Nearly half of the world is suitable for diversified farming for sustainable intensification

The paper addresses how diversified farming systems (DFS) can contribute to sustainable intensification (SI)—increasing food production per unit area without degrading the environment—amid growing food demand and environmental concerns. It situates the study within debates over intensification versus extensification and the land sharing versus land sparing dichotomy. The authors posit that SI is context-specific and that DFS can support both intensification and extensification depending on local socio-economic conditions. The research question is to identify where, globally, profitable DFS are suitable under current socio-economic conditions and how these areas align with biophysical potentials for cropland intensification and expansion, thereby informing strategies for SI.

The paper reviews evidence that current agricultural practices drive land degradation, biodiversity loss, and greenhouse gas emissions, while rising demand pressures land to intensify or expand. The land sharing versus sparing debate highlights trade-offs: intensification can reduce land conversion but may harm on-farm biodiversity, whereas extensification can improve on-farm biodiversity but may spur expansion and deforestation. Some scholarship suggests an optimal window of SI minimizing trade-offs. Evidence shows DFS can enhance ecosystem services (e.g., pest control, pollination, soil carbon, water quality), reduce groundwater depletion, and sequester carbon, with co-benefits for yield and stability; however, DFS profitability varies and can be lower than simplified systems, though quality premiums can offset this in some markets. Prior suitability studies often focus on biophysical crop requirements; less is known about DFS profitability suitability influenced by socio-economic factors such as market access and infrastructure. Farmers’ decisions frequently follow profitability, underscoring the need to map where DFS are likely to be profitable.

The study predicts global suitability of profitable diversified farming systems (DFS) using a Maximum Entropy (MaxEnt) modeling approach with socio-economic predictors and known locations of profitable DFS. Occurrence data: From a global meta-analysis (Sánchez et al., 2022) comprising 3192 comparisons across 119 studies, the authors classified positive effect sizes as profitable DFS and negative as unprofitable. After removing duplicates, 114 profitable presence and 93 unprofitable records were compiled; however, MaxEnt was run as a presence-only model, treating absence records as not true absences due to context-dependence of profitability. Predictors: Fourteen socio-economic and environmental variables were considered (e.g., cropland area, soil organic carbon, population metrics, electricity coverage, travel time to urban centers, ICT coverage, HDI, GDP per capita, and governance indicators such as voice and accountability, rule of law, political stability, government effectiveness, regulatory quality). Highly collinear variables (Pearson r>0.8) were excluded, resulting in eight uncorrelated variables for modeling at 2.5 arc-minute resolution. Modeling and calibration: Using the kuenm package in R, 155 candidate models were generated by combining 8 variables, 5 regularization multipliers (0.5–4), and 31 feature-class combinations (linear, quadratic, product, threshold, hinge). Model selection criteria included statistical significance via partial ROC (500 iterations, 50% bootstrap), omission rate ≤5%, and ΔAIC ≤2. The initial best model (simple features, lq) achieved AUC=0.878 but had a slightly elevated omission rate; a rerun using the top five predictors (from permutation importance: accessibility, cropland area, voice and accountability, nighttime lights, GDP per capita) with feature classes lqpt and multiple regularization multipliers reduced omission to 0 and satisfied all criteria. Outputs used cloglog transformation (0–1). Binary suitability maps were produced using four thresholds: maximum training sensitivity plus specificity (mtss), balanced training omission (bto), equal training sensitivity and specificity (etss), and 10th percentile training presence (ptp). The authors aggregated thresholded maps to account for uncertainty. Integration with SI pathways: To contextualize DFS suitability for SI, they overlaid the DFS suitability map with global maps of integrated potential for cropland intensification and expansion from Zabel et al. (2019), which combine biophysical potentials and socio-economic conditions for 17 major crops. Bivariate maps were created in R (classInt) to identify areas with concurrent high DFS profitability suitability and either intensification or expansion potential.

• Model performance: The selected MaxEnt model using the five most relevant predictors achieved AUC=0.878 initially; the refined model met all selection criteria with omission rate 0 and feature classes lqpt. - Global suitability extent: Depending on threshold, suitable areas for profitable DFS ranged from 29 to 93 million km². Aggregating thresholds, high-suitability areas accounted for 19.56% of land, while approximately 47% of the world was suitable overall and about 53% not suitable. - Spatial patterns: High suitability is concentrated in North America, Europe, and South/East Asia; in the Global South, high suitability clusters near cities and coastlines. - Variable importance and relationships: Accessibility (distance to cities/markets) and cropland area are the most important drivers. Infrastructure variables (electricity coverage, ICT/cell towers) and existing cropland extent strongly influence profitability suitability. Suitability increases with 30–60% of land in agriculture, then slightly decreases beyond 60%. Nighttime lights above ~60 nW cm−2 sr−1 and higher governance scores (voice and accountability) are associated with higher suitability. GDP per capita shows a negative relationship; above USD 60,000, suitability probability falls below 60%. - SI pathways: Areas with high DFS suitability and high intensification potential include sub-Saharan Africa, Brazil’s east coast, parts of India and Tajikistan, Australia, and Canada. Areas with high DFS suitability and high cropland expansion potential include western Europe, India, China, parts of Brazil, and eastern Europe. Regions suitable for both intensification and expansion while also having high DFS suitability include West Africa near the Atlantic coast and stretches from eastern to southern Africa.

The findings demonstrate that under current socio-economic conditions, the Global North is generally more suitable for profitable DFS due to better infrastructure, market access, and potential price premiums (e.g., certified products). In the Global South, suitability is higher near urban centers and transport corridors, underscoring the importance of accessibility, electricity, ICT, and market development for DFS profitability. Governance (voice and accountability) also correlates positively with suitability, suggesting institutional quality supports diversified, profitable agriculture. Strategically, high DFS suitability in already intensified regions (e.g., much of Europe and North America) could enable extensification through diversification (e.g., agroforestry, hedgerows, mixed plantings) to reduce environmental pressures while maintaining production. In regions with high DFS suitability and high biophysical potential for intensification (e.g., sub-Saharan Africa, parts of South America and Asia), DFS could help close yield gaps on existing cropland, potentially reducing pressure for expansion in biodiverse areas. However, enhancing infrastructure and market access may create feedbacks that increase land-use pressure and opportunity costs for conservation, requiring careful policy design to balance SI goals with biodiversity protection.

The study introduces a global, socio-economics-based suitability assessment for profitable diversified farming systems and connects it to biophysical potentials for cropland intensification and expansion. It shows that nearly half of the world has conditions suitable for profitable DFS, with distinct regional pathways to sustainable intensification: supporting extensification via diversification in highly intensified regions, and enabling intensification on existing agricultural land where biophysical potential is high. The authors conclude that adopting DFS reframes the land sharing versus sparing debate by offering context-specific pathways to achieve SI without a strict either-or framing.

• Predictor scope: The modeling focused on socio-economic variables and did not explicitly include bioclimatic variables, land cover, crop choices/combinations, or adoption rates, due to the diversity of DFS and crop-specific requirements at global scale. - Data integration: Input datasets were from different years and resolutions, introducing scale and harmonization issues. - Uncertainty and scale: Compounded uncertainties arise from predictor data quality, complex system interactions, and the integration with external maps (e.g., Zabel et al.). Scale effects may influence predictions and validity. - Modeling assumptions: Presence-only MaxEnt relies on imperfect occurrence data for profitability; absence records were not treated as true absences. To address uncertainty, multiple thresholds were used instead of a single cutoff, but residual uncertainty remains.