Future Spatially Explicit Patterns of Land Transitions in the United States With Multiple Stressors

Abstract

Climate change, income and population growth, and changing diets are major drivers of the global food system with implications for land use change. Land use in the U.S. will be affected directly by local and regional forces and indirectly through international trade. In order to investigate the effects of several potential forces on land use changes in the U.S., we advanced capabilities in representing the interactions between natural and human systems by linking a multisectoral and multiregional socio‐economic model of the world economy to a model that downscales land use to a 0.5°grid scale. This enables us to translate regional projections of future land use into higher‐resolution representations of time‐evolving land cover (effectively spatially explicit land use transitions). We applied the framework over the U.S., with a particular interest in the Mississippi River Basin and its four sub‐basins, to consider how a range of global drivers affect land use and cover in the target regions. Our results show that under scenarios of high pressure on the world food system a comparative advantage in livestock production amplifies the recent trend toward less cropland and more pastures in the U.S. Under low pressures on the world food system agricultural land is used less intensively. However, there can be key differences among the various land‐use transitions at the sub‐ basin scale. Overall, these results highlighted the need for high resolution details to explicitly understand the implications of land use change on environmental impacts such as carbon storage, soil erosion, chemical use, hydrology, and water quality.

Introduction

Climate change, income distribution, population growth, and altering diets are major drivers of the global food system with implications for land use change at global, regional and local scales (Alexandratos & Bruinsma, 2012; Hertel, 2011; Nelson et al., 2014; Springman et al., 2018). As a consequence land use in any particular region will be affected directly by local and regional forces and indirectly through international trade. There are wide-ranging views on how climate change will affect agriculture, how diets will change, and prospects for economic and population growth (Nelson et al., 2014; Popp et al., 2017). There are debates and uncertainty about the general direction of these forces as it could affect land use globally and in the U.S (Hertel, 2011). Conventional studies suggest relatively little change in cropland use in the near-term to 2050 (Schmitz et al., 2014; Villoria, 2019). Some have asked whether the various forces operating globally are a “perfect storm” in the making that lead to transformative consequences on land use (Hertel, 2011). Others have explored a broad range of potential land allocation outcomes (Alexander et al., 2017; Stehfest et al., 2019). Both of these perspectives emphasize the need for a range of exploratory scenarios, including those that may involve more extreme changes. This allows users of scenarios to judge which scenarios are more or less likely.

Land use and land-use change are important considerations in projecting future agricultural and bioenergy supply, ecosystem losses and conservation, climate and environmental change (Hasegawa et al., 2021; IPCC, 2023; Roe et al., 2019; Smith, 2005). Different types of land cover store varying amounts of carbon, and transitions among land cover types can act as either a carbon source or a carbon sink (Houghton & Nassikas, 2017). Different land uses and associated practices can change methane and nitrous oxide emissions as well (Smith, 2005; Tian et al., 2015). Changes in land cover can also affect the earth′s albedo and the water cycle through changes in evaporation, evapotranspiration and run-off (te Wierik et al., 2021). These varying changes can have effects on the global climate and local and regional climates. Large scale integrated assessment models (IAMs) and Multi Sector Dynamics (MSD) models often include projections of land use and cover that are driven by changes in human activity and/or efforts to protect or restore natural land cover (Popp et al., 2017; Reilly et al., 2012; Schmitz et al., 2014). Often, however, these projections are relatively coarse, spatially, with representation of large countries or multi-country regions, whereas General Circulation Models - models simulating the Earth′s climate system - can have spatially gridded resolutions of 0.5° lat. x long. or less (O’Neill et al., 2016). Moreover, IAMs and MSD models often do not track or report specific transitions of individual land parcels, yet ecosystem science finds that land has a long memory such that a parcels ability to take up carbon or act as a sink depends on its use history over decades to centuries (Melillo et al., 2009).

Projecting land use change with needed fidelity at high spatial resolution including the transition of individual parcels of land is a demanding task. There are some efforts to downscale such projections from IAMs and MSD models to finer resolutions. The predictive power of such projections is limited (Alexander et al., 2017; Luo et al., 2023). However, even if there are uncertainties, tools that can downscale land use projections in a numerically efficient manner provide a basis for testing how different downscaling assumptions might affect local and global climate and other environmental metrics. An example of downscaling is that of the Hurtt et al. (2020) methodology which produces maps of broad land use but not of different types of vegetation which are often modeled by IAMs. We overcome this limitation by downscaling major land uses (crops, pasture, forests) as well as specific types of crops and vegetation. As such, the downscaled results provide the ability to assess how land use changes affect emissions projections, the yields of crops, and other factors that could, in principle, provide a feedback loop into an IAM′s projections of land use itself. Here our focus is the first step in this process: testing an efficient downscaling tool.

Developing an efficient framework for fine-resolution land use projections enables the evaluation of future scenarios affecting agriculture, land use, and the environment. Gurgel et al. (2021a) previously investigated how major drivers of agricultural market forces will impact land use changes with a model that resolved the U.S. as a single region, taking a multisector, multisystem dynamics (MSD) perspective (Moss et al., 2016). In the present study, we extend that analysis to a more refined geographical level by combining the strengths of a multiregional economy-wide model (the Economic Projection and Policy Analysis (EPPA) Model or EPPA) with a versatile and efficient land use downscaling model (Demeter) focusing in particular on 4 sub-basins of the Mississippi River basin (MRB). As such, we investigate how global stressors might, alone or in combination, affect local and regional land use change. Assessing the effects of global stressors on the land use in the MRB is a first step toward future investigation of its implications for environmental concerns such as carbon storage, soil erosion, chemical use, hydrology, and water quality.

Various procedures have been developed to downscale land use changes from global IAMs to more refined geographical resolutions (Chen et al., 2020; Hasegawa et al., 2017; Hurtt et al., 2020; Melillo et al., 2009; Reilly et al., 2012; Verburg et al., 2006). Most prominent amongst these methods is that of Hurtt et al. (2020) which downscales land use land cover (LULC) projections from several IAMs for the specific purpose of coupling with Earth System Models (ESMs) which operate at higher resolutions. There are also attempts to build partial equilibrium models with spatially refined economic and environmental driven decisions on land use (Baldos et al., 2020). However, a limit in some of these attempts is the difficulty to adapt the downscaling algorithm to alternative modeling structures and spatial resolutions, or the lack of general equilibrium effects and competition among alternative land uses. Le Page et al. (2016) attempted to minimize such limitations by developing the Demeter downscaling model to produce finer resolution land use projections using coarser projections produced by an IAM. The approach was developed for a specific IAM, but with a goal of being flexible enough to be adapted to other models and varying grid resolutions. Here we extend the efforts of Le Page et al. (2016) to spatially assess land use effects in the U.S. and the MRB from global drivers projected by Gurgel et al. (2021a).

Our contribution is twofold. We developed an integrated Multi-Sectoral Dynamics (MSD) framework combining an economy-wide model with a land-use downscaling model to investigate interactions between natural and human systems, which overcomes challenges related to flexibility, reproducibility, interoperability and connectivity (Moss et al., 2016). It incorporates economy-wide competition among land uses, providing consistent socio-economic feedback and capturing demand-side responses, and uses grid-level spatial characteristics to guide land allocation. Second, we applied the framework to evaluate fine-scale land use implications in the U.S. of a broad range of global socio-economic and climate forces, exploring interactions between demographics, climate, agriculture, and the economy. We utilize previously published results from a MSD model (Gurgel et al., 2021a) that purposely investigated a wide range of land use change scenarios with a goal of investigation the potential for tipping points under extreme land use change scenarios. Previous studies projected land use trajectories at fine resolution for alternative Shared Socioeconomic Pathways (SSP) and Representative Concentration Pathways (RCP) (Chen et al., 2020; Hurtt et al., 2020), where assumptions on several land use forces (such as population, economic growth, trade, climate policy) differ among scenarios. These studies do not allow understanding how each specific force affects land use changes nor to identify which driver is most relevant. To gain a clearer understanding of how each specific force could impact land use allocation in the U.S. we tested changes in one force at a time while holding other forces constant among scenarios.

Methods

The Socio-Economic Model (EPPA)

At the world and regional level, our modeling approach explicitly represents socio-economic behavior and natural resources, and also interactions and feedbacks among them. The multi-region, multi-sector model of the global economy model is an extension of the MIT EPPA model, version 6, a recursive-dynamic computable general equilibrium (CGE) model (Chen et al., 2016; Gurgel et al., 2016). The model is expanded to include links to natural resources, including energy and land resources. It tracks changes in these resources, maintaining consistent accounts of both value and physical terms, and includes estimates of emissions of pollutants and greenhouse gasses from industrial, energy, and land use sources. EPPA represents 18 regions and 28 sectors, including 11 crop and livestock sectors plus a forestry industry, and a number of primary factor inputs (see Table 1). Among them are both depletable and renewable natural capital inputs, as well as produced capital and labor. Among produced capital, EPPA treats cropland, pastures, and managed forest land as “produced” from natural capital of forest areas and grasslands. The economic data is sourced from the Global Trade Analysis Project Version 8 (GTAP 8) database (Narayanan et al., 2012). The model simulates historical economic trajectories recursively to the year 2010 and 2015, and then projects future economic pathways at 5-year intervals from 2015 to 2100. Economic development is benchmarked to IMF's historical data and short-term Gross Domestic Product (GDP) projections (IMF, 2019). Figures S1 and S2 in Supporting Information S1 illustrate the main features and functioning of EPPA.

Table 1 Regions, Sectors and Primary Factor Inputs in EPPA

Regions	Sectors	Primary factor inputs
United States (USA)	Paddy Rice (PDR)	Depletable Natural Capital:	Conventional Oil Resources
Canada (CAN)	Maize, Coarse Grains (GRO)		Shale Oil
Mexico (MEX)	Soybean, Oil Seeds (OSD)		Conventional Gas Resources
Japan (JPN)	Wheat (WHT)		Unconventional Gas Resources
Australia and New Zealand (ANZ)	Sugar Crops (SGR)		Coal Resources
Europe (EUR)	Vegetables & Fruits (V_F)
Eastern Europe (ROE)	Plant Based Fibres (PBF)
Russia (RUS)	Other Crops (OCR)	Renewable Natural Capital:	Natural Forest
East Asia (ASI)	Bovine Cattle		Natural Grasslands
South Korea (KOR)	Poultry and Pork		Solar and Wind Resources
Indonesia (IDZ)	Other Livestock		Hydro Resources
China (CHN)	Forestry
India (IND)	Wood Products
Brazil (BRA)	Food Products	Produced Capital:	Convent.Capital (Bldgs & Machin.)
Africa (AFR)	Coal		Cropland
Middle East (MES)	Crude Oil		Pasture and Grazing Land
Latin America (LAM)	Refined Oil		Managed Forest Landa
	Gas
	Electricity
	Non-Metallic Minerals		Labor
	Iron & Steel
	Non-Ferrous Metals
	Other Energy-intens. Ind.
	Other Industries
	Construction
	Other Services
	Transport
	Ownership of dwellings

EPPA is an economy-wide model which explicitly solves all inter-industry interactions and bi-lateral trade in goods. It represents domestic production, exports, imports, government expenditures, investment and household demand for final goods, and the ownership and supply of labor, capital and natural resources. Microeconomic and macroeconomic balances and consistencies are assured in the benchmark and after any simulation. Physical quantities of energy (exajoules), emissions (tons), land use (hectares), population (billions of people), natural resource endowments (exajoules, hectares) and efficiencies (energy produced/energy used) of advanced technologies are linked with the economic transactions by supplemental accounts. These allow to represent technical efficiencies and availability of renewable resources as also as the physical depletion and use of natural resources. As an illustrative example, scenarios assuming increase in population will lead to higher demand for food and other goods, but also a larger labor supply. These will generate a cascade of economic and environmental effects, from higher agricultural output and more demand for agricultural land, to changes in economic growth and international trade, which are fully accounted for and will lead to a new equilibrium in all markets and economies with a consequent impact on land use and natural resources. Similarly, shocks such as income growth, trade openness, and increase in crop yields will have cascading effects on food demand, economic growth, and inter-industry transactions in several sectors of the economy.

The economic accounting of land-use in the model retains consistency with the supplemental physical accounts on land so that simulated changes in economic use of land translate into hectares of land, preserving total land area constraints in each region. The approach considers five broad land use categories: cropland, pasture, forest, natural forest and natural grass. Several world-scale data sources are reconciled for the purpose of this study. These include the GTAP8 Land Use and Land Cover Database Version 2 which is built from FAOSTAT production data and additional cropland and pasture data (Ramankutty, 2011). Data from the Terrestrial Ecosystem Model (Felzer et al., 2004), using historical land use transitions (Hurtt et al., 2006), complements the land use database. Land and the transformation of natural lands into managed land types in physical terms are represented by a land use transformation approach (Gurgel et al., 2016), which is well suited to longer term analysis where demand for some land uses could expand substantially. It represents conversion costs associated with preparing the soil, spreading seeds and managing the creation of a new agricultural system. Natural areas transformation to agricultural areas are calibrated to a land supply response based on rates of conversion reviewed in the literature (Hertel, 2011).

To assess the value for natural forest that are not currently used we develop a “non-use value” for these land areas using data on the value of standing timber (Sohngen, 2007; Sohngen & Mendelsohn, 1998; Sohngen & Tennity, 2004). This approach assumes that, at the margin, the cost of access to remote timber land must equal the value of the standing timber stock plus that of future harvests as the forest regrows. The net present value of the land and timber is calculated using an optimal timber harvest model for each region of the world and for different timber types. Setting the access costs to this value establishes the equilibrium condition that observed current income flow (i.e., rent and returns) from currently non-accessible land is zero because the timber there now, and in the future, can only be obtained by bearing costs to access it less than or equal to its discounted present value. From these data, we calculate the value of an average standing stock of timber for each region and the separate value of the land based on the discounted present value of timber harvests of forest regrowth after the initial harvest of the standing stock. We further assume that the natural grassland rent relative to pasture is the same as the natural forest rent relative to managed forest.

We assume that land is subject to an exogenous productivity improvement of 1% per year, reflecting assessments of potential productivity improvements showing similar historical crop yields growth, albeit with variations among regions, crops and time (Gitiaux et al., 2011; Ray et al., 2013). Agricultural output can grow by intensification of land use through partial substitution of other inputs and other primary factors in the agricultural production functions as relative prices change over time. Such representation of farmer's behavioral responses under land constraints and food scarcity reflect common strategies on crop and livestock intensification practices which have increased output per unit area over time, notably when food prices trends show persistent rise. These responses are often ignored in agricultural models dealing primarily with bio-chemical-physical aspects.

Most of the output of primary land use sectors (crops, livestock and forestry products) end up as inputs in the food, energy, and other sectors of the economy. Food and agriculture production, and hence the amount of land used is strongly influenced by the growth in population and incomes. Expenditure shares on food tend to decrease as income grows although food consumption levels may increase, which is represented in EPPA by a Stone-Geary preference system (Chen et al., 2016).

The economic behavior represented in EPPA can be summarized into a few economic parameters which, together, determine whether land used for agricultural purposes might increase or decrease under global drivers of agricultural land changes (Gurgel et al., 2021a; Hertel, 2011). An “elasticity of land supply” captures how much agricultural land would expand by conversion of other existing uses given an increase in agricultural land prices. An “intensification elasticity” measures how much farmers would increase the use of other inputs (capital, fertilizers, etc.) in response to rising prices for their agricultural products. An “elasticity of demand” determines how consumers would react to changes in prices of agricultural products. Together, these parameters will determine how much land will increase under changing conditions of the demand and supply of land, population and income growth, changes in diet patterns, yield and productivity changes due to research achievements, higher input uses or climate change. EPPA captures these behavioral parameters, which are further tested in our exploratory scenarios described in Section 3.

Leveraging the EPPA model's strengths in capturing interactions and feedbacks between human and natural systems, Section 3 introduces scenarios previously published by Gurgel et al. (2021a). These scenarios encompass a wide range of global drivers affecting agricultural market and land-use changes in the continental U.S., which are then further downscaled using the EPPA-Demeter framework.

The Spatial Downscaling Model (Demeter)

The downscaling model, Demeter, is an open-source model designed to distribute regional land use projections from economic models to finer spatial resolutions by explicitly modeling land transitions at a pixel level in fractional and physical units (Chen et al., 2019; Le Page et al., 2016; Vernon et al., 2018). Specifically, Demeter combines regional land use projections with pixel level suitability characteristics and user defined transition rules, to model fine resolution transitions between land types. Demeter also operates from time step to time step, where the maps produced for a given time step influence future land cover. Thus, Demeter produces land maps that are spatially and temporally resolved across land types. The main advantage of utilizing a model such as Demeter, is that rather than downscaling LULC using a statistical approach, the LULC is explicitly tracked and downscaled according to relationships (described below) between natural and managed land types at the pixel level. Demeter has been used previously with other IAMs such as the Global Change Analysis Model (Calvin et al., 2019) and GCAM-USA (Binsted et al., 2022) to produce fine resolution projections of Land Use and Land Cover Change (LULCC) for alternative socio-economic and climate scenarios (Chen et al., 2020; Vernon et al., 2018).

Demeter can be used to translate the aggregated regional land use change as represented by EPPA (e.g., loss of corn land, increase in pastures) to gross land use transitions at a pixel level (e.g., increase in pastures from forests, increase in pastures from corn etc.). The modeling framework in Demeter can be expressed by Equation 1. 1 $${\text{Gross}\,\text{land}\,\text{use}}_{l,p,t}=f\left({Map}_{l,p,t-1},{Land\,Use}_{l,r,p,t},{\text{Constraints}}_{l,p,t},{\text{Transition}\,\text{Rules}}_{l},{\text{Process}}_{l}\right)$$ where, the gross land use change of a land type l, at each pixel p, in timestep t is a function of:

• Spatial Map of land in the previous timestep (t − 1) which at t = 1 is the base data map and for t > 1 is the spatial map as modeled by Demeter.

• Regional Net-Land Use change for the same land type l from the regional model (in this case EPPA) at timestep t for the EPPA region r that the pixel belongs to.

• Spatial Constraints defined at the pixel level p for each land type l. Currently, the spatial constraint applied is the proximity of a given land type in a pixel to similar land types in other pixels which is calculated as a kernel density or proximity ratio.

• Transition Rules defined globally by land type l. The transition matrix used in this paper is available in Supporting Information S1 Section 2.

• Process Order can be chosen by the user and determines the sequence in which land types are assigned areas.

For this study, we largely used the base value for parameters prescribed by Chen et al. (2019) who extensively tested the downscaling algorithm. Further discussion is provided in Supporting Information S1 Section 2 including the admissible range for each parameter and the values used in this study. The grid resolution in Demeter can be varied, the limit being the resolution of the base year data that are an input to the model (see Section 2.3). Here our base data are at a 0.5-degree grid resolution.

Linking the Socio-Economic Model (EPPA) With the Downscaling Model (Demeter)

Land use changes at local (grid) level depend on economic forces and drivers solved at regional and global levels, which are represented by EPPA, while sub-regional differences in our EPPA-Demeter integration are captured in the base map used by Demeter, which reflects land cover and land use at the benchmark year. Downscaling rules and assumptions in Demeter assure that land use projections at the U.S. level are distributed at sub-basins and grid cell levels consistently with existing spatial patterns, since Demeter uses grid-level spatial characteristics to guide land allocation and has been parameterized and applied to continental and global LU downscaling (Chen et al., 2020). Future investigation may be performed to test if spatial disaggregation in the CGE framework would change the results, bearing in mind that economic drivers and responses span beyond fine resolution scales.

The basic steps involved in producing downscaled projections in the Demeter-EPPA framework depicted in Figure 1 are:

• Demeter requires an initial base map of LULC for t = 1. We utilize a map from the Community Land Model (CLM 5.0) which is harmonized with fine resolution crop harvest data from IFPRI MapSPAM. See in Supporting Information S1 Section 3 on construction and validation of the base map used to initialize Demeter for this project.

• This base map is used by Demeter to initialize spatial Constraints (specifically the kernel density) which will determine how land will change in the future (e.g., land type l is most likely to expand in or near grids that already have some of that land type).

• Demeter is provided with the regional LULCC projections (from EPPA, in this case). These ultimately dictate the net regional land use across land types.

• Demeter can also be provided with other fine resolution constraints (e.g., Demeter can be provided with a map of soil quality so that crops are most likely to be planted where nutrients are available). We currently don't provide any explicit constraints other than the kernel density in point number 2 above.

• We also provide Demeter with global level parameters, such as Transition Rules so that certain land types are given priority over others when land transitions occur that along with the regional LULCC determine the new allocation of land types in each grid. The overall regional land use area allocations provided by EPPA are not affected by the Demeter transition rules. We note that the global transitions are used in conjunction with regional economic signals (as seen in point 3), spatial suitability determined at the pixel level (as seen in point 1) and any other spatial constraints (as seen in point 4).

• Demeter utilizes steps 1–5 above to generate “gross” land use maps for all land types at a grid level.

• Demeter also produces transitions between land types for each period, such as illustrated in the Sankey plot in Figure 1 below.

  Note that the final output is a map of LULC at time t which forms the base map for time t + 1.

[IMAGE OMITTED. SEE PDF]

The downscaling of EPPA regional results by Demeter requires mapping land use categories between the two models (Table 2). The base map in Demeter treats forest areas as a broad land use category, so we map EPPA's natural forests and managed forests to forests in Demeter. Similarly, Demeter does not distinguish between natural grasslands and pastures, which are combined as a single category from EPPA to Demeter. Demeter tracks separately land use by each EPPA agricultural sector, which allows us to break-up gross transitions by different crop types from cropland projections in EPPA.

Table 2 Land Use Categories Represented in EPPA and Demeter and Land Use Mapping Between Them

Land use categories in EPPA	Land use categories in Demeter	Further aggregated land use categories
Natural Grassland	Pasture and Grazing Land	Pasture and Grazing Land
Pasture and Grazing Land
Natural Forest	Forests	Forests
Managed Forest
Cropland
Paddy Rice	Paddy Rice	Paddy Rice
Coarse Grains	Corn	Corn and Oil Seeds
Oil Seeds	Oil Seeds
Wheat	Wheat	Wheat
Sugar Crops	Sugar Crops	Other Crops
Vegetables & Fruits	Vegetables & Fruits
Plant Based Fibres	Plant Based Fibres
Other Crops	Other Crops
Other Land Uses	Other Land Uses	Other Land Uses

Scenarios

Gurgel et al. (2021a), through a survey of the literature, determined a high and low value for each of 8 different stressors intended to capture potential divergence in the strength of these global forces from the BAU projection. Stressors are alternative socio-economic and environmental forces and drivers influencing decisions on land use, which may lead to higher or lower needs for agricultural land. The paper also considers the combined sets of forces that together would put the greatest pressure on the U.S. land use (high all) or put the least pressure on the U.S. land use (low all) as shown in Table 3. These, together with a business-as-usual (BAU) configuration where all values are at median levels, creates 17 separate scenarios (Table 3).

Table 3 Scenarios (Shocks Applied to All Regions of the Model)

Name	Brief explanation
BAU	Baseline scenario
low trade	Less trade due to higher import tariffs globally (tariffs 50% higher)
low clim. imp. crops	Positive climate impacts on crop yields from Global Gridded Crop Models (GGCMs)
low clim. imp crops&livest.	Positive climate impacts on crop and pasture yields from GGCMs
low yield constraint	Higher annual increase in crop yields (1.5% per year)
low meat demand	Changing diets toward lower income elasticity on meat demand
low pop. growth	Lower population growth (1% lower than BAU)
low econ. growth	Lower GDP growth (20% lower than BAU)
low all	All “low” impacts together
high trade	More trade due to lower import tariffs globally (tariffs 50% lower)
high clim. imp. crops	Negative climate impacts on crop yields from IPCC local crop models
high clim. imp crops&livest.	Negative climate impacts on crop yields and livestock from IPCC local crop models
high yield constraint	Lower annual increase in crop yields (0.5% per year)
high meat demand	Changing diets toward higher income elasticity on meat demand
high pop. growth	Higher population growth (1% higher than BAU)
high econ. growth	Higher GDP growth (20% higher than BAU)
high all	All “high” impacts together

The shocks applied in our counterfactual scenarios cover most of the possible drivers and stressors affecting future land use changes discussed in the literature (Hertel, 2011; Stehfest et al., 2019) and assume changes in parameters of the model from its initial calibration point. In the low trade scenario, trade barriers were increased by 50%. The barriers were reduced by 50% to induce high trade. In the low climate impacts scenario crop yields and/or livestock productivity and pasture yields were shocked by the average impacts from global gridded crop models (Blanc, 2017) that project higher crop yields in most cases (Gurgel et al., 2021b). The high climate impacts scenario assumes a central value of crop yield impacts from the Intergovernmental Panel on Climate Change (IPCC) 5th Assessment report (Porter et al., 2014), with mostly negative effects on yields but varying by region. We assumed changes in yields from Global Gridded Crop Models and IPCC for rice, wheat, maize and soybean, while for Other Crops (OCR) and pastures and livestock, we assumed the simple average change in yields from those four major crops, following the approach in Gurgel et al. (2021b). The low yield constraint scenario assumes a higher rate of increase in exogenous crop yields due to faster research and development and diffusion, while the opposite (lower rate of increase) is assumed in the high yield constraint. Upper and lower levels of improvements in yields in these scenarios are in agreement with the literature (Ray et al., 2013) Decreases in the income elasticity of demand for meat affect the preferences for meat in developed countries in the low meat demand scenario, and the opposite change in income elasticity is considered in the high meat demand scenario. The low population growth and high population growth scenarios assume population growth 1% lower and higher than in the BAU, respectively. Similarly, low and high economic growth scenarios assume 20% lower or higher growth in GDP than BAU.

Results and Discussion

We report the results in 4 sections. In Section 4.1 we summarize results for net land use change for the U.S. and the total global land use for all 17 scenarios from EPPA. In Section 4.2 we report Demeter results for the U.S. Section 4.3 provides mapped results for the U.S. at the grid level. Section 4.4 focuses further on the MRB, plotting land use change results for the major sub-basins in the river system (Upper Mississippi, Lower Mississippi, Ohio River, and Missouri River basins).

Scenario Results From EPPA (Net Land Use)

Some of the simulated forces affected primarily the agriculture sector, such as changes in yields. These initially drive farmers' responses on land intensification and land conversion among land use categories. As EPPA captures the inter-relationship among sectors and regions, these responses will indirectly impact the international agricultural competitiveness of each country and drive changes in the rest of the economy. Other simulated forces, such as population and economic growth, changes in diets, and trade policies, initially impact the entire economy and are transmitted to land allocation by shifts in the demand for agricultural goods and overall competitiveness. The land use resulting from these forces highlight the dynamics and interactions in a complex system.

Our land use projections for the world under BAU conditions in the EPPA model suggested increases in pasture and cropland globally through conversion of natural forests and natural grasslands (Figure 2, top panel). In the U.S., however, the BAU trajectory is characterized by expansion of natural forests and pastures and decreases in total cropland, coincidentally following recent trends as reported by FAO and the USDA (see Figures S7 to S10 in Supporting Information S1). Considering multiple future scenarios of diverging strength of forces affecting land use, these trends are intensified under higher pressures for agriculture land or reduced under lower pressures. As an example, FAO data suggests pasture areas in the US have grown by 7.7% from 1990 to 2022, and our results show an additional 7.5% growth in pasture areas from 2020 to 2050 under the BAU scenario and 12% under the high population scenario.

[IMAGE OMITTED. SEE PDF]

By focusing on changes in land use, Figure 2 may lead to the impression that large changes are projected. In fact, most scenarios show negligible change in the overall allocation of land uses in the U.S. when total land use by type over time are plotted (See Figures S11 and S12 in Supporting Information S1). The high all is something of an exception as it entails considerable growth in global meat demand and, with it, increased pasture area especially in the U.S. owing to its comparative advantage in livestock production (Adcock et al., 2006; Chen et al., 2017; Crespi & Saitone, 2018). The economic principle of comparative advantage refers to a country's higher relative productivity (Ricardo, 1817) due to aspects such as resource endowments, technology, labor skills and productivity, infrastructure, economies of scale, climate and geography, business practices and government policies (Krugman et al., 2018). Dating back to Ricardo (1817), economists have recognized the advantages of trade even when one country has an absolute advantage in producing all goods, if other countries have only a relative advantage in producing some goods. In such a case, a country with an absolute advantage in all goods would benefit by focusing more production on goods in which it has a relative advantage, exporting those goods, and importing goods in which it has a relative disadvantage. The high all scenario, as originally published, is an intentionally extreme scenario as all the driving forces are operating in a direction that increases pressure on land use change, particularly on pasture expansion since the scenario assumes negative climate effects on livestock productivity, changing diets toward meat consumption, and more rapid economic and population growth biasing diet choices even more toward meat consumption.

The BAU scenario is in line with conventional views on global agriculture development through 2050, similar in broad trends to projections by the OECD and FAO (FAO, 2018; OECD-FAO, 2023). In general, we did not find evidence of strong deviations toward large deforestation or toward land abandonment in the scenarios tested (Figure 2). And the increase in pasture area and reduction in cropped areas in the US is consistent with USDA reported trends since ∼ year 2000 (See figure S7 in Supporting Information S1). While in the BAU scenario we saw a continuation of recent trends in the U.S. the high- and low- pressure scenarios affected the strength of this shift but neither reversed it nor magnified it dramatically (Figure 3), with some few exceptions. For most forcings, the low scenarios showed less pasture, more cropland, and some additional forest areas compared with the BAU. Conversely, the high scenarios showed more pasture, less cropland, and somewhat less forest area than the BAU. In no case were these differences enough to change the general direction of change in the BAU of more pasture and less cropland. This conventional view embodied in various projections as reviewed earlier thus appears robust to the range of global driving forces investigated. The biggest divergence from the BAU in these projections is the substitution among managed agricultural areas, from cropland to pastures under high pressures, most significantly under high econ. growth and high all (Figure 3). In general, the high scenarios lead to higher global demand for livestock products, and with comparative advantage in the U.S. over other regions in livestock production, an increase in the trend toward more pasture that has been observed over the past few decades.

[IMAGE OMITTED. SEE PDF]

A general increase in pasture and grazing areas, resulting from growing demand for livestock products as incomes increase, is well aligned with the economic theory and recent empirical findings. Among the many factors driving diets toward more protein-rich food, income growth is often seen by economists as the main driver, as confirmed by recent empirical research (Bodirsky et al., 2015; Milford et al., 2019). The high econ. growth and high all scenarios enhance this trend. Population growth and attendant urbanization has been pointed to as a second major driver (Milford et al., 2019), which is captured by the high pop. growth and high meat demand scenarios. Note that the high and low values of drivers are considered one-by-one except in the high all and low all. Thus, the results suggest that economic and income growth is by far the strongest force for increased livestock product demand. Other factors driving diets, such as changes in prices and their impacts on demand and supply of alternative food sources, are endogenously accounted for in our socio-economic model. The low meat demand scenario was intended to capture potential trends toward lower consumption of meat, perhaps driven by environmental, ethical, or diet concerns but, at least as explored here, it is more than offset by increasing income. A more expansive investigation of changes in dietary preferences is warranted in future studies.

Although results across studies are not easily comparable given differences in modeling approach and set up, regional coverage, definition of land use categories, and scenario design, some broad numbers and trends from studies indicate both similarities to our findings and differences. Stehfest et al. (2019) compared global land use projections across six models under alternative socio-economic pathways (SSPs). Each individual model estimated between −10% and +10% cropland changes in 2050 going from SSP2 to SSP1 and SSP3, respectively, while only one model achieved −20% and another +18% changes. In our study, cropland area in the U.S. changed by −8.3% in the BAU scenario by 2050, while in most of the individual stressor scenarios it varies between +0.3% and −14.8% under alternative drivers (or stressors). In our high econ. growth and high all scenarios we found cropland changes approached −20%. Thus, like the Stehfest et al. (2019) comparison, we find such large changes unlikely. Our estimates on global cropland changes through 2050 are narrower than those in Stehfest et al. (2019), varying from −5% to +6%. In another model comparison study, Schmitz et al. (2014) shows that four out of ten models projected decreases in cropland in North America, which may reach −16% by 2050 under the SSP2 scenario, while six out of ten models project decreases under SSP3. Another set of studies focused only on the U.S. land use also project decreases in cropland. These include studies using a set of econometric estimates with fine spatial resolution under alternative policy scenarios (Radeloff et al., 2012), varying assumptions of future crop demand aligned with historical trends (Lawler et al., 2014) and under climate impacts on agricultural prices (Haim et al., 2011). Different from our findings, these studies projected decreasing pasture and rangeland areas, while most of our scenarios suggest larger pasture areas in the future, consistent with recent trends. A possible explanation for this difference is that most of these studies cover a relatively short period (1992–1997) when pasture areas were decreasing in the US. Statistical or extrapolative scenarios based on that period would have missed the overall trend of pasture increases after 1997 and from a larger historic period (1978–2017). Further, as these studies focus only on the US, they may not account for changes in trade due to growing meat demand outside the U.S. with consequential changes in prices and income, and the U.S. competitive advantage in meat production. Their approach also does not account for general equilibrium effects, such as endogenous intensification responses.

EPPA trends of expanding pasture areas in the U.S. are in agreement with projections from the Land-Use Harmonization 2 (LUH2) (Hurtt et al., 2020) for scenarios SSP4-RCP3.4, SSP4-RCP6.0 and SSP3-RCP7.0, since pasture areas grow by 15%, 24%, and 14% by the end of the century relative to 2015 in these scenarios, respectively. Also, Figures 2 and 3 in our paper show that pasture and grassland areas in our projections are fairly stable or even decreasing under some scenarios applying “low” levels of stressors, which is consistent with Chen et al. (2020) scenarios SSP1-RCP6.0, SSP3-RCP6.0, SSP4-RCP6.0, and SSP5-RCP8.5. Although our research goal and scenarios are not directly comparable to those in Chen et al. (2020) and LUH2, we noticed that expansion or relatively stability in pasture area is a common result under RCP scenarios associated with higher greenhouse gas emissions combined with SSP scenarios with higher challenges on mitigation and/or higher population and economic growth. These are consistent with our scenarios, since we don't assume climate policy targets in any of them. Two aspects help to explain the common finding of pasture expansion in some SSP and RCP scenarios by those studies and in our research: (a) higher population and/or economic growth are associated with higher demand for livestock-based food; (b) as GHG emissions from livestock and meat-based diets are not penalized when climate policies are neglected, livestock production does not face constraints to increase and require larger areas.

The increase in pasture area, consistent with LUH2 results (Hurtt et al., 2020) for scenarios SSP4-RCP3.4, SSP4-RCP6.0 and SSP3-RCP7.0, obviously reflect a relative increase in demand for meat. Many of the “high” stressors imply relatively greater demand for meat than for food grains or other vegetable crops. Reflecting existing econometric evidence (Reimer & Hertel, 2004), the baseline parameterization of EPPA includes a somewhat higher income elasticity for meat consumption than for other food consumption. Thus, it is not surprising that faster economic growth, implying faster income growth results in an increase in pasture area. The high climate impacts scenario assumes a 10% reduction in livestock productivity, and thus requires more pasture land for a given amount of livestock production. And finally, assumptions that this bias toward meat is either increased or decreased. The high all scenario includes the assumption that the meat bias increases. Looking at the results for individual scenarios, we can see that the strongest effect on meat demand and on pasture is the higher rate of economic growth.

EPPA differs from other similar IAM models used in previous studies (Chen et al., 2020; Hurtt et al., 2020; Stehfest et al., 2019) in some aspects. The general equilibrium economic approach in EPPA distinguishes from the partial equilibrium approach in several IAMs by considering the full “circular economic flow” of money, goods, and services between households and all industries in the economy, which means that supply and demand of goods, services and resources need to consistently match in all domestic and international markets simultaneously. It includes the proper balance between savings and investments in capital markets and the balance on international bilateral trade and flow of capital among regions. As such, EPPA may provide different results and insights from partial equilibrium models used in other studies although the general trends in land use changes may move in the same direction under similar scenarios, as discussed in the previous paragraphs.

Aggregated LULC From Demeter-EPPA

EPPA results from the 17 scenarios were downscaled by Demeter to 0.5°grid scale to investigate the effects of several potential forces on land use changes in the U.S. Figure 4 shows the U.S. trajectories for 10 land-use categories generated after running Demeter. We highlight three scenarios, the BAU, the high all and the low all (as defined in Table 3). The range across all 17 scenarios is shown as the box plots. In the BAU, pasture and grassland areas in the U.S. is the land category changing the most, increasing by 200 thousand km² by 2050 from 2500 thousand km² in 2020, but remained flat at 2020 levels in the low all scenario. Deviations from the BAU range from −6% to +13% by 2050 for pasture and grazing areas, except in the low all and high all scenarios. Under the high all scenario, pasture and grazing areas increased by 600 thousand km² (24%). Areas of forest, corn, oil seeds and other cereals also faced variable trends under alternative stressors. By 2050, forest areas increased by 25 thousand km² in the low all scenario from 2,975 thousand km² in 2020, or decreased by as much as 150 thousand km² in the high all scenario. A similar range of changes is seen for other cereals. Corn and oil seeds areas increased by up to 25 thousand km² during any period of the simulated time horizon, but overall, corn area is lower by 2050 compared to 2020 in all scenarios, decreasing between 25 thousand km² and 100 thousand km². The high all and low all scenario are meant to explore the extremes as it is unlikely that all driving forces will reinforce each other to maximize or minimize land use pressures. Regarding individual crop types, changes in single forcing scenarios relative to BAU were in the range of ±7.8% for plant-based fibers, −3.3% to +6.3% for wheat, −5.1% to +3.5% to other cereals, −2.7% to +5.1% to oil seeds, −1.9% to +3.7% for sugar crops, ±2.7% for vegetables and fruits, −1.2% to +2.4% for corn, and ±2.2% for rice.

[IMAGE OMITTED. SEE PDF]

Downscaled Maps of LULC From Demeter-EPPA

Here we explore the spatial results of our EPPA-Demeter downscaling framework at a 0.5°grid resolution. Figure 5 shows the downscaled net-land use allocation in the U.S. as the differences in fractional land cover in 2050 as compared to 2020 for selected high scenarios that show the most change. Figure 6 shows the same plots for similar low scenarios. For presentation purposes, we have grouped some crops so that we have 5 categories of land uses (see last column in Table 2). These categories are: Wheat, Corn and soybean (coarse grains and oilseeds), Pasture and grazing, Forest, and OCR (Paddy Rice, Sugar Crops, Vegetables and Fruits, Plant Based Fibers and OCR).

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

Overall, the downscaled projections indicate the areas with increasing pasture faced more pronounced changes than other categories of land use, consistent with the U.S. regional projections from EPPA (i.e., Figures 2 and 3). Since land use changes must balance in the aggregate, the increase in pasture and grazing means that most other uses are decreasing with some minor exceptions. Comparing the alternative forcing scenarios with the BAU, the high all scenario caused the greatest change in land use, not surprisingly as it combines all the incentives and weaker impediments. The high economic growth scenario also created substantial change, suggesting that forcing carries the dominant effect as compared to the others. When these high forces are in place, Pasture and grazing land expanded mostly in areas where it already exists, and mostly at the expense of cropland. Most of these transitions were observed in the MRB (highlighted in the upper left map of Figure 5). As noted in the previous section, focusing on changes only can give the impression of a widespread and extensive change when, in fact, most land is not changing. Even though there is a substantial increase in pasture, for example, in the high forcing scenarios the predominant land uses in the 2050 remained the predominant land uses in the BAU (see Figures S13 and S14 in Supporting Information S1 with maps displaying absolute land use shares). For example, Corn and soybeans are still the predominant land use in the corn belt. The increase in pasture areas in the high scenarios is consistent with the current technology mix where livestock and beef production heavily rely on pasture/grazing areas. The EPPA model assumes some substitution among land and other resources and inputs, capturing potential intensification and switches toward more feedlots, but these seem limited in our results and may ignore potentially radical changes to livestock production (e.g., less pasture/grazing and more feedlot production), which may be a caveat to our approach. On the other hand, grass fed beef has some environmental and dietary benefits which may favor a persistence of this system in the future.

The low scenarios (Figure 6) reduced the need for more pasture areas. By 2050 (compared to 2020) forest and pasture/grasslands were nearly unchanged in the low all scenario. The underlying EPPA projections show some net changes in natural forest and managed forests. The Demeter base map does not distinguish between managed and natural forest, showing total forest changes to be small relative to total forest area in that scenario. The same happens in the case of pasture and natural grasslands. EPPA projected some transitions between these two categories, but the aggregated pasture and grazing lands remained fairly stable from 2020 to 2050 in the low all scenario.

Overall, the high and low impacts are not symmetric (Figures 5 and 6), since several high scenarios lead to land use changes toward less forest areas, while low scenarios don't increase such areas. That is, in low scenarios where the pressure on land use for agricultural purposes was reduced, land was not abandoned but rather just used less intensively, implying lower output per unit area.

Comparing Land Transitions Across Sub-Basins

In the prior section, we found an extensive range of land-use changes are possible across the U.S. under a variety of climate and socioeconomic factors - and that many of the more salient responses take place within the MRB. Taking a closer look, land uses are quite different across the MRB that includes drier areas in the West, cooler areas in the North, greater precipitation in the East and warmer conditions in the South. To provide some quantification of how these different underlying conditions may affect land use change, we aggregated grid level results for 4 sub-basins of the MRB (the Upper Mississippi, Lower Mississippi, Ohio River, and Missouri River sub-basins).

We calculated the gross land transitions between land types (19 different transition types). For simplicity, we aggregated all transitions between 2020 and 2050. Figure 7 shows these cumulative land transitions for the high all scenario, the low all scenario and the BAU for all four basins over 2020 to 2050. We focus on these extreme scenarios with the intent of bounding the range of outcomes, and perhaps challenge conventional thinking. As should be evident from the previous results, there is very little difference from the BAU for most of the scenarios, and in aggregate the BAU largely extends recent trends in U.S. land use. Note that Figure 7 is in effect a more detailed breakdown in the land use change as shown in Figure 3. A major conclusion of comparing MRB sub basins is that there can be different levels and types of transitions in any one scenario or across scenarios. The latter is hardly surprising as the differences in net land use change from EPPA will necessarily create differences across scenarios when downscaled. Less obvious is the extent to which salient differences occur across basins for a single scenario. However, such differences show up because of differences in base conditions (what land uses are present in different grid cells in the base year map) which is a major driver of the allocation rules in Demeter.

[IMAGE OMITTED. SEE PDF]

In more detail, under the BAU scenario, the total area of land change is largely the same (i.e., within 25% of one another) across three of the MRB sub-basins, except for the Ohio River that sees the lowest total area of transitions. The largest transition under the BAU scenario is that of more Pasture and grazing land area converted from Corn and soybean area in the Missouri basin (∼3 thousand km2). There are also increases in Pasture and grazing land in the other three Mississippi sub-basins from conversion of Corn and soybean land. At the same time, there is ∼2 thousand km² of land converted to Corn and soybean from OCR in the Upper Mississippi sub basin. That is, overall areas in corn and soybeans are decreasing in the U.S. but are increasing in the Upper Mississippi and the Ohio River basin. The low all scenario on the other hand sees less land transition compared to the BAU and a net decrease in Pasture and grazing land for all sub- basins. This is expected given the relatively low levels of regional land changes (provided by EPPA projections) in that scenario.

Under the high all scenario however, there are larger gross land changes. Firstly, large amounts of Corn and soybean land are converted to Pasture and grazing land, most notably across the Missouri, Ohio, and Upper Mississippi sub-basins. This is expected since there is a large amount of land that is suitable for pasture and grazing in these sub-basins and reflects Demeter's settings that favor expansion of a land type in or near a grid where it already exists. Overall, the Missouri river sub basin sees the largest changes in land in absolute terms (∼30 thousand km2) compared to all other basins. In the Lower Mississippi basin, more of the land in OCR is replaced with Pasture and grazing land (∼4 thousand km2) to accommodate the high levels of meat demand projected by EPPA in this scenario. In the Ohio river sub-basin, there is relatively less Corn and soybean land converted to uses other than Pasture and grazing land, and some Forest land transitions to Corn and soybean land (∼1.6 thousand km2). Similarly, in the Upper Mississippi sub-basin, there are additional ∼8 thousand km2 of Corn and soybean land mainly through conversion of OCR and Forest lands.

Overall, the sub-basin scale results are driven by land suitability and availability at the grid level as well as regional demands. Firstly, there is an increasing demand for Pasture and grazing land in the U.S. under the high all scenario with increasing demand for meat and dairy products. In the Missouri river sub-basin, which already has land suitable for Pasture and grazing, there is a large increase in that land use. In the Lower Mississippi sub-basin, with land in OCR more proximate to Pasture and grazing land, there is a loss of that land given the lack of regional demands. The EPPA projections have demand for land in Corn and soybeans in the U.S. rising through 2035 and then falling under the high all scenario (Figure 4). This trajectory results in some regions with land suitable for Corn and soybeans production, continuing to transition land in OCR toward Corn and soybeans (e.g., Upper Mississippi and the Ohio River basin). Understanding the grid level evolution of LULC allows us to understand the heterogeneity of responses underlying regional land change.

Conclusion

In this study, we investigated how different global forces could affect future land use in the U.S. at the regional levels, with a particular interest in four sub-basins of the MRB. This is a large area, with a large proportion of diverse agricultural uses including corn and soybeans dominant in the Ohio and Upper Mississippi River sub-basins, wheat and pasture and grazing more dominant in the Missouri River sub-basin, and a mix of OCR and forest in the Lower Mississippi sub-basin. We combined a multi-sectoral and multi-regional socio-economic MSD model of the world economy with an open-source downscaling model, which allows us to translate the regional land-use projections (net land use change) from the socio-economic model to higher-resolution, gridded representations of time-evolving land cover (gross land use). We investigated 17 scenarios that considered how various drivers of the global food system could affect land use worldwide. These included ranges of economic and population growth, climate impacts on crop and livestock productivity, changes in yields, international trade, and changes in diets.

Our results show that recent past trends in the U.S. are likely to continue in the future under BAU conditions, with pasture and grazing lands increasing through 2050, a trend that has existed in the recent past, while cropland decreases (Section 4.1). This result reflects the U.S. comparative advantage in livestock production (Adcock et al., 2006; Chen et al., 2017; Crespi & Saitone, 2018). The alternative forcing scenarios affected the strength of these changes, but with minor to moderate impacts in land use changes at the grid and sub-basin level. The main exceptions were under a scenario of high economic growth, or when we combine all pressures related to higher needs for agricultural land. We then see an amplification of the trend toward more pasture and grazing land and less cropland, again reflecting the comparative advantage the U.S. has in livestock production. When observing the downscaled results of gross LULC from the Demeter-EPPA framework (Section 4.4) we find that there is considerable heterogeneity in land transitions at the grid level and, therefore, also at the basin level for any given scenario and across scenarios. In a scenario when all global pressures are at a high level, there is a relatively large transition of corn and soybean land to pasture and grazing land, especially in the Missouri river basin. However, in the Lower Mississippi basin, OCR (other than corn, soybeans and wheat) are transitioned to pasture and grazing land. In the Ohio River basin, while there is a transition toward pasture and grazing land, there are also relatively large transitions toward corn and soybean at the expense of forest cover. These results are thus driven not only by regional economic pressures but also the evolving patterns of land use that are determined by Demeter, which tends to convert land to a given type when and where it is already prevalent.

There remain various uncertainties in our projections, which require further investigation. In particular, how results depend on fundamental behavioral parameters in both models can be further tested. In the case of EPPA, previous work has shown low sensitivity to most parameters and moderate sensitivity to the agricultural intensification elasticity parameter (see in Supporting Information S1 Section 7 and Gurgel et al., 2021a). While the general low sensitivity of results suggests our general findings are robust, if the intensification elasticity is much lower the result could be greater land-use changes than currently projected. In the agricultural economics literature there are different views as to whether yield growth is largely an exogenous force, unresponsive to pressure for more output (low intensification elasticity), or if it responds to those pressures (higher intensification elasticity). And, because any response can be slow it has been difficult to statistically determine this elasticity. Another aspect which deservers future investigation is that EPPA does not capture urban area expansion competition with agricultural uses. Overall urban areas in the U.S. occupy only about 3.5% of the total land area (Figure S16 in Supporting Information S1) so further expansion would have limited impact on the total of other land uses which are much larger. However, for downscaled projections at fine resolution such changes could have important effects in urban fringe areas. Application of our results in such areas would require further work to consider urban land expansion. Other potential areas of investigation include changes in land ownership, the role of conservation programs, and explicit consideration of irrigation and water availability.

Natural grassland areas in EPPA are very unresponsive among scenarios in the U.S., which means that changes in pasture and grazing land in Demeter are due almost entirely to changes in pasture areas in EPPA. Combining natural grassland and pasture when downscaling land use is common practice in the literature reflecting the lack of downscaled data. Distinguishing pastures from natural grassland areas in Demeter is one of our current efforts to improve the EPPA-Demeter connection.

For Demeter, it will be useful to develop methods within the algorithm to consider how changing climate could affect suitability for different uses of land. In addition, there is room to test Demeter's downscaling parameters. We included a sensitivity analysis of the intensification ratio (which defines how much of an expanding land use category will be placed on grid cells where it already exists, and how much will be placed in nearby grid cells) in Supporting Information S1 Section 3. Exploring other sensitivities and methodological advances is a focus of ongoing work. In the end, the value of detailed land-use scenarios is in using them in physical land system models to investigate the earth system consequences of these changes. This is where a finer resolution of land cover changes is particularly important. While that work is ongoing, the transition toward more permanent cover (e.g., pasture and grazing) and away from row crops (corn and soybeans) could mean reduced erosion and run-off, imply changes in the hydrological cycle as corn is more water intensive than grass, as well as having implications for ammonia use and air quality. However, as we saw, the changes can be regionally heterogenous, so generalizing the direction of change may oversimplify the implications of these scenarios. Moreover, land area is only one part of the story. If reductions in land area for corn and soybeans is accomplished by increasing yields, in part, through a more intense use of fertilizer, there would be an offsetting increase in ammonia use and attending air quality concerns.

Developing an efficient downscaling approach is a first step toward in creating dynamic connections between human and ESM components that operate across different spatial scales, and ultimately enables more explicit multi-sector feedbacks between socioeconomics, bio-geophysics, biogeochemistry, atmospheric chemistry, and land use changes and subsequent effects on land-energy-water resources. In particular these results are relevant to further understanding the implications of land use change on carbon storage, soil erosion, chemical use, hydrology, and water quality. The employed downscaling model facilitates interoperability among models and across various spatial scales and can be readily applied to other basins or regions of interest.

Acknowledgments

This material is based upon work supported by the U.S. Department of Energy, Office of Science, under Award Number DE-FG02-94ER61937. This report was prepared as an account of work sponsored by an agency of the United States (USA) Government. Neither the USA Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the USA Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. The authors also gratefully acknowledge financial support for Integrated Global Systems Model (IGSM) development by other government, industry, and foundation sponsors of the MIT Center for Sustainability Science and Strategy. For a complete list of sponsors and US government funding sources, please visit . Mr. Narayan's time on this project was supported by the U.S. Department of Energy, Office of Science, as part of research in Multi Sector Dynamics, Earth and Environmental System Modeling Program.

Data Availability Statement

The Demeter model and all 17 EPPA runs used in this study are available at Zenodo via (Narayan et al., 2024).