Uncertainty in land-use adaptation persists despite crop model projections showing lower impacts under high warming

The study investigates how the global land-use system can dynamically adapt to climate change impacts on crop yields and resource availability, and what costs are associated with such adaptation. The context is rising greenhouse gas concentrations leading to unprecedented changes in temperature, precipitation, and extremes, which affect biophysical determinants of crop production. Adaptation can occur at farm scale (e.g., irrigation, cultivars, planting dates, nitrogen inputs) and at the land-use system scale (e.g., cropland expansion or shifts between rainfed/irrigated areas, crop mix changes, R&D investments, and trade adjustments). Prior agro-economic studies focused on specific measures and often excluded CO2 fertilization, yielding pessimistic projections. This study aims to assess combined and individual effects of multiple adaptation strategies at the land-use system level, explicitly including CO2 fertilization and uncertainty across climate and crop models, to quantify adaptation potential and costs under contrasting socio-economic and emissions scenarios (SSP1-RCP2.6 and SSP5-RCP8.5).

The paper situates its contribution within work showing climate change impacts on crop yields and the need for agricultural adaptation, including meta-analyses and recent advances in climate and crop modeling. Previous agroeconomic studies typically analyzed single adaptation levers such as trade liberalization, cropland expansion, or R&D investments, often without accounting for CO2 fertilization, leading to predominantly negative impact projections. Iizumi et al. estimated farm-level adaptation costs including CO2 fertilization but without transformative land-use changes or regional dynamics. Studies ignoring CO2 fertilization tend to overestimate negative impacts under high emissions. The authors highlight gaps: limited integration of multiple adaptation mechanisms, insufficient consideration of regional dynamics, and lack of comprehensive cost assessments at the land-use system level.

The authors use the global land system model MAgPIE (v4.4.0), a recursive dynamic global partial equilibrium model for AFOLU sectors that minimizes total costs given spatially explicit productivity, demand (food, feed, materials, bioenergy), and trade. The model operates on 13 socio-economic regions with inputs clustered from gridded 0.5° biophysical data to 200 clusters via k-means. Demand projections follow SSP narratives with assumptions on population, GDP, diets, food waste, and bioenergy (coupled to REMIND). Livestock demand uses region-specific feed baskets and productivity; trade is based on historical patterns and regional comparative advantage. Climate effects enter via crop yields, water availability/requirements, and soil carbon. Biophysical inputs: Harmonized and calibrated yield projections for four staple crops (maize, soybean, rice, temperate cereals) from nine GGCMs (CYGMA1p74, EPIC-IIASA, LPJmL, CROVER, ISAM, LandscapeDNDC, PEPIC, PDSSAT, PROMET) forced by five CMIP6 GCMs (GFDL-ESM4, MRI-ESM2-0, UKESM1-0-LL, MPI-ESM1-2-HR, IPSL-CM6A-LR), including CO2 fertilization, from ISIMIP3b/GGCMI Phase 3. Additional impacts for other crops, water, and carbon use LPJmL. Historical harmonization and trend extraction use cubic smoothing splines (approx. 30-year running mean) and a limited calibration method to align future trends to an observed-climate baseline (GSWP3-W5E5), avoiding unrealistic amplification when baselines are underestimated. Yields are calibrated to FAO regional levels at the initial time using limited calibration. Scenarios: Two climate-socioeconomic combinations are analyzed: SSP1-RCP2.6 (sustainable, low emissions, strong protections, healthy diets) and SSP5-RCP8.5 (resource-intensive, high emissions, unhealthy diets). To isolate climate-driven adaptation from socio-economic change, SSP1-NoCC and SSP5-NoCC maintain 2015 biophysical conditions. Adaptation mechanisms in MAgPIE: relocation of cropping areas intra/inter-regionally, shifts in crop mix, investments in irrigation, cropland expansion, and technological intensification via a technological change (TC) factor increasing yields proportionally with R&D and management investments. Post-processing computes counterfactual production without adapting cropland patterns and/or TC to quantify supply-demand gaps. Self-sufficiency ratio (SSR) is computed as value-weighted production/demand for traded commodities. Shifts in crop mixes are quantified as the share of cropland allocation distribution differing from NoCC. Costs: Four categories are tracked—land conversion (with calibrated rewards for cropland reduction), intensification (inputs, management, technology via TC; labor per ton fixed; capital dependent on stocks and depreciation; fertilizer/chemicals in other modules), equipping irrigation areas, and trade/transport. Adaptation costs equal absolute differences in aggregated costs between climate-impact runs (SSP1-RCP2.6, SSP5-RCP8.5) and their NoCC baselines, normalized by total crop production to yield average costs in US$05 per ton of dry matter (tDM). Simulations span 1995–2100 with 5–10 year time steps.

• Under RCP8.5, model ensemble shows large uncertainty in yield impacts increasing over time; median global aggregated yield change (maize, soybean, rice, temperate cereals) in 2100 is about -3.8% vs 2015, with maize and soybean most sensitive.
• SSP5-RCP8.5 adaptation responses in 2100 relative to SSP5-NoCC (median [range]): rainfed cropland area +4.2% [-4.5%, +24%]; irrigated cropland area -4.6% [-20%, +33%]; technological change factor (TC) +0.23% [-4.2%, +6.6%]. Lower or negative changes reflect CO2 fertilization and blue-water effects reducing input needs.
• Not adapting in 2100 (using SSP5-NoCC cropland patterns and TC with impacted yields) yields a modest overproduction vs demand: median +1% [−15%, +14%]. In 2050, slight underproduction: median −1% [−11%, +4%]. Rainfed cropland pattern adjustments have the largest effect on supply-demand balance (range about −8% to +8%), followed by TC (−5% to +4%) and irrigated cropland (−4% to +2%).
• SSP1-RCP2.6 shows smaller adaptation needs: slight increase in irrigated area, slight decrease in rainfed area vs SSP1-NoCC; not adapting leads to slightly higher overproduction (median +2% [−2%, +5%]). Positive yield effects in some simulations due to CO2 fertilization and water availability reduce required cropland and intensification.
• Regional dynamics under SSP5-RCP8.5 vary widely across GCM-GGCMs. Example LPJmL–MRI-ESM2-0: global irrigated area −10%, rainfed +0.3%, TC −1.1%; REF sees higher livestock (+26%) and crop (+12%) output and SSR↑ (0.98→1.06). Large crop mix shifts occur in Sahel (up to 85%), Equatorial Africa (64%), parts of Afghanistan/Pakistan (70%), and the Mediterranean (~65%). Under pessimistic CYGMA1p74–UKESM1-0-LL: TC +6%, irrigated +32%, rainfed +21%; USA production −34% despite adaptation; some regions (LAM, SSA) expand cropland to raise output despite yield declines. Under optimistic PROMET–MRI-ESM2-0: global irrigated −19%, rainfed −2.6%, TC −4%, with modest production changes.
• Adaptation costs in 2100: SSP5-RCP8.5 average +4.8 US$05/tDM (range −1.5 to +19 US$05/tDM), translating to −17 to +209 billion US$05 per year depending on impacts and production volume. SSP1-RCP2.6 average ≈ +0.31 US$05/tDM (range −0.62 to +1.4 US$05/tDM). Uncertainty in SSP1 is largest around 2050 (average −4 US$05/tDM; range −10 to +2.5), linked to peak population and demand.
• Cost drivers in SSP5-RCP8.5: intensification costs average +1.3 US$05/tDM (0 to +5.5) and land conversion +2.4 US$05/tDM (−1 to +13). Extremes: PROMET–MRI-ESM2-0 yields −1.3 US$05/tDM (lower conversion, trade/transport, and intensification), CYGMA–UKESM1-0-LL +9.8; highest cost case PDSSAT–IPSL-CM6A-LR +19 due to large land conversion in the USA.
• Key insight: Adaptation costs toward century’s end depend more on regional distribution of impacts, rates of change, and prior land-use adjustments than on global average yield impacts in a given year.

The analysis demonstrates that when CO2 fertilization and improved water availability are considered, land-use adaptation can buffer or even capitalize on climate impacts, moderating the need for cropland expansion and intensification, particularly under low emissions (SSP1-RCP2.6). Under high emissions (SSP5-RCP8.5), despite modest median global yield declines by 2100, large inter-model and regional variability leads to diverse adaptation portfolios—primarily reallocation of rainfed cropland and crop mixes, with supplemental investments in irrigation and technological change. Not adapting can lead to overproduction in some cases by 2100, risking unnecessary ecosystem pressure and price declines, while mid-century underproduction highlights timing sensitivity. The findings address the research question by quantifying how multiple system-level adaptation levers combine to balance supply and demand and at what cost under uncertainty. Importantly, costs are driven by where and how impacts unfold regionally and by the cumulative trajectory of adjustments, implying that robust adaptation planning must be flexible and region-specific to avoid maladaptation. The study highlights the necessity for policies and markets that facilitate mobility of production, technology diffusion, and efficient water and land allocation to manage risk across a wide range of possible futures.

Including CO2 fertilization in impact assessments lowers the median need for cropland expansion and intensification and reduces adaptation costs compared to studies that ignore it. Nevertheless, uncertainty in high-emissions futures remains large, especially at regional scales, complicating effective planning and raising maladaptation risks. Toward the end of the century, adaptation costs hinge more on regional impact distributions, change rates, and prior adjustments than on global average yield changes. A more flexible global food system—on both supply and demand sides—will be required, potentially involving new technologies, more liberalized markets, and deeper economic transformations. Future research should further integrate multiple adaptation levels (farm- to system-scale), expand crop coverage beyond the four staples, and refine representation of management adaptations and policy-driven (planned) adaptation constraints.

• The analysis focuses on autonomous land system-level adaptation, not radical transformations (e.g., degrowth scenarios, large-scale alternative proteins) or planned adaptation policies beyond those embedded in SSP1/SSP5 (bioenergy, afforestation, protection).
• Impact inputs cover only four staple crops (maize, soybean, rice, wheat/temperate cereals); other 15 crop types use LPJmL data, which may be relatively optimistic and vary only across GCMs, potentially understating cross-model variance for non-staples.
• Farm-level management adaptations (e.g., planting dates, nitrogen, cultivar changes) are underrepresented in the GCM–GGCM dataset, possibly underestimating total adaptation potential.
• Extreme weather events, geopolitical instability, and alternative socioeconomic scenarios are not explicitly modeled; effects on prices and trade may be underestimated.
• Potential impacts on labor and animal productivity from heat stress are not included.
• Average values should not be interpreted as the most probable outcomes; results reflect a range of potential futures given large uncertainties in climate and crop model projections and CO2 fertilization responses.