Coupled social and land use dynamics affect dietary choice and agricultural land-use extent

The study investigates how socially driven dietary choices interact with agricultural land-use dynamics to influence global land-use trajectories. Prior work often treats demand as exogenous, while supply-side models dominate land-use projections. The authors posit that social learning and socio-economic factors affect adoption of eco-conscious diets, which in turn feed back on global land use, creating two-way coupling. The purpose is to introduce and analyze a minimal coupled social-and-land use model at the country level for 153 countries to: (1) demonstrate dynamics emergent from coupling social behavior with land use that are absent when modeled separately; and (2) gain insight into how these couplings modify projections of global land use and dietary trends. The importance lies in potentially large, policy-relevant changes in projected agricultural land use when social feedbacks are included.

The paper situates itself within research on sustainable food systems and land-use modeling, noting that many models emphasize supply-side factors while using exogenous demand scenarios. Ensemble land-use models (e.g., those used in IPCC assessments) often prescribe dietary patterns rather than modeling their evolution. Empirical and theoretical work indicates diets strongly influence land use, and individuals incorporate environmental considerations into dietary decisions. Theoretical frameworks from evolutionary game theory and public goods games have been applied to social behaviors (e.g., vaccination, fisheries exploitation) and to coupled socio-environmental systems (e.g., socio-climate models), but have not been used to explicitly couple global dietary social dynamics with land-use feedbacks. The authors draw on this literature to justify modeling social learning in dietary adoption as endogenously driven by utility that depends on income, non-income factors, and observed global land use.

The authors develop a coupled country-level model linking dietary social dynamics to land-use demand. Populations in each country i are partitioned into below-poverty (fraction p_i(t)) consuming a sustenance diet, and above-poverty (1 − p_i(t)) who can adopt either an eco-conscious or land-intensive diet. Land-use demand per capita is defined for sustenance, EAT-Lancet (eco-conscious reference), and maximum land-impact diets, converted from caloric food group compositions to land area using a previously published model that accounts for domestic production and imports, thus capturing effective global land required by a country's consumption. Total land use from country i is L_i(t) = [c_s,i(t) p_i(t) + c_i(t) (1 − p_i(t))] P_i(t), where c_s,i is sustenance per-capita land and c_i(t) is the average per-capita land use of the above-poverty population, inferred from the share of eco-conscious vs. land-intensive diets. Social dynamics: Let x_i(t) be the proportion of above-poverty individuals practicing an eco-conscious diet. Behavior evolves via replicator-style imitation dynamics: dx_i/dt = κ_i x_i (1 − x_i) Δe_i(t), where κ_i is the social learning rate. The utility difference favoring eco-conscious over land-intensive is Δe_i(t) = β_i m_i(t) + α_i + γ_i L_G(t). Here m_i(t) is per-capita income; α_i captures net non-income utility differences (social, psychological, cultural, etc.); β_i captures income-related effects; and L_G(t) is global land use, providing a feedback where higher global land pressure increases the utility of eco-conscious diets. Parameters κ_i, α_i, β_i are estimated for 153 countries by fitting model outputs to historical global land use attributable to each country (1961–2013) generated with the earlier land-use accounting model (excluding land equivalent of food waste). Normalization of L_G and m_i is applied for stable inference across countries. Projections to 2100: Population and income follow five SSP scenarios (SSP1–SSP5) using the OECD Env-Growth model from the IIASA SSP database. Agricultural yield futures are parameterized by f ∈ [0,1], a tuning parameter applied to extrapolate the per-capita land-use bounds (sustenance, EAT-Lancet, and maximum) beyond 2013: lower f implies higher future yields (lower land per calorie), higher f implies lower yields. Country-level poverty fractions are extrapolated linearly from historical trends with a floor at zero. For each scenario combination (SSP and f), dynamic simulations compute x_i(t), c_i(t), L_i(t), and aggregate to global L_G(t). Sensitivity analyses vary κ_i, α_i, β_i uniformly above and below baseline fitted values (−100% to +100%) to assess impacts on peak global land use and peak year.

• Two-way feedback drives non-monotonic social response: Across all yield scenarios under SSP2, the proportion of eco-conscious consumers (above-poverty) rises until mid-21st century, then declines as global land use and population decline. Peak eco-conscious share ranges roughly 45–53% depending on yield scenario. The decline occurs when land pressure falls and the utility difference becomes negative for most of the global population. - Peak global land use magnitude: Projected peak global land use for food production ranges between approximately 3.25 and 4.5 billion hectares depending on yield scenario. Including social-land feedbacks can change peak land use by up to about 2 billion hectares under parameter variations. - Yield-social interplay: Higher yields (low f) reduce land pressure but also dampen the social shift towards eco-conscious diets; thus, high-yield futures have lower land use and a lower eco-conscious share than low-yield futures. Under lower yields (f = 0.6–0.8), eco-conscious share peaks earlier than land use; under higher yields (f = 0.2–0.4), it peaks later. - Regional heterogeneity (SSP2): • Africa: Continuous increase in land use through 2100 across yields; eco-conscious share declines or remains flat due to parameter values (often positive β_i and low κ_i), and steady population growth. Land use in Africa increases from about 500 million to ~1 billion hectares above 2013 levels in baseline. • Asia: Land use rises to ~2030 then declines to 2100; eco-conscious share starts higher than Europe but later saturates or falls. • Europe: Land use peaks ~2040 then declines; eco-conscious share increases steadily through 2100, aided by less negative α_i and higher β_i median values, making transitions easier with global land pressure. • Oceania and the Americas: Land use generally declines most of the period; eco-conscious share increases in Oceania and is yield-dependent in the Americas. - Sensitivity and synergies: Under low yields (f = 0.8), peak land use and timing are far more sensitive to social parameters than under high yields (f = 0.2). If incentives (α_i, β_i) favor land-intensive diets (more negative), higher social learning rates can rapidly propagate land-intensive behavior, causing peak land use to jump (e.g., from ~3.5 to ~6 Gha) and occur earlier. Conversely, when incentives favor eco-conscious diets (α_i, β_i above baseline), increased social learning can delay the peak and reduce magnitude in some regions of parameter space. Jointly increasing both income-related and non-income net benefits (β_i and α_i) of eco-conscious diets is more effective at reducing and delaying the peak than changing either alone; if only one can be improved, increasing non-income net benefits (α_i) yields a more consistent delay in peak timing. - Policy-relevant insight: Modest improvements in incentives for eco-conscious adoption around baseline can yield large reductions in peak land use, especially under low-yield futures; combined with adjustments to social learning dynamics, reductions up to ~2 Gha are possible.

The findings demonstrate that explicitly coupling social learning about diets with land-use feedbacks materially alters projections of agricultural land extent. The model explains a mid-century rise and subsequent decline in eco-conscious diets as a consequence of utility shifting with global land pressure: when land use grows, social incentives to adopt eco-conscious diets increase; when land use later declines (aided by demographic transitions and yield improvements), incentives weaken, and eco-conscious adoption recedes. This addresses the research question by showing that endogenizing demand via social dynamics yields qualitatively different trajectories than models with exogenous diets. The sensitivity analyses reveal that socio-economic parameters—particularly net non-income and income-related benefits of eco-conscious diets—can produce giga-hectare differences in peak land use and shift peak timing by decades, especially under low-yield, high-population contexts. These results highlight the importance of policy levers that enhance the perceived utility of eco-conscious diets (e.g., cultural norms, education, pricing, and access) and the nuanced role of social learning speed, which can either accelerate beneficial or detrimental behaviors depending on incentive direction. Regional divergences underscore that global signals (global land use) can drive local behavior in distinct ways due to heterogeneous socio-economic baselines and demographic trajectories.

The study introduces and applies a minimal coupled social-and-land use model to 153 countries, showing that endogenizing dietary social dynamics with feedback from global land use can substantially change projections of agricultural land extent—by up to approximately 2 billion hectares—and produce non-monotonic global trends in eco-conscious diet adoption. It identifies leverage points: increasing both non-income and income-related net benefits of eco-conscious diets and, in some regimes, managing social learning rates can reduce and delay peak global land use. Future work should reduce overfitting risk by integrating richer social data, expand behavioral realism beyond binary diets to continuous spectra and demographic structure, include institutional and cross-country social learning effects, and incorporate economic mechanisms such as price-yield interactions and rebound effects (Jevons Paradox). Incorporating climate risk perceptions and livestock-specific considerations could further refine projections and policy relevance.

• Potential overfitting due to limited social data and multiple inferred parameters, motivating a minimal model. - Simplifying assumptions: binary dietary categories; homogeneous behavior within countries; no explicit modeling of institutions, social norms inertia, or cross-country social learning. - Exclusion of aquatic food sources. - Economic mechanisms such as price responses to yield changes and rebound effects are not explicitly modeled; β_i is not endogenized to yield or prices. - Assumes behavioral response to global (not local) land use; cultural variation may alter this sensitivity. - Land spared by eco-conscious diets may be repurposed for other extractive uses, so reduced agricultural land use does not automatically imply reduced pressure on natural lands.