, on the other axis. The combined policy packages performed well in this analysis, achieving relatively high levels of both economic activity and environmental protection.

However, the analysis revealed that if decision-makers cared only about maximizing economic activity and placed no value on environmental protection, then the preferred policy package would consist of removing environmentally harmful subsidies and investing the savings in agricultural research and development, without implementing any payment for ecosystem services programs. Some policies, such as providing subsidies directly to landowners as lump-sum payments, actually increased GDP but reduced the number of hectares preserved, because landowners used the additional income to invest in agricultural expansion.

In contrast, PES policies had the effect of moving outcomes into what can be characterized as win-win space, where both economic and environmental outcomes improve relative to the baseline. This win-win outcome was especially pronounced when PES programs were combined with other policies like agricultural R&D investment and subsidy reform. The various policy combinations can be visualized on the trade-off curve, with each point representing a different balance between economic and environmental objectives.

Spatial Visualization of Policy Impacts

Each policy scenario corresponds to specific spatial patterns of land use and land cover on the landscape, which can be visualized through maps generated by the SEALS model. For example, policy scenarios that include robust PES programs tend to maintain large, intact blocks of forest, particularly in tropical regions where deforestation pressure is high and biodiversity values are substantial. In contrast, scenarios that lack PES policies or other strong conservation incentives show more extensive agricultural expansion and greater fragmentation of remaining natural habitats.

It is important to emphasize that these maps are not intended as predictions of exactly what future landscapes will look like. Rather, they represent causally linked pathways that connect specific policy incentives to land use decisions and ultimately to landscape outcomes. The models trace through the logical consequences of different policy choices, showing how changes in economic incentives alter land use patterns, which in turn affect ecosystem service provision and economic wellbeing. This causal framework provides a scientific basis for evaluating policy options and understanding their likely consequences.

Future Directions for Earth Economy Modeling Research

Looking toward the future trajectory of this research area, several directions hold particular promise for advancing the field and providing better tools for decision-making. These future directions are especially relevant for students considering potential research topics for papers, dissertations, or other projects. Five main areas represent frontiers where substantial progress could be made with dedicated research effort.

The first area involves increasing the number of land-intensive sectors that are explicitly represented in earth economy modeling frameworks. Currently, the impacts that are modeled flow primarily through changes in cropland, pasture, and managed forestry expansion, which are the traditional land-using sectors in agriculture. However, other sectors are becoming increasingly important drivers of land use change and environmental impact. The energy sector, particularly renewable energy installations like solar and wind farms, requires substantial land areas and is expanding rapidly as countries transition away from fossil fuels. The extraction sector, including mining for materials like lithium for batteries and aggregates for construction, also has significant and growing land footprints.

Recent work has demonstrated that energy and extractive sectors combined have a land footprint that is nearly as large as agriculture, yet conservation efforts and policy attention have focused predominantly on agricultural land use. Including these additional sectors in earth economy models requires detailed spatial data on where these activities are suitable and what their potential for future expansion might be. Developing these datasets and integrating them into the modeling framework represents an important research opportunity.

The second area for future development involves embedding country-specific models within global earth economy modeling frameworks. Global models, while useful for understanding broad patterns and international linkages, are not always suitable for analyzing country-level policies or providing decision-relevant information to national policymakers. By embedding national CGE models within the global framework, it becomes possible to maintain the international trade linkages and spillover effects that are captured in global models while simultaneously allowing for more detailed representation of national land use change processes and domestic policies.

This embedded modeling approach also has practical advantages for research impact, as countries and their policymakers are generally more receptive to using and trusting models that are specifically designed for their national context rather than generic global models that may not capture important local institutional details or policy instruments. The embedded approach allows for the best of both worlds, maintaining global consistency while providing national detail.

The third area requiring continued development involves improving the quality and resolution of spatial crop data that underlies much of the ecosystem service modeling. The pollination modeling described earlier depends critically on accurate information about where different crops are grown and in what quantities. The EarthStat dataset, while groundbreaking when it was developed, is now becoming outdated as agricultural patterns shift. Newer data products such as SPAM (Spatial Production Allocation Model) and CropGrids offer improved gridded raster datasets of crop production that incorporate more recent information and better methodologies.

Access to better spatial crop data supports more precise modeling not only of agricultural expansion patterns but also of the effectiveness of targeted policies like payments for ecosystem services that may focus on particular crops or regions. Continued investment in developing and maintaining these global agricultural datasets should be a priority for the research community.

The fourth area for advancement involves endogenizing climate change processes within earth economy models. The current GTAP Invest framework takes climate change scenarios as exogenous inputs, meaning that the model runs under assumed climate trajectories but does not model the complete pathway from policy choices to emissions to climate change and ultimately to economic damages. Integrating climate-economy models such as DICE (Dynamic Integrated Climate-Economy model) or FUND (Framework for Uncertainty, Negotiation and Distribution) would allow for more comprehensive analysis of the interactions between land use policies, climate change, and economic development.

This integration would enable the modeling framework to capture feedback loops where land use policies affect emissions, which affect climate, which in turn affects agricultural productivity and land use suitability. These feedbacks are important for accurately assessing the full consequences of alternative policy pathways, particularly for policies that aim to address both biodiversity conservation and climate mitigation simultaneously.

Expanding Ecosystem Service Coverage

The fifth major direction for future research involves expanding the range of ecosystem service linkages that are explicitly modeled within the earth economy framework. Ongoing work continues to improve both the number of ecosystem services that are included and the sophistication with which their economic impacts are represented. For example, current research is exploring ways to link biodiversity changes to natural pest control services and to quantify the economic value of pest control to farmers in monetary terms.

This type of research requires synthesizing field data from ecological studies that document relationships between biodiversity and ecosystem function, and then expressing the resulting changes in ecosystem service provision as shocks within the macroeconomic model. Each ecosystem service presents unique challenges in terms of both measuring the biophysical relationship and translating it into economic impacts, requiring collaboration between ecologists, earth scientists, and economists.

Machine Learning Applications

Finally, a brief note on the potential role of machine learning techniques in advancing earth economy modeling is warranted. New methodological approaches are being explored for improving the SEALS land use change model, including the application of convolutional neural networks and generative adversarial networks (GANs). These machine learning architectures offer promising new approaches for predicting land use change patterns based on complex spatial relationships and for generating artificial landscapes that maintain realistic spatial structures while exploring scenarios that may not have historical precedent.

These computational tools have the potential to improve both the predictive accuracy of land use models and their ability to support modeling of future scenarios that may involve novel combinations of conditions. As machine learning methods continue to develop and as computational resources become more powerful, their integration into earth economy modeling frameworks represents an exciting frontier for methodological innovation.

Closing Remarks and Student Presentations

The semester has been marked by strong engagement from students in grappling with complex modeling concepts and beginning to develop their own research applications. For the final day of class, each student will present their project work, with presentations aimed at approximately five minutes in length. Presentations should include interesting figures that visualize key aspects of the work, discuss the methods that have been employed or are planned, and share any preliminary results that have been obtained thus far.

Students should feel free to reach out if they need an extension beyond the presentation date to continue developing their projects, as the goal is to support high-quality research rather than to enforce arbitrary deadlines. The flexibility that has characterized the course will continue through the final project phase, ensuring that students have adequate time to produce work that represents genuine progress in applying earth economy modeling methods to questions of interest.

Transcript

All right, let’s get started. Welcome to Lecture 13, our last content-bearing lecture for this half-semester course. Today, we’ll be discussing GTAP Invest and the culmination of the earth economy modeling linkage we’ve been working towards.

We’ll review all the different parts of earth economy modeling, highlighting the area that still needs more detail: dependencies modeling. Then, we’ll take a detailed look at the GTAP Invest model and examine some specific results. This is important because it demonstrates key findings and how to express results in this complex modeling space.

I also want to reserve time to discuss future directions for this research and possibilities for students to pursue as research papers, perhaps for your dissertation or other projects.

First, a quick note on the schedule. Today, we’ll focus on topics within CGEs and the GTAP Invest model we’ve been building towards. Regarding deadlines, I’ve seen the outlines come in and they’re looking great. The goal of this course, as stated in the syllabus, is to give you hands-on skills for research in this area. The final project isn’t about producing a polished manuscript with a literature review, but rather showing evidence of good progress in using these tools.

If you find this project interesting, you can turn in and present something on Thursday, the last day of class. It can be a rough work in progress, as long as it shows effort and progress. Extensions are available, even past the end of the grading period. If you’re making good progress, you’ll get an A in this methods course. If you want to take more time with your paper, I’m happy to grade an intermediate step so you don’t feel forced to rush. Just reach out to me individually. We’ve covered a lot of content in this compressed half-semester, so flexibility is important.

Any questions on timing or logistics? Great. Bilateral emails on research ideas have been critical, so thank you for those discussions.

Let’s dive back in and come full circle on earth economy modeling. We’ve repeatedly returned to the diagram of GTAP and INVEST, which underlies many figures we’ve seen. Earth economy modeling is a large space, inclusive of many models that can have two-way linkages between the earth and the economy. GTAP Invest is just one example, plugging in specific models for each step, but we’re expanding to try out different models. Today, we’ll focus on the model we’ve developed in-house.

We’ll discuss GTAP and the land supply functions. We’ve talked extensively about SEALS and INVEST, but today we’ll focus on upscaling ecosystem service changes and expressing them as economic shocks in the economic model. This is how we model the dependencies of the economy on nature.

The reading for today is the 2023 PNAS article, “Investing in Nature Can Improve Equity and Economic Returns.” When publishing in top journals like Nature, Science, and PNAS, the focus is on results rather than methods or models. These journals aim for general interest, so findings must be centered on results. However, many big results rely on new models or methodological advances, so you have to present the methods in supplemental information and focus the manuscript on the key result. In our official GTAP Invest publication, we hid the methods in a long appendix and focused on the keynote result: investing in nature can improve equity and economic returns.

Before diving into results, let’s discuss the steps we took to get there and how they’re similar to what we’ve learned in class. Publishing GTAP Invest involved linking GTAP and INVEST, starting with a soft linkage and moving towards a hard linkage. The first versions were comparative static models, comparing the equilibrium before and after a shock. This approach allows results to be exchanged between model experts without deep knowledge of each other’s models, often ending at the soft linkage.

For recursive dynamic formulations, emailing results for each time step isn’t practical, so we developed a software stack to automate the process. We borrowed the software development approach from NATCAP, creating a user-friendly interface, code implementation, and a GitHub repository for users to fork. Success in open science means scaling up so others can run and adapt the code independently. All parts of GTAP Invest are open source except for the database, which is expensive, and the model solution method, which relies on Gempak. The RunGTAP program uses a limited version of Gempak, a set of mathematical equation solvers custom-built for CGE models. While we’ve made everything else open source, replacing Gempak is challenging. That’s why we’ve started using Julia, specifically the JuMP library, which is promising for fully open-source equation solving. Economists are increasingly moving to Julia for modeling, skipping Python, as open-source development is limited in Stata and MATLAB.

Gempak implements the Tableau language, similar to GAMS, for defining equations at scale and coupling them with efficient solvers. Gempak also includes utilities for tasks like aggregating the GTAP database and sensitivity analysis. It’s both a language and an environment.

Now, let’s focus on the last remaining part of the model linkage: the dependencies model, or upscaling of ecosystem services. We’ll use pollination as an example and discuss how we link it to GTAP. First, let’s consider the broader concept of a shock. How do CGE modelers express a shock to the economy?

One way is shifting supply curves to reflect changes in production, such as a hurricane in Florida shifting the supply curve for a crop. Another common way is implementing a factor-augmenting technical change, which adjusts the effectiveness of one factor relative to another in the CES production function. For example, a technological change making capital more effective causes substitution between capital and labor and changes input-output relationships.

A third approach is a factor-neutral productivity shock, which is industry and region-specific and changes overall productivity. There’s a science to choosing the right shock, covered in CGE courses and textbooks like Burfisher’s. If you’re interested, reach out after class.

One important area for us is endogenizing specific resource pathways, such as land use change. In existing models, you can’t shift a supply curve for land if there’s no supply curve for land—if land is just a fixed endowment. We had to change the models to make land endogenously calculated, so it can be shocked meaningfully.

Let’s make this more specific using pollination. Our goal, as shown in the PNAS article, is to use land use/land cover maps from SEALS, derived from endogenously determined land use change at the regional scale, as inputs to INVEST. Different land use maps result in different ecosystem service provision, such as pollination sufficiency. Above a threshold of 0.3 pollination sufficiency, yields are unaffected, but below that, yields decline. The calculation uses convolution, considering cropland and nearby natural habitat. Large contiguous cropland without fragments of pollinator habitat results in reduced yields for pollination-dependent crops.

This biophysical output from INVEST isn’t enough; we need to convert it to specific changes in yield. We used the EarthStat dataset, which provides gridded yield and production data for 175 crops at 10-kilometer resolution. To determine overall production changes, we considered each crop’s pollination dependency and whether it’s grown in grid cells lacking pollination sufficiency. High-resolution data on crop locations allowed us to sum up pollination-dependent crops and calculate the total dollar value produced in affected grid cells. This gives a change in dollar-level production, which can be aggregated to a percentage change for each region. For example, we mapped the oilseed sector productivity shock.

In the map, white areas indicate no crop production, so pollination sufficiency isn’t estimated there. This map shows service value to people, not where the ecosystem service is supplied. The supply may occur in one location, but the impact on people may be elsewhere.

Once we have the percentage change, we export it as a CSV, convert it to a HAR file, and plug it into the GTAP code. The Tableau language, used by Gempak, defines equations for each region’s price and activity. The A0 variable is adjusted for each region, and the model is re-solved with the new percentage change defined in the shock file. The process is straightforward once the correct variable is identified.

The dependencies linkage is the part of this research area needing the most work. Other areas have established disciplines, but ecosystem services research has focused on valuation from individual preferences, not productivity and macroeconomic impacts. The process for pollination will differ for every ecosystem service, with unique impact pathways. For example, carbon storage has a well-understood pathway from changes in storage to concentrations, warming, damages, and GDP. Other services like sediment retention, erosion, water quality, and habitat provision are harder to model, both economically and biophysically. There’s significant research potential in linking changes in ecosystem services to macroeconomic impacts.

Now, let’s discuss what we can learn from this work. I’ll organize results into three key questions for earth economy modeling:

  1. How much does nature benefit the economy?
  2. Who loses or wins when nature is depleted?
  3. What policies can protect nature’s value and equity?

For the first question, we worked with the World Wildlife Fund on the Global Futures Project, developing estimates using earth economy modeling for different scenarios. We compared business as usual, with continued land use change and emissions, to a global conservation scenario with habitat protection and stabilized land use. The main result was a net present value of cumulative damages from 2011 to 2050 of $9.8 trillion under business as usual. While this is less than the $44 trillion per year estimated by the World Economic Forum, our approach was conservative, considering only well-supported linkages.

Nature matters—a $10 trillion value is significant. More importantly, when we broke down results by who benefits from investing in ecosystem services, we found the percentage change in GDP by 2050 under business as usual was most negative for low-income countries. Under global conservation, the percentage increase was greatest for these countries. This shifted our focus to the equity question: who loses or wins when nature is depleted?

The Dasgupta Review, commissioned by the UK Treasury, asked us to assess distributional impacts of ecosystem service collapses. The review defines demand as human population times economic activity per capita, divided by production efficiency, and supply as the biosphere’s regeneration rate. We provided applied, decision-relevant science, running approved model versions and finding that low-income countries are hurt most by degraded natural capital. We also modeled ecosystem collapse scenarios, finding up to a 10% reduction in GDP and annual damages of $2.7 trillion. These estimates remain conservative, but the key point is that nature matters most for low-income countries.

The third question is what can be done to protect nature’s value and equity. We worked with the World Bank on “The Economic Case for Nature,” analyzing specific policies to mitigate losses from ecosystem services. They wanted actionable policies, not just abstract trade-off frontiers. We identified four main policy options:

  1. Repurposing agricultural subsidies: Removing environmentally damaging subsidies improves market efficiency and reduces externalities. To address opposition, we proposed reallocating subsidy savings as lump-sum payments to landowners, aiming for revenue neutrality.

  2. Investing in agricultural research and development: Increasing yields on existing land reduces the need for expansion and deforestation. We modeled increased government investment in R&D, sometimes using subsidy savings for this purpose.

  3. Payments for ecosystem services (PES): Paying for positive environmental outcomes is an effective market-based tool. We modeled global forest payments, with transfers from high-income to low-income countries to incentivize land protection. We also considered PES policies without global transfers, reflecting most real-world programs.

Policy effectiveness depends on compliance, which is increasingly measurable with remote sensing. PES programs are easier to monitor due to ongoing payments and incentives.

Combining these policies, we found that protecting nature benefits low-income countries most. The combined policies led to the greatest increase in regional welfare for low-income groups, sometimes at the expense of high-income countries. Policies to protect nature are especially beneficial for those most dependent on it.

We also examined trade-offs using a Pareto efficiency curve, plotting economic revenue against environmental quality (mean hectares preserved). The combined policies performed well, but removing subsidies and investing in R&D was preferable if only economic activity mattered. Some policies, like subsidies to landowners, increased GDP but reduced hectares preserved, as landowners invested in expansion. PES policies moved outcomes into the win-win space, especially when combined.

Each policy corresponds to specific land use/land cover maps. For example, PES policies maintain intact forest blocks, while their absence leads to more agricultural expansion. We’re not predicting exact future landscapes, but modeling causally linked pathways of incentives and outcomes.

Any questions on these results? There are more details in the slides, but I’ll move to concluding thoughts.

Where is this research going? This is especially relevant for potential research topics:

  1. Increasing the number of land-intensive sectors in earth economy modeling: So far, impacts are mainly through cropland, pasture, and managed forestry expansion. Other sectors, like energy (especially renewables) and extraction (lithium, aggregates), are increasingly important. Our work shows energy and extractives have a footprint nearly as large as agriculture, but conservation has focused mainly on agriculture. Including more sectors requires detailed data on suitability and expansion potential.

  2. Embedding country-specific models in global earth economy models: Global models aren’t always suitable for country-level policies. Embedding national CGEs within the global framework allows for more detailed national land use change models and policies. Countries are more receptive to using their own models, increasing impact.

  3. Improving spatial crop data: Pollination modeling depends on accurate crop location data. The EarthStat dataset is outdated, but newer options like SPAM and CropGrids offer improved gridded rasters of crop production. Better data supports more precise modeling of agricultural expansion and payments for ecosystem services.

  4. Endogenizing climate change: GTAP Invest currently takes climate scenarios as inputs but doesn’t model the pathway from policy to emissions to climate change and damages. Integrating models like DICE or FUND would allow for more comprehensive analysis, linking with growth models and others.

  5. Expanding ecosystem service linkages: There’s ongoing work to improve the number of ecosystem services modeled. For example, linking biodiversity to pest control and quantifying the economic value to farmers requires synthesizing field data and expressing changes as shocks in the macroeconomic model.

Finally, a note on machine learning. New methods for SEALS, such as convolutional neural nets and generative adversarial networks (GANs), offer promising approaches for predicting land use change and generating artificial landscapes. These tools can improve predictive accuracy and support modeling future scenarios.

Thank you for your engagement this semester. For the last day of class, each of you will present your project, aiming for about five minutes. Include interesting figures, discuss methods, and share preliminary results. Reach out if you need an extension to develop your project further.

Any other questions? Have a great rest of your day, everyone.