The Global Spatial Co‐Variation Between Crop Diversity and Landscape Heterogeneity

ABSTRACT

The influence of crop and landscape heterogeneity on different components of biodiversity in agricultural landscapes has been assessed at multiple scales. However, how crop species diversity relates to landscape heterogeneity remains underexplored at the global scale. We apply independent global spatial datasets to test the relationship between crop and landscape heterogeneity across 19,505 agricultural landscapes worldwide. We first examine the spatial patterns in crop diversity and landscape diversity (compositional heterogeneity), defined as the effective number of crop species and land cover types, respectively, based on Shannon entropy. Median crop diversity increases on average by 0.36–0.48 effective species for each unit increase in land cover diversity globally. We use quantile generalized additive models (QGAM) to statistically test this relationship. The QGAM approach confirms that crop diversity has a positive but complex relationship with landscape diversity, particularly for landscapes with ≥ 4–5 non‐crop cover types. However, this positive trend is context‐dependent, as the clearest median response generally corresponds to landscapes with moderate cropland extents (25%–75% cropland). We also examine how other components of landscape compositional and configurational heterogeneity, including dominant agricultural field size and patch size, as well as topographic heterogeneity are associated with crop diversity. The relatively highest crop diversity tends to correspond to very small dominant agricultural field sizes (crop configurational heterogeneity) when controlling for cropland extent and non‐agricultural land cover diversity. Mean patch area (landscape configurational heterogeneity) has a strong negative association with crop diversity until patch sizes of > 150 km² while topographic heterogeneity has a consistent positive association with crop diversity. Our findings therefore demonstrate how the diversity of non‐agricultural land covers in conjunction with configurational heterogeneity has relevance to understanding existing patterns of spatial crop diversity. Such insights could help inform efforts to design more multifunctional agricultural landscapes, including landscape‐scale farm diversification strategies.

Are more heterogeneous landscapes also characterized by greater crop species diversity? We examine this question across thousands of agricultural landscapes globally. Crop diversity tends to be higher in landscapes with greater overall land cover complexity, including smaller patch sizes and a greater number of non‐crop cover types, although this relationship is variable and often context dependent. Overall, our findings add further support to considering the role that landscape‐scale factors could play in current patterns of spatial crop diversity and thus broader farm diversification efforts.

1. Introduction

Crop diversity has an important role in the sustainability and resilience of agricultural landscapes. Crop diversity—the number of crop species grown in a specified area—has been linked to ecosystem services such as pest and disease regulation (Redlich et al. 2018; Larsen and Noack 2021), improved water and nutrient cycling (Renwick et al. 2019; Smith et al. 2008), crop yield and food supply stability (Gaudin et al. 2015; Renard and Tilman 2019; Egli et al. 2021), and resilience to climate variability (Langridge et al. 2021; Khoury et al. 2022). Efforts to conserve and enhance crop genetic diversity have become a major component of recent international agreements, including the Aichi Biodiversity Targets adopted under the Convention on Biological Diversity (CBD), the International Treaty on Plant Genetic Resources for Food and Agriculture (ITPGRFA), and the United Nations Sustainable Development Goals (SDGs). For example, SDG Target 2.5 calls to conserve crop genetic diversity as a prerequisite for food and nutritional security (UN 2017). Many studies have addressed crop diversity in relation to agri‐environmental, biodiversity, and food supply variables at more local (e.g., Degani et al. 2019; Khan et al. 2023; Priyadarshana et al. 2024) and national/international scales (e.g., Aguilar et al. 2015; Renard and Tilman 2019; Martin, Cadotte, et al. 2019a; Mariani et al. 2021). Yet spatially explicit analyses of crop diversity at continental to global scales have only emerged recently (e.g., Aramburu Merlos and Hijmans 2022; Machefer et al. 2024).

Recent agricultural intensification has often been accompanied by a reduction in crop diversity as well as a simplification in agricultural landscape structure, such as reducing the diversity of other land‐use types and remnant semi‐natural habitats (Benton et al. 2003; Tscharntke et al. 2005; Reckling et al. 2023). One example is the specialization of agricultural production in the United States, where the diversity of crop types has reduced sharply over time: 88% of counties grew > 10 crops in 1940 whereas only 2% of counties grew > 10 crops by 2017 (Crossley et al. 2021). Additionally, agricultural intensification has coincided with a general shift toward larger fields (White and Roy 2015; Clough et al. 2020) and farm sizes in some regions over the past 50 years (e.g., in the United States, MacDonald et al. 2013). As the proportion of cultivated area in a landscape increases, the likelihood of contiguous crop fields also increases, reducing the availability of diverse non‐crop habitats (Landis 2017). Such reduction can also impact species vital for crop production, including pollinators and natural predators of crop pests (Martin et al. 2016; Sirami et al. 2019; Sánchez et al. 2022). These changes may be related to more mechanized farming practices, land‐use planning decisions, farm consolidation, and market demands (Lowder et al. 2016; Mehrabi 2023).

Transitioning to more sustainable landscapes could therefore entail some movement toward more diversified agricultural systems and away from highly specialized agricultural landscapes, where often fewer large farms focus on producing a small number of crops (Lin et al. 2011). One approach is through crop diversification strategies, such as intercropping and crop rotations, which can be beneficial for biodiversity and soil quality (Beillouin et al. 2019). In turn, many of the ecosystem services essential to agriculture rely on ecological processes at the landscape scale. For example, the provision of pollination services increases with greater landscape heterogeneity, particularly when natural areas within the landscape are restored (Duarte et al. 2018). Research is therefore increasingly focused on how agricultural landscapes can be multifunctional—effectively balancing food production with biodiversity conservation and the enhancement of human well‐being, including livelihood potential (Kremen and Merenlender 2018; Frei et al. 2020).

The diversity and complexity of agricultural landscapes have been shown to enhance biodiversity and ecosystem services, which can in turn benefit agriculture (Estrada‐Carmona et al. 2022). Increasing uniformity and homogeneity in agricultural landscapes may be associated with the loss of biodiversity and a reduction in semi‐natural habitats (Khan et al. 2023). Such changes can alter ecosystem services including: (1) pollinator diversity (Nicholson et al. 2017; Sánchez et al. 2022); (2) natural pest control (Rusch et al. 2016); (3) crop yields (Grab et al. 2018); (4) functional diversity of taxonomic groups (Emmerson et al. 2016; Martin, Dainese, et al. 2019b); and (5) plant assemblages, including lower species richness in simplified landscapes (Lecoq et al. 2021). More complex landscape mosaics can therefore increase habitats available for various species and may decrease reliance on external agricultural inputs (Kremen and Merenlender 2018; Garibaldi et al. 2021; Rasmussen et al. 2024). Higher crop yields associated with landscape complexity (e.g., landscapes with a higher number of land cover categories) could also affect decisions about which crops to grow (Nelson and Burchfield 2021). Likewise, agricultural landscape structure (such as the length of hedgerows and wooded areas) has been hypothesized to influence farmer decision‐making about crop choice and cropping patterns (Cooke et al. 2013).

Landscape heterogeneity is a fundamental concept in landscape ecology that has been commonly used to assess the relationship between crop diversity and biodiversity (Priyadarshana et al. 2024). It encompasses both the variety and proportions of different land cover types (compositional heterogeneity) and their spatial arrangement (configurational heterogeneity) (Fahrig and Nuttle 2005; Fahrig et al. 2011; Jeanneret et al. 2021). Fostering more heterogeneous agricultural landscapes could help to mitigate biodiversity loss and has been suggested as a strategy to increase biodiversity without reducing the amount of land used for agricultural production (Fahrig et al. 2015; Sirami et al. 2019; Jeanneret et al. 2021; Estrada‐Carmona et al. 2022).

Increasing attention is being given to landscape drivers of crop diversity, including how patterns of crop diversity may be shaped by social‐ecological factors and agricultural geographies (Goslee 2020). Past work has examined the socioeconomic and biophysical predictors of crop diversity (Spangler et al. 2022) as well as the influence of landscape diversity on crop yields in the United States (Nelson and Burchfield 2021). Landscape diversity is a component of landscape heterogeneity, measured with metrics such as Shannon diversity index to account for the number and distribution of different land cover types (Nagendra 2002). While it is understood that more complex agricultural landscapes tend to host greater species diversity in general (e.g., Estrada‐Carmona et al. 2022; Priyadarshana et al. 2024), evidence suggests that this could apply to crop diversity as well (e.g., Nelson and Burchfield 2021, 2023).

The specific features within landscapes that may facilitate higher levels of crop diversity, and to what degree, are yet to be fully explored at the global scale. Global studies can provide important insights as there are pronounced differences in the distribution of crop diversity across regions (Martin, Cadotte, et al. 2019a). In their global analysis, Aramburu Merlos and Hijmans (2022) found large ‘gaps’ in crop diversity (defined as the difference between current crop diversity and ‘attainable’ crop diversity), especially in the Western Hemisphere. They found that 84% of croplands globally have attainable crop diversity ‘gaps’ exceeding 50%, meaning that crop species could be doubled to maximize diversity while factoring in both crop suitability and current crop demand (Aramburu Merlos and Hijmans 2022). Better understanding the factors influencing patterns of crop diversity across agricultural landscapes worldwide is therefore an important step.

Here, we apply independent, globally consistent data to examine how crop diversity (a component of landscape compositional heterogeneity) varies in relation to the overall configurational and compositional heterogeneity of agricultural landscapes. We hypothesized that crop species diversity would be positively associated with landscape compositional and configurational heterogeneity. For example, the inclusion of semi‐natural habitats such as hedgerows and riparian corridors could support more diversified farming systems (Kremen et al. 2012) and greater configurational heterogeneity through smaller crop field sizes could be expected to increase the number of crop cover types (Fahrig et al. 2011; Clough et al. 2020). Our intent is to examine the global spatial co‐variation of crop and landscape diversity, not to identify potential causal drivers underlying it or to statistically predict crop species diversity. We therefore first examine the global spatial patterns of crop and landscape diversity, including by using quantile generalized additive models (QGAMs) to statistically test the crop‐landscape diversity relationship. The results suggest that crop diversity is positively associated with both overall landscape diversity and non‐agricultural land cover diversity, but this relationship is complex and varies across landscapes with different cropland extents. We also consider how crop and landscape diversity compare to other components of landscape heterogeneity (mean patch size and agricultural field size) as well as underlying topographic heterogeneity (terrain ruggedness). Overall, our study illustrates how the compositional heterogeneity of crops compares to land cover complexity across thousands of agricultural landscapes globally, capturing a wide range of geographic contexts. This provides further support for considering the role of landscape‐scale factors, including landscape composition and complexity, in farm diversification efforts (Rasmussen et al. 2024). Our findings could thus help direct future research into the drivers of this co‐variation with respect to spatial crop diversity.

2. Methods

Our analysis considers two aspects of landscape heterogeneity that we hypothesized could be relevant to crop diversity outcomes: compositional heterogeneity and configurational heterogeneity (Table 1). Figure S1 illustrates our analytical framework with exemplary landscapes from our study. First, we focus on landscape composition to analyze the relationship between crop diversity and landscape diversity based on spatial patterns in a bivariate map and then statistically test it. Second, we assess crop diversity in relation to other dimensions of compositional and configurational heterogeneity, including dominant field size and patch size, as well as topographic heterogeneity. Combining these approaches helps to assess how crop diversity varies across a gradient of landscape heterogeneity. Input data are temporally variable due in part to the limited availability of global datasets, being representative of approximately the 2000–2010 period for crop diversity and 2010–2017 for other landscape variables.

TABLE 1.

Variables used in the analysis, including crop diversity and landscape heterogeneity.

Variable	Description	Base resolution	Year	Aggregation method	Source
Crop diversity	Effective number of crop species	~10 km	~2010	Mean of effective crop species diversity in hexagon (~14 grid cells per hexagon)	Aramburu Merlos and Hijmans (2022)
Crop diversity gap	Relative difference between current and ‘attainable’ diversity (the effective number of crop species when the proportion of land planted to the best‐adapted crops is largest, holding market demand constant)	~10 km	—	Mean crop diversity gap (%) in hexagon (~14 grid cells per hexagon)	Aramburu Merlos and Hijmans (2022)
Indicators of landscape compositional heterogeneity
Landscape diversity	Effective number of land cover classes, computed with and without cropland	100 m	2015	Computed in this study as the effective number of land cover classes within each hexagon	Calculated from Buchhorn et al. (2020)
Percent cropland and Percent semi‐natural land cover	% cropland and % of semi‐natural land cover classes	100 m	2015	Computed in this study as the percentage of total land area in each hexagon	Calculated from Buchhorn et al. (2020)
Indicators of landscape configurational heterogeneity
Dominant agricultural field Size	Dominant field size class: Very small (< 0.64 ha), small (0.64–2.56 ha), medium (2.56–16 ha), large (16–100 ha), very large (> 100 ha)	1 km	2015–2017	Majority value of the dominant agricultural field size class per hexagon	Lesiv et al. (2019)
Mean patch area	Size in km2 of contiguous land cover patches	100 m	2015	Computed in this study as the mean area of land cover classes within each hexagon	Calculated from Buchhorn et al. (2020)
Topographic heterogeneity
Terrain ruggedness	Index from 0% to 100%, capturing absolute differences in elevation (flat = 0%)	1 km	2010	Mean terrain ruggedness index per hexagon	Amatulli et al. (2018)

Note: There are two compositional heterogeneity variables and two configurational heterogeneity variables in addition to our main variables of crop diversity, diversity gap, and cropland extent. We also consider underlying biophysical heterogeneity due to topography.

2.1. Crop Diversity Data and Crop Diversity Gaps

Crop species diversity was obtained from Aramburu Merlos and Hijmans (2022), who used a combination of two widely used gridded cropland area datasets to calculate the effective number of crop species based on Shannon entropy: ‘SPAM’ (International Food Policy Research Institute (IFPRI) 2019) and ‘Monfreda’ (Monfreda et al. 2008). Shannon entropy (herein, ‘Shannon diversity index’) is calculated as the weighted average of the logarithm of each crop's proportional abundance (made positive with a negative sign); taking the exponential of this value gives the effective number of species. Shannon diversity is commonly applied in landscape pattern analysis to capture the ‘evenness’ and number of different types, or the relative abundance of each crop species in relation to others (Turner et al. 2001; Ramezani and Holm 2011).

SPAM and Monfreda are gridded cropland datasets with crop‐specific areas at the 5 arc‐minute spatial resolution (~10 × 10 km at the equator; Monfreda et al. 2008 and International Food Policy Research Institute (IFPRI) 2019). Both datasets were produced by combining different data sources, including national and subnational agricultural statistics, remote sensing, and crop suitability based on local climate and soil conditions. SPAM v.2.0 contains the physical areas for 42 crop categories made up of 33 individual crops and nine crop groups for the year 2010. Aramburu Merlos and Hijmans (2022) disaggregated the crop categories from SPAM v.2.0 using the more detailed crop‐specific harvested areas from Monfreda for the circa‐year 2000 period. The result is a combined dataset estimating the physical areas for 171 crops used to compute crop diversity (Aramburu Merlos and Hijmans 2022).

We also summarize the crop diversity gaps from Aramburu Merlos and Hijmans (2022), which represent theoretical levels to which crop diversity could be raised. We use the crop diversity gaps based on ‘attainable’ diversity (mean of the two methods), which is determined by the environmental suitability of crops and further constrained to match current societal demand. We reclassed some values to ensure that all ranged from 0% to 100%, with higher values indicating a larger gap. This provides a simple benchmark for the degree to which current crop diversity could hypothetically be enhanced and is thus informative for understanding the relative degree to which strategies addressing agricultural diversification could raise crop diversity in different regions.

2.2. Global Land Cover Data, Including Cropland Extent

We use an independent global land cover dataset to enable a consistent assessment of landscape composition and configuration for our analysis at relatively high spatial resolution from Copernicus' CGLS‐LC100 (Buchhorn et al. 2020). CGLS‐LC100 includes 22 land cover classes from 2015 to 2019 with around 80% or greater overall accuracy (Buchhorn et al. 2020); we selected the year 2015 for closer comparison to the crop diversity data. One of the land cover classes represents agriculture (“Cultivated and managed vegetation/agriculture (cropland)”) but does not differentiate crop types. CGLS‐LC100 provides a good indicator for cropland area compared to recent global maps of cropland extent; however, it is known to overestimate cropland areas in some places, such as due to misclassification of pastures (Potapov et al. 2022). To maintain a greater consistency across our global landscape‐level analysis and comparability with our focus on cropping systems, we therefore omitted landscapes based on two conditions: (1) landscapes with < 5% cropland based on CGLS‐LC100 that may have very limited cropland extents, and (2) landscapes with ≥ 75% permanent pasture/rangeland based on the Historic Land Dynamics Assessment+ (HILDA+; Winkler et al. 2021). While pasture and rangelands add heterogeneity to agricultural landscapes, grazing lands have distinct agroecological characteristics that are different from cultivated systems producing crops (Ramankutty et al. 2008). Cropland extents refer to the land area, excluding water. All CGLS‐LC100 variables (Table 1) were computed using the base spatial resolution (100 m) before summarizing to the landscape level.

2.3. Landscape Diversity and Heterogeneity Variables

Land cover diversity was computed in two ways by using CGLS‐LC100 as the effective number of land cover classes in each hexagon based on the Shannon diversity index (accounting for both the number of unique land cover types and their proportional abundance in a landscape). For Shannon diversity, we follow the general approach of Martínez‐Núñez et al. (2023), except that we removed the ‘Open Sea’ class given our emphasis on terrestrial agriculture. For consistency with the crop diversity calculation from Aramburu Merlos and Hijmans (2022), we then converted this to the effective number of land covers by taking the exponential of Shannon diversity. Given the clear conceptual overlap between crop diversity and cropland as a land cover class in CGLS‐LC100, with potential confounding effects (such as endogeneity), we use two measures of land cover diversity to help add robustness to our analysis:

1. Overall landscape diversity, which includes all available land cover classes.

2. Non‐agricultural (non‐crop cover) landscape diversity, which omits cropland.

We further consider topographic heterogeneity and two other common landscape metrics to quantify the composition and configuration of land cover within each landscape: percent semi‐natural land cover and mean patch area. The percentage of semi‐natural areas represents the proportions of “any habitat within or outside of the crop containing a community of non‐crop plant species”, following Holland et al. (2017). From CGLS‐LC100, this includes land covers that are not extensively modified by human activities such as forests, shrubland, herbaceous vegetation, herbaceous wetland, moss and lichen, bare and sparse vegetation. We calculated mean patch area to assess landscape configuration. This metric is commonly applied to quantify the average size of individual, homogenous land cover patches in a landscape (Kupfer 2012). We vectorized CGLS‐LC100 to compute the area of homogenous land cover patches then calculated the mean area per hexagonal landscape, excluding water; for this, we used ‘Queen's contiguity’ to form patches with diagonal grid adjacency and allow land cover patches to cross multiple hexagons (Wolff et al. 2021). As an independent measure of underlying biophysical heterogeneity due to topography, we use the terrain ruggedness index derived from global digital elevation model products by Amatulli et al. (2018) (Table 1).

We use dominant agricultural field size as an independent landscape crop configuration variable to compare with crop diversity (compositional heterogeneity). Field size is a common metric to assess configurational crop heterogeneity, which is itself a component of broader landscape heterogeneity (e.g., Collins and Fahrig 2017; Hass et al. 2018). To our knowledge, no seamless global field boundary datasets were available during the timeframe of our study. We therefore used the crowdsourced agricultural field size classification from Lesiv et al. (2019), which includes five dominant field size categories globally ranging from very small fields (area of < 0.64 ha) to very large fields (> 100 ha) (Table 1). Different field sizes contribute to the spatial arrangement of croplands, so we hypothesized that these would be analogous to our calculation of patch area but more explicitly focused on croplands.

2.4. Hexagonal Agricultural Landscapes: Visual Analysis and Descriptive Statistics

Hexagons are commonly used in global land system classifications for their beneficial geographic properties (Ellis et al. 2021), such as minimizing distortion in grid cell area across different latitudes. We use the open‐source H3 Hexagonal Hierarchical Geospatial Indexing System as a consistent spatial unit for our analysis (available at: https://h3geo.org/). This system provides a robust geospatial framework for delineating the Earth's surface into hexagonal cells for analysis at nested spatial extents (Bousquin 2021). We chose a moderate landscape extent with an average area of ~1770 km² (H3‐4), as it allows for aggregating several patches and grid cells in our composition and configuration variables (Table 1) and thus capturing a range of features within each hexagon.

Spatial datasets were aggregated to the hexagon level using ArcGIS Pro v.3.1 (ESRI, Redlands); all other calculations and visualizations were conducted in the R statistical programming software v.4.5.1 (R Core Team 2025), mainly using the ‘data.table’ package. Patch area was computed in ArcGIS Pro then summarized as a mean in R. We summarized the data in each hexagon according to the aggregation method listed in Table 1. Hexagons with no data for any variable were omitted.

For our visual and descriptive statistical analysis, we use three approaches. First, we assess Spearman's rho (non‐parametric, rank‐based) correlations across all numeric variables, including assessing the degree to which non‐independent variables derived from CGLS‐LC100 are correlated—as we expected these might capture similar landscape characteristics. Second, bivariate maps were generated based on the quartile distributions of both crop and landscape diversity to assess spatial patterns. We also used cross‐tabulations to assess the changes in quartile rankings for overall landscape diversity and non‐agricultural land cover diversity (e.g., a landscape might be in the top quartile for overall landscape diversity but the bottom quartile for non‐agricultural land cover diversity). Third, we used box plots to examine the distribution of crop diversity and crop diversity gaps across a gradient of effective land cover diversity and dominant field sizes, then map their geographic patterns globally (described further in Section S1).

2.5. Statical Analysis With Quantile Generalized Additive Models (QGAM)

We apply non‐parametric quantile generalized additive models (QGAMs) (Fasiolo et al. 2021) as crop and landscape diversity showed non‐linearity and were heavily skewed with outliers. GAMs are analogous to regression but can flexibly assess complex, nonlinear relationships that are common in agriculture and agrobiodiversity data by using spline smoothers (e.g., Kleijn et al. 2009; Wellington et al. 2023). A key advantage of non‐parametric QGAMs is that they do not require assumptions about the underlying distribution of the response variable (De Rosa et al. 2024). QGAMs allow for analysis of partial effects (effect of each predictor on the response variable holding other predictors fixed) at different quantiles of the response variable distribution (Olivetti et al. 2024). This provides a different entry point for the analysis of the relative relationship between the independent and response variables, especially when extreme values are present, as the fitting process of QGAMs helps diminish the influence of outliers (De Rosa et al. 2024). We fit the QGAMs using the ‘qgam’ package and plot them using the ‘mgcViz’ package in R (Fasiolo et al. 2020). We focus on the results corresponding to the median (qu = 0.50) quantile of crop diversity (response variable), but assessed three other quantiles (25th, 75th, and 90th percentiles) to compare how the effects of landscape diversity differ across different parts of the distribution of crop diversity.

To help account for possible spatial dependence, we added a smoother term for the interaction between latitude and longitude (x and y, decimal degrees) following Martínez‐Núñez et al.'s (2023) assessment of landscape diversity and bird diversity. We use AIC (Akaike Information Criterion) as a simple measure to compare models with and without spatial effects to assess tradeoffs between model fit and model complexity (Vrieze 2012).

Our goal was not to predict crop diversity or to assess causality but rather to assess the potential shape and direction of the crop‐landscape diversity relationship. We therefore visually inspect partial dependence plots that show the partial effect of landscape diversity on crop diversity, while holding geography constant based on the spatial variable (sensu Hastie et al. 2009; De Rosa et al. 2024; Olivetti et al. 2024). We tested several QGAM models based on our objective including the two landscape diversity computations as well as the potential effects of other composition and configuration variables, including cropland extent. Our model with multiple variables from Table 1 aimed to reduce AIC as well as potential multicollinearity among predictors assessed using the degree of correlation with Spearman's rho. Model testing showed that the QGAM models with the spatial smoother (interaction between latitude and longitude) have lower AIC, so we focus on those in the Results. Our final models test the following relationships with crop diversity:

• Global models with nonlinear smoothers for overall landscape diversity and non‐agricultural land cover diversity, with a spatial smoother.

• The same models, above, for each of three cropland extent subsets to capture broad differences in the area of landscapes devoted to cropland (< 25% cropland, 25%–75% cropland, and > 75% cropland landscapes).

• Global additive effects of non‐agricultural land cover diversity and cropland extent, the nonlinear interaction of these variables, as well as the linear effect of dominant field size and the spatial smoother.

• Global model with a nonlinear smoother for log‐transformed mean patch area and the spatial smoother.

• Global model with a nonlinear smoother for terrain ruggedness and the spatial smoother.

3. Results

Our analysis encompasses 19,505 hexagonal landscapes globally. Most compositional and configurational heterogeneity variables have only weak correlations with crop diversity (Spearman's rho < 0.2 and > −0.2). However, many of these variables have moderate‐to‐strong correlations with each other and thus may capture similar landscape contextual features (Figure 1). Cropland extent is almost perfectly negatively correlated with semi‐natural vegetation cover (Spearman's rho = −0.97, p < 0.001) and has a moderate correlation with overall landscape diversity (effective number of land covers, including cropland) (Spearman's rho = −0.57, p < 0.001)—yet both variables are only weakly correlated with non‐agricultural land cover diversity (effective number of land covers, excluding cropland). Terrain ruggedness has a moderate positive correlation with both crop diversity and overall landscape diversity (Spearman's rho of 0.27–0.28, p < 0.001).

FIGURE 1.

Correlation matrix showing Spearman rank‐based (rho, ρ) correlations for the numeric variables in this study. % Cropland refers to the cropland extent based on the CGLS‐LC100 cropland class.

The relationship between crop and landscape diversity at the global scale is complex (see scatterplot in Figure S2) with a relatively weak correlation (Spearman's rho of 0.16–0.19, p < 0.001). However, our descriptive analysis (detailed in Section S1) shows a general trend of increasing median crop diversity and decreasing crop diversity gaps with each per‐unit increase in effective land cover diversity, which differs slightly when considering the effective number of non‐agricultural land covers only. On average, the median crop diversity increases by 0.48 and 0.36 effective crop species for each unit increase in overall land cover diversity and non‐crop cover diversity, respectively (Figure S3a). More diverse landscapes also tend to have slightly lower median crop diversity gaps, with a reduction in the median crop diversity gap of 2.0%–2.8% for each unit increase in effective land cover diversity (Figure S3b).

3.1. Global Spatial Patterns of Crop and Land Cover Diversity

Crop diversity and overall landscape diversity have strong spatial patterns within and across regions. The bivariate map in Figure 2a shows the spatial patterns of crop diversity and overall landscape diversity by quartiles. Landscapes with the relatively highest crop diversity (≥ 9.2 effective crop species) and overall landscape diversity (≥ 5.2 effective land covers, including cropland, shown as the darkest green color in the bivariate legend corresponding to the top quartile in Figure 2a) occur on every continent, but are most prevalent in Europe and Asia (> 80% of these most diverse landscapes globally occur on these two continents). Europe has a particularly high co‐occurrence of crop and overall landscape diversity: 18.9% of the landscapes in Europe fall in the top quartiles for both variables, with a mean overall landscape diversity of 4.4 effective land covers and a mean crop diversity of 8.7 effective species. By comparison, 9.3% of landscapes in Asia fall in the top quartile for both variables, though this continent has a slightly lower mean crop diversity (8.2 effective species) and mean overall landscape diversity (3.3 effective land covers, including cropland). Relatively high crop diversity yet low landscape diversity subregions are prevalent in Asia, especially Western Asia (e.g., Iran and Saudi Arabia). The African continent also tends to be comprised of similarly high overall diversity landscapes, but again slightly lower than Europe (mean crop diversity of 7.5 effective species and mean landscape diversity of 3.6 effective land covers, respectively).

FIGURE 2.

The global distribution of the crop and landscape diversity variables. Panel (a) shows a bivariate map of the spatial patterns between crop diversity (effective number of crop species) and overall landscape diversity (effective number of land covers, including cropland). The bin breaks in the legend are based on the quartiles for each variable (n = 19,505). Panel (b) depicts how non‐agricultural land cover diversity compares to overall landscape diversity based on a cross‐tabulation of the quartile rankings for each, including little difference (same quartile for both), relatively lower (non‐cropland 1 quartile decrease), relatively higher (1 quartile increase), and relatively much higher (≥ 2 quartile increase) non‐crop land cover diversity. See Figure S4 for the bivariate map of non‐cropland diversity with crop diversity. Map lines delineate study areas and do not necessarily depict accepted national boundaries.

While such altogether diverse landscapes (top quartile for both variables in Figure 2a) can be found in parts of the Western Hemisphere (e.g., the southeast United States, southern Brazil, and Cuba), North and South America tend to be characterized by much lower absolute crop diversity (as reported by Aramburu Merlos and Hijmans 2022). Still, overall landscape diversity is quite spatially variable. Close to half (47.1%) of the landscapes corresponding to the lowest quartiles of both crop and overall landscape diversity globally (grey color in Figure 2a) occur in North and South America. The midwestern United States, central Argentina, and parts of northeast Brazil are examples of subregions with low crop and overall landscape diversity. Around three‐fifths (60.7%) of the global distribution of landscapes with high overall landscape diversity but low crop diversity (≥ 5.2 effective land covers and ≤ 4.3 effective crop species, respectively) is in the Western Hemisphere (dark blue color in Figure 2a). However, smaller subregions with overall low landscape diversity occur on every continent, including parts of Europe and Asia.

Figure 2b shows how non‐agricultural land cover diversity (excluding cropland cover) compares to overall landscape diversity (with the cropland class from CGLS‐LC100). About 43.5% of landscapes (n = 8491) remain in the same relative quartile grouping when calculated for non‐agricultural land cover diversity compared to overall landscape diversity (shown as the grey color in Figure 2b). Non‐agricultural land cover diversity diverges most sharply (≥ 2 quartile increase) from overall landscape diversity in areas with the highest cropland extents (e.g., parts of the midwestern United States, eastern Europe, and northern India). For example, the 10.1% of landscape areas where non‐agricultural land cover diversity is relatively much higher than overall landscape diversity (Figure 2b; n = 1974) have on average > 75% of their land area devoted to cropland. By comparison, areas where non‐agricultural diversity is relatively lower (1 quartile decrease) than overall landscape diversity (n = 6825, 35.0% of landscapes) have on average < 25% of their land area devoted to cropland.

3.2. Global Statistical Relationship Between Crop and Landscape Diversity

We focus on the QGAM models with the spatial smoother term (see Table S2 for details on the global QGAM model parameters, including for the non‐spatial global models). The QGAM analysis confirms a positive crop–landscape diversity relationship globally, although the spatial term explains most of the variation in crop diversity based on the r ² (coefficient of determination). The partial dependence plots in Figure 3 show how crop diversity changes in relation to increasing effective land cover diversity for the median (0.50) quantile across all landscapes globally, controlling for latitude and longitude. The global relationship is positive and slightly nonlinear (effective degrees of freedom of ≥ 3.0 indicating a corresponding number of bends) when considering both overall landscape diversity (Figure 3a) and non‐agricultural land cover diversity (Figure 3b). The estimated partial effect shows a more pronounced positive association between overall landscape diversity and crop diversity at the median level (p < 0.001), typically exceeding 1.0 effective crop species as overall landscape diversity increases above ~5.0 effective land covers. While non‐agricultural land cover diversity also has an upward trend in the estimated smoother (p < 0.001), the partial effect is smaller with a more pronounced inflexion and wider confidence intervals as effective land cover values exceed ~5.0 (Figure 3b). The shape and direction of the smoothers are generally consistent for the other quantiles of crop diversity, with slight variation especially for the lowest (25th percentile) and highest (90th percentile) quantiles (Figure S5).

FIGURE 3.

Partial dependence plots from the quantile generalized additive models (QGAM) showing the relationship between crop and landscape diversity across all landscapes globally. Panel (a) shows the model with overall landscape diversity, which includes cropland area as a land cover type; panel (b) shows the results for non‐agricultural land cover diversity, which omits the cropland land cover class. The y‐axis shows the partial effect for the median quantile (qu = 0.5), that is, how crop diversity changes along the distribution of effective land cover diversity, holding the spatial effect constant. Shaded areas depict the 95% confidence interval. ‘edf’ refers to effective degrees of freedom and describes the bendiness of the spline smoother (1.0 is perfectly linear). Figure S5 shows detailed results for the other three quantiles. QGAMs with a spatial smoother predictor variable are shown here (see Table S2 for non‐spatial model results).

The QGAM results for the broad cropland extent subsets (maps are shown in Figure S6) depict how the relationship between land cover diversity and crop diversity varies across agricultural landscape contexts. Moderate cropland landscapes (25%–75% cropland area out of the total landscape area; n = 8995) (Figure 4b,e) show the clearest positive relationship of both overall and non‐agricultural landscape diversity with crop diversity (p < 0.001) when controlling for geographic location. Conversely, landscapes with the relatively lowest (< 25% cropland, n = 7381; Figure 4a,d) and highest (> 75% cropland, n = 3129; Figure 4c,f) cropland extents often show more nonlinear functional forms (edf ≈4); Figure 4f shows a sharp but significant inverse relationship (p < 0.001), which may be attributable to cropland‐dominated landscapes that have relatively high non‐crop cover diversity but low crop diversity, such as the Midwest United States (Figure 2b). The partial dependence plots for the other quantiles (25th, 75th, and 90th percentiles; Figure S7) generally show the same shape but some differences in the size of the estimated partial effects, suggesting heterogeneity in how landscapes with the relatively lowest or relatively highest crop diversity relate to a gradient of increasing land cover diversity across different cropland extents.

FIGURE 4.

Partial dependence plots from the quantile generalized additive models showing the relationship between crop and landscape diversity across broad subsets of cropland extent in the landscape (< 25%, 25%–75%, and > 75% cropland area). Panels in the left column (a, b, c) show the models with overall landscape diversity, including cropland; panels in the right column (d, e, f) show the results for non‐agricultural land cover diversity, omitting the cropland cover class. Shaded areas depict the 95% confidence interval. See Figure S7 for detailed results of the models including the other quantiles.

3.3. Cropland Extent and Field Size Moderate the Crop‐Landscape Diversity Relationship

Median crop diversity increases incrementally across dominant field size classes from largest to smallest globally (Figure S8)—from 3.7 effective crop species in landscapes characterized by very large fields to 8.2 effective crop species in those with very small fields. Crop diversity and field size also have distinct spatial patterns with non‐agricultural land cover within and among continents (see Section S1.2 and Figure S9). Our follow‐up analysis therefore examines the joint association of non‐agricultural land cover diversity and cropland extent with crop diversity while accounting for field size and geographic location. Figure 5 shows results of this multivariable model, with the additive effects of non‐agricultural land cover diversity and cropland extent—plus their interaction—as well as the parametric effect of dominant field size on the median crop diversity response. We find a significant nonlinear interaction (p < 0.001) between the diversity of non‐agricultural land covers and cropland extent as well as a clear distinction in crop diversity among the largest and smallest dominant field size classes (Figure 5).

FIGURE 5.

Combined effects of non‐agricultural land cover diversity, cropland extent, and field size on the median crop diversity response variable. Results are shown for the quantile generalized additive model for the median quantile with partial dependence plots for the additive (smooth) effects of non‐agricultural land cover diversity (a), cropland extent (b), and their nonlinear interaction (c). Panel (c) shows the nonlinear interaction as a two‐dimensional smoothed surface with contours corresponding to the estimated bivariate partial effect at the median quantile of crop diversity. Panel (d) shows the linear parametric effect for the categorical dominant field size variable. This model also included the spatial smoother.

The combined effect of non‐agricultural land cover diversity and cropland extent on the crop diversity response (Figure 5c) displays two main intervals of positive associations: (1) contours of high cropland extents (> 75% cropland) with relatively low non‐agricultural land cover diversity (from 2.5 to 5.0 effective land covers), which are characteristic of parts of South Asia, and (2) low cropland extents (< 25% cropland) with high land cover diversity (> 7.5 effective land covers), which occur in several regions (see Figure S6). After factoring in the combined effect of these variables, only the smoother for cropland extent is significant (p < 0.001), with a clear negative relationship with crop diversity at the median quantile. Dominant field size also has a clear structure, with a strong positive association of very small fields with relatively higher crop diversity and the lowest median response on crop diversity for very large fields (Figure 5d).

3.4. Association of Crop Diversity With Configurational and Topographic Heterogeneity

The spatial patterns of mean patch area are displayed in Figure 6a. Mean patch area has a strong negative association with crop diversity at the median level (p < 0.001) until reaching patch areas of ~150 km², after which the effect dissipates and reverses direction (Figure 6b). This nonlinear inflexion corresponds to areas with some of the largest patch sizes globally (orange and pink colors in Figure 6a) yet relatively high crop diversity at or above the global median—the majority of which are in Asia (65% of n = 2258 landscapes with > 150 km² mean patch area). These landscapes tend to have mixed dominant field sizes, with both very small and small fields (n = 1328) as well as large and very large fields (n = 912) represented, indicating how mean patch area can poorly reflect dominant field size in some regions (Figure S9).

FIGURE 6.

Map showing the global distribution of mean patch area calculated from CGLS‐LC100 as the average size of individual, contiguous land patches within each hexagonal landscape (a). Mean patch area (km²) is shown on a logarithmic scale given the large variation in patch sizes. Panel (b) shows the partial dependence plot resulting from the quantile generalized additive model (including the spatial smoother) with a smooth term for mean patch area (log‐transformed) at the median quantile of crop diversity; values on the x‐axis are back‐transformed from the log‐scale form for ease of interpretation. Note that in (b), landscapes with very large mean patch areas > 10,000 km² (n = 14) were excluded to improve model stability. Map lines delineate study areas and do not necessarily depict accepted national boundaries.

Mean patch area has a relatively strong negative correlation with overall landscape diversity (Spearman's rho = −0.70, p < 0.001) and a moderate negative correlation with non‐agricultural land cover diversity (Spearman's rho = −0.48, p < 0.001) (Figure 1). Given this negative correlation, smaller mean patch sizes (the lowest quartile is < 0.53 km²; Figure 6a) coincide with relatively diverse landscapes in places such as central and southern Europe, southeast China, and the southeast United States (Figure 2a). Overall, European agricultural landscapes have the smallest mean patch area (16.7 km²) while Africa has the largest patch area (245.9 km²).

Higher topographic heterogeneity (as indicated by greater terrain ruggedness) closely corresponds to mountainous regions in parts of Central America, Southern Europe, Eastern Africa, and South Asia (Figure 7a). These areas tend to also have relatively high crop diversity in a global sense. The QGAM model confirms that topographic heterogeneity has a strong, positive association with crop diversity when controlling for geographic location (Figure 7b).

FIGURE 7.

Map showing global topographic heterogeneity based on the terrain ruggedness index (0%–100%) values from Amatulli et al. (2018) within each hexagonal landscape (a). Flat areas have values closer to 0% whereas high values indicate larger mean absolute differences in local elevation. Panel (b) shows the partial dependence plot resulting from the quantile generalized additive model (including the spatial smoother) with a smooth term for terrain ruggedness at the median quantile of crop diversity. Map lines delineate study areas and do not necessarily depict accepted national boundaries.

4. Discussion

To our knowledge, this study is the first to examine the association specifically between crop diversity and landscape heterogeneity at the global scale. While many studies have considered the positive association of crop diversity as a component of landscape heterogeneity for biodiversity in agricultural landscapes (as shown in the meta‐analysis of 122 studies by Priyadarshana et al. 2024), our findings suggest that the consideration of landscape heterogeneity is relevant to the large‐scale assessment of crop diversity. Specifically, we demonstrate that more diverse landscapes tend to have relatively higher crop diversity, although this association is variable and context dependent: globally, it is clearest for landscapes that have more moderate cropland extents (~25% to 75% cropland area). Cropland extent (inversely correlated with semi‐natural vegetation) also appears to have a moderating role in the association between non‐agricultural land cover diversity and crop diversity, as demonstrated by their nonlinear interaction (Figure 5c). Crop diversity also differs clearly among dominant field sizes, with the relatively greatest median (6.6–8.2 effective species) and absolute crop diversity typically coinciding with inter‐regional variation in the locations of small and very small field sizes (< 2.56 ha; described in Section S1.2). However, while our findings suggest that landscape compositional and configuration variables are relevant to current spatial patterns of crop diversity, geographic location (latitude and longitude) and topographic heterogeneity are likely even more important (Figure 7 and Table S2).

Our study does not assess the potential causal mechanisms that could help explain the linkages between more heterogeneous landscapes and high crop diversity. These are important next steps, especially confirming the direction of the effects we show. Sociopolitical factors ranging from agricultural policies to cultural norms influence regional disparities in crop diversity and the adoption of diversification strategies (Blesh et al. 2023). However, considerable evidence exists from the literature to hypothesize about potential mechanisms underlying how crop diversity may be linked to landscape complexity (e.g., Rasmussen et al. 2024). Diverse landscapes characterized by a variety of semi‐natural elements could provide conditions conducive to supporting a wider range of crops (Fahrig et al. 2011), including different niches and habitats (Krauss et al. 2004; Tscharntke et al. 2012; Duarte et al. 2018). Landscape heterogeneity could also foster pollinators and pest regulation (Shackelford et al. 2013; Botzas‐Coluni et al. 2021) that may be favorable for cultivating a diverse mix of crops. Compositional heterogeneity of agriculture, including via higher crop diversity, could also enhance crop yield outcomes (Burchfield et al. 2019). The heterogeneity and overall complexity of landscapes could thus have a role in enhancing crop productivity and more multifunctional landscapes (Nelson and Burchfield 2021).

4.1. Landscape Features as a Constraint or Facilitator of Crop Diversity

Landscape diversity would not be expected to increase crop diversity without the addition of new crop species by farmers; however, our findings reinforce the role that agricultural landscape structure could have in either constraining or facilitating agricultural diversification strategies. For example, Rasmussen et al. (2024) found that landscape composition, particularly the presence of semi‐natural habitats, moderates the effect of different on‐farm diversification strategies. Taken as a whole, our QGAM analysis shows that moderate cropland extent (25%–75% cropland) landscapes that maintain ≥ 4–5 non‐crop cover types, as well as landscapes characterized by smaller field and land cover patch sizes, often have relatively higher crop diversity. On the other hand, more homogenous landscapes with higher cropland extents, less non‐agricultural land cover diversity, and larger field sizes are often associated with lower crop diversity. Our results therefore support the hypothesis of Clough et al. (2020) that field size can constrain crop diversity by reducing the presence of certain crops on the landscape. Similarly, we find that patch size provides a useful indicator of landscape configurational heterogeneity, as crop diversity generally declines with mean patch area (Figure 6b)—though this association may ultimately be dwarfed by regional differences in field size (Figure S9). The incorporation of diverse field sizes and non‐crop covers could therefore be useful considerations in land‐use planning to support crop diversification.

Our results with the QGAM approach show considerable nonlinearity and variability across quantiles of low to high crop diversity. How landscape heterogeneity may be related to crop diversity across thousands of different landscape contexts globally is therefore understandably complex. When cropland extents are relatively low (< 25% cropland), compositional heterogeneity may be too great to discern a crop‐landscape diversity relationship; for example, we observe that very large patches of semi‐natural vegetation may have a confounding effect that reduces overall landscape diversity (Figure 6). Both crop and landscape diversity are influenced by a myriad of factors including climate and topography, as shown by our analysis of terrain ruggedness (Figure 7). Conversely, when cropland extent is very high (> 75% cropland), overall land cover diversity is always low in our analysis (Figure 2a) yet non‐agricultural land cover diversity can be unexpectedly high (e.g., in the US Midwestern ‘Corn Belt’, Figure 2b), which may explain the observed nonlinear shape of the crop‐landscape response (Figure 4f). Since landscape diversity is moderately to highly correlated with some other landscape compositional and configurational heterogeneity variables, including mean patch area and semi‐natural vegetation, it encompasses aspects of both (Figure 1; Martínez‐Núñez et al. 2023). However, the CGLS‐LC100 land cover dataset does not capture patches < 100 m², which may miss strips of remnant semi‐natural vegetation in agricultural landscapes that are important components of landscape complexity (e.g., hedgerows; Rasmussen et al. 2024); this may be influential on our results for regions such as South Asia, where our approach resulted in very large contiguous patches of some land cover classes. For example, we find a mean patch area of ~162 km² in India compared to ~78 km² in the United States (Figure 6).

4.2. Limitations and Future Directions

Our findings are influenced by several factors, including the choice of input data and limited data to assess crop diversity at higher resolutions globally. The later date of the landscape heterogeneity data (2015–2017) in comparison to the crop diversity data (circa 2010) may impede interpretations related to landscape heterogeneity shaping crop diversity. Global cropland datasets can also vary greatly in areal estimates, mainly due to definitional issues; for example, Tubiello et al. (2023) compared six high‐resolution (10–30 m) global maps of cropland extent, showing that uncertainty was around 25% with global cropland area in 2020 ranging from 1.1 to 1.9 billion hectares. This uncertainty can be due to areas where cropland is mixed with pasture or in places where landscapes are highly fragmented (Lu et al. 2020; Tubiello et al. 2023), which is often the case in our analysis (i.e., landscapes with high effective land cover diversity or small mean patch area). The CGLS‐LC100 data used in our study (100 m) is generally but not always consistent with recent high‐resolution maps developed specifically to assess global cropland extent. For example, CGLS‐LC100 often misclassifies pasture in Brazil and Paraguay as cropland relative to Potapov et al.'s (2022) dataset. We attempted to mitigate this by using the HILDA+ data to mask landscapes with > 75% permanent pasture from our analysis.

Another constraint is our reliance on a single crop diversity metric, which may not capture important crop traits or facets of agrobiodiversity that are important for landscape multifunctionality. While Shannon diversity is a common measure of species richness and evenness, other diversity metrics capture different properties. For example, Simpson's diversity emphasizes dominant species and is thus a good measure of crop concentration (Aguilar et al. 2015). Aramburu Merlos and Hijmans (2022) found that global values for both current and attainable diversity are lower when calculated with Simpson's diversity due to the highly uneven proportions of major crops produced and consumed globally; crop diversity gaps are thus also lower when computed with Simpson's diversity (exceeding 50% across 67% of global croplands compared to 84% for Shannon diversity). Functional diversity has also been assessed in many landscape‐scale studies, accounting for the functional traits of different crop species (Cadotte et al. 2011). Temporal crop diversity (e.g., crop sequences and rotations over multiple growing seasons or years) is another important indicator for crop diversity assessment (Smith et al. 2008; Renwick et al. 2019; Aramburu Merlos and Hijmans 2020). To our knowledge, no global datasets currently exist to analyze temporal crop sequences. Though we were unable to consider temporal changes in either crop or landscape diversity explicitly, our analysis can be interpreted as a “space‐for‐time” approach (sensu Macchi et al. 2020), as the current spatial variation across the world in crop and land cover diversity could provide insight on the different temporal trajectories of agricultural landscapes, including the effects of land use and land cover change on cropland extent and semi‐natural vegetation. Our study is also at a relatively large landscape extent (~1770 km²) with mean crop diversity across multiple grid cells, and crop diversity is known to be scale dependent (Aramburu Merlos and Hijmans 2020).

Our study captures contrasting field sizes across regions but does not include farm size. Farm size is one of the most important variables across studies of the drivers and constraints of on‐farm diversity (Tacconi et al. 2022). Smaller farms often harbor greater crop diversity and have higher yields at the landscape level (Ricciardi et al. 2021), in addition to influencing farm management practices (Liebert et al. 2022). Relatively smaller farms (< 50 ha) contribute 51% to 77% of the different food nutrients from key food crops harvested for human consumption globally (Herrero et al. 2017). Field size and farm size cannot be viewed interchangeably, as large farms can include small fields that may lead to overestimates of smaller farms when field size is used as a proxy for farm size (Su et al. 2022). Other important farm management factors that might affect crop diversity include the area equipped for irrigation (which may support a more diverse mix of crops) (Waha et al. 2020), socioeconomic factors such as land tenure, subsistence versus market farming, and sociocultural considerations in crop choice (Ricciardi et al. 2021; Tacconi et al. 2022). Recognizing the importance of these aspects and investigating their interactions with crop diversity are important next steps for related research.

Growing attention is also being given to the role of landscape diversity, as mediated by crop diversity, and how it could contribute to the diversity of the food supply (Nicholson et al. 2021) and human diets (Gergel et al. 2020). Crop diversity has an important role in nutritional production, particularly crops grown by the world's smallholder farmers (Herrero et al. 2017). Our analysis empirically demonstrates how the structure of agricultural landscapes (including field size and the abundance of semi‐natural vegetation) may be an important component of this. Broader landscape structure and land cover complexity are therefore relevant not just to the diversity of production but possibly also to nutritional and food security.

5. Conclusions

Our study applies consistent global spatial datasets to explore hypothetical links between the geographic patterns of crop diversity and key aspects of landscape heterogeneity. Overall, landscapes with greater land cover diversity tend to exhibit higher crop diversity, especially when accounting for landscape composition, particularly cropland extent. This finding builds on extensive previous research examining crop composition and landscape complexity effects on biodiversity in agricultural landscapes. By analyzing thousands of landscapes representative of agricultural geographies worldwide, our study highlights how landscape heterogeneity variables can be relevant to contextualizing current patterns of crop diversity. There are clear spatial patterns of crop and landscape diversity that reflect inter‐regional variation in dominant field size and to some extent mean patch area, so our findings suggest that landscape configuration warrants further attention in assessments of crop diversification. Future studies could address causal mechanisms underlying the crop‐landscape diversity relationship, as well as the temporal diversity of crops and the underlying land‐use and land‐cover changes that alter landscape complexity. Our findings provide a foundation for such future work, including untangling directionality of the effects—whether the compositional diversity of agriculture enhances land cover diversity, or the other way around. Landscape diversity provides a useful indicator to consistently assess global landscapes toward both greater agrobiodiversity and multifunctionality.

Author Contributions

Erin Gleeson: conceptualization, data curation, formal analysis, visualization, writing – original draft. Graham K. MacDonald: conceptualization, formal analysis, methodology, supervision, visualization, writing – original draft, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: gcb70583‐sup‐0001‐Supinfo.pdf.

Gleeson, E. , and MacDonald G. K.. 2025. “The Global Spatial Co‐Variation Between Crop Diversity and Landscape Heterogeneity.” Global Change Biology 31, no. 11: e70583. 10.1111/gcb.70583.

Data Availability Statement

The data that support the findings of this study are openly available on figshare at https://doi.org/10.6084/m9.figshare.30395617. R code will be made available upon request. The CGLS‐LC100 land cover data (Buchhorn et al. 2020) for the year 2015 can be accessed at https://doi.org/10.2909/c6377c6e‐76cc‐4d03‐8330‐628a03693042. Crop diversity data from Aramburu Merlos and Hijmans (2022) are available at https://github.com/aramburumerlos/globcropdiv. Terrain ruggedness (Amatulli et al. 2018) data are available at https://doi.pangaea.de/10.1594/PANGAEA.867115. Dominant agricultural field size data (Lesiv et al. 2019) can be accessed at https://pure.iiasa.ac.at/id/eprint/15526/.