Introduction

The current rate of biodiversity loss exceeds any previously recorded in human history1 and this decline extends to agricultural biodiversity (agrobiodiversity)—the variety and variability of plants, animals, and microorganisms essential for food and agriculture2—posing a threat to global food security3. Agrobiodiversity is multidimensional, encompassing crop species diversity, genetic variation, and agroecosystem functions, yet these dimensions are often conflated or inadequately measured in empirical research4. Evidence highlights the importance of context-specific management of agrobiodiversity to enhance production stability, reduce risk, and support sustainable food supply over time5—aligning with the United Nations Sustainable Development Goals, particularly SDG 2 (Zero Hunger) and SDG 3 (Good Health and Well-being). Failures in agrobiodiversity management—rather than diversity loss alone—can exacerbate vulnerabilities in food availability, access, and diet quality, contributing to persistent nutritional inequities affecting an estimated 2.8 billion people who cannot afford even the least expensive healthy diet6.

In response, functional agrobiodiversity has emerged as a key strategy for building resilient food systems that support ecosystem sustainability and improve access to diverse and nutritious food7,8,9. When strategically managed, agrobiodiversity can play multifunctional role and contribute to improved food and nutrition security, by supporting diverse agricultural production and linking it effectively to consumption through well-functioning markets and institutions10,11,12,13. Enhancing functional agrobiodiversity within agricultural landscapes can improve nutritional stability and access to micronutrient-rich foods, addressing hidden hunger14,15, particularly in rural settings with limited market access.

Despite this potential, agricultural development over the past half-century has largely prioritized maximizing yields of a few staple cereals, often at the expense of more nutrient-rich food crops16. Today, just three grains—rice, maize, and wheat—provide over half of global caloric intake17. In South Asia, where crop diversity has declined18,19,20, potentially undermining the diversity and nutritional adequacy of local diets. Recent studies recognized both the importance and the decline of agrobiodiversity and its uneven expression, underscoring the need for robust metrics to guide planning and intervention—particularly in South Asia, where cereal-dominated systems coexist with high burdens of undernutrition3,17,21,22,23,24.

However, most existing studies15,21,25,26,27 rely on national-level, coarse data typically from FAOSTAT and focus narrowly on species richness. Few efforts have mapped agrobiodiversity at high spatial resolution within countries or incorporated multiple dimensions, particularly functional indicators linked to dietary and nutritional diversity for women and children. Consequently, large-scale spatial patterns of agrobiodiversity in South Asia remain poorly understood.

In this study, we address that gap by assessing large-scale spatial patterns in the production dimension of agrobiodiversity to address where agrobiodiversity is well-expressed, where it remains underutilized, and to identify opportunity spaces for nutrition-sensitive agricultural interventions. Despite the importance of other dimensions of agrobiodiversity—such as conservation, market access, and consumption28,29,30—we focus on dietary food groups and the nutritional content as proxies for dietary and nutritional potential.

While other factors—including women’s education, employment, market access, non-farm income, input use and landholding size—also shape rural household diets31,32,33, production diversity remains a strong predictor of dietary diversity and national food supply diversity21,26,34. Evidence, from South Asia and beyond, shows that increased production diversity is positively associated with dietary diversity35,36, particularly for women and children37,38. Panel data indicate that each additional food group produced contributes a modest (5–20%) improvement in household dietary diversity with crop group diversity positively linked to both the household dietary diversity score (HDDS) and the minimum dietary diversity for women (MDD-W)39,40,41.

We use publicly available subnational crop production data covering 326 crops across 906 districts from Bangladesh, India, Nepal and Pakistan, linked with dietary food group classifications (MDD-W and HDDS) and eight nutrients from country-specific food composition tables, to construct high-resolution indicators for species diversity (SD), dietary functional diversity (DFD) and nutritional functional diversity (NFD). We introduce two composite indices: the agrobiodiversity index score (ABDIS), integrating SD, DFD and NFD to capture current agrobiodiversity status, and the agrobiodiversity potential score (ABDPS), quantifying untapped potential between taxonomic (SD) and functional diversity (DFD and NFD). Spatial statistical models identify hotspots and coldspots to distinguish realized and unrealized diversity, enabling benchmarking and guiding nutrition-sensitive interventions.

This study examines the relationships between taxonomic and functional dimensions of agrobiodiversity and their spatial expression in South Asia. First, we assess whether species diversity aligns with dietary and nutritional functional diversity, or whether systematic mismatches persist. Second, we test whether areas of high agrobiodiversity potential (ABDPS) coincide with regions facing high child stunting, reflecting unrealized opportunities for nutrition-sensitive diversification. Third, we evaluate how crop composition, particularly cereal dominance versus pulses, vegetables, and oilseeds, explains observed trade-offs between species diversity and functional outcomes. Together, these hypotheses guide a spatially explicit analysis to distinguish realized agrobiodiversity from unrealized potential and inform targeted interventions.

Results

Regional crop area and production patterns

We analyzed 906 districts across Bangladesh, India, Nepal, and Pakistan, excluding 50 with missing data, primarily from the disputed areas of Jammu and Kashmir (Fig. 1). Figure 1 shows a clear difference in crop areas and production patterns, which are shaped by geographical characteristics, agroclimatic conditions, irrigation availability, market dynamics, and technology use42. India, with 654 districts (72.2% of the sample), dominates high-production zones, particularly in the Indo-Gangetic Plain, southern states, and eastern regions, where multiple cropping cycles and high-yielding varieties boost output. Bangladesh, with 7% of total sample districts, also has high production, especially in northern regions, due to intensive farming practices. In contrast, Pakistan, which accounts for 12.5% of the sample districts, has larger cultivated areas but fewer cropping seasons, resulting in moderate to low production intensity. Nepal, which account for 8.3% of the total sample districts, predominantly operate low-output, extensive smallholder systems. Excluding 50 districts may limit generalizability for disputed regions, and using data from a single year may not capture temporal variability. To increase food security, policies should promote irrigation and high-yielding varieties in Pakistan’s large-area districts and support smallholder intensification in Nepal, while sustaining intensive systems in India and Bangladesh.

Fig. 1: Crop production and area patterns.
Fig. 1: Crop production and area patterns.
Full size image

Map of district-level area-production typologies based on crop production (metric tons) and area (hectares) from national censuses (2020–21 for Bangladesh, Nepal, Pakistan; 2016–21 for India), classified into low, medium, and high terciles(quantiles) for both production and area.

Agrobiodiversity metrics and regional variation

Agrobiodiversity was evaluated using SD, DFD, NFD and composite indices: ABDIS and ABDPS (see details in methods and supplementary information). The regional median SD is 0.56; DFD and NFD are 0.44 and 0.49, respectively, showing moderate diversity with regional differences (Fig. 2). Nepal shows the highest median SD (0.75) and ABDPS (0.84), reflecting robust smallholder systems with opportunities for diversification. India’s scores align with the regional average (ABDPS: 0.62), with the high-potential districts where cereals and tubers dominate but pulses, oilseeds, fruits, and vegetables are underrepresented. Bangladesh and Pakistan exhibited lower diversity (SD: 0.43 and 0.42; ABDPS: 0.53 and 0.58), indicating specialized cropping systems. Cereals occupy 58% of farmland, particularly in high-production regions, whereas pulses are concentrated in southern and central India, oilseeds in central India; fruits and vegetables (11% of cropland) in parts of Pakistan and India; and non-food crops (e.g., jute, cotton) in Gujarat, Haryana, and southern Bangladesh (more discussion are provided in the Supplementary material). Data limitations, such as missing districts and single-year data, may affect the robustness of these patterns. Policies should promote nutrient-rich crops such as pulses and oilseeds in India’s high-potential districts and encourage diversification in Bangladesh’s and Pakistan’s specialized systems to enhance nutritional outcomes.

Fig. 2: Distribution of agrobiodiversity indicators.
Fig. 2: Distribution of agrobiodiversity indicators.
Full size image

Variability in district-level species diversity (SD), dietary functional diversity (DFD), nutritional functional diversity (NFD), agrobiodiversity index score (ABDIS) and agrobiodiversity potential score (ABDPS) are shown.

Spatial clustering of agrobiodiversity

Spatial autocorrelation analysis revealed distinct clustering of agrobiodiversity indicators and crop classes across South Asia (Fig. 3). High area–production districts cluster in the Indo-Gangetic Plain, southern India, and northern Bangladesh, whereas coldspots—districts with significantly below-average agrobiodiversity indicators—are prevalent in western, southern, and eastern India (e.g., Chhattisgarh, Odisha), eastern Bangladesh, and much of Pakistan. Significant clustering occurred for SD (582 districts: 296 hotspots, 286 coldspots, 280 km distance band), DFD (235 districts: 103 hotspots, 132 coldspots, 240 km), and NFD (535 districts: 191 hotspots, 344 coldspots, 240 km) (Fig. 4). The composite indices show similar patterns, with ABDIS clustering in 490 districts (251 hotspots, 239 coldspots) and ABDPS in 484 districts (235 hotspots, 249 coldspots) (Fig. 5). Fewer than 1% of districts lack neighborhood connectivity, affirming robust spatial patterns. NFD exhibited the strongest clustering, followed by ABDPS and SD. Targeted interventions, such as promoting diversified cropping in coldspots (e.g., eastern Bangladesh, Chhattisgarh) and enhancing market access in hotspot regions (e.g., Bihar, Uttar Pradesh), could strengthen nutrition-sensitive agriculture.

Fig. 3: Spatial autocorrelation of crop classes and agrobiodiversity indicators.
Fig. 3: Spatial autocorrelation of crop classes and agrobiodiversity indicators.
Full size image

Spatial autocorrelation (global Moran’s I) shows a significant clustering or dispersion pattern for crop groups (e.g., cereals, pulses), species diversity (SD), dietary functional diversity (DFD), nutritional functional diversity (NFD), agrobiodiversity index score (ABDIS) and agrobiodiversity potential score (ABDPS).

Fig. 4: Agrobiodiversity status and hotspot analysis.
Fig. 4: Agrobiodiversity status and hotspot analysis.
Full size image

a–f, Maps of the district-level spatial distributions of diversity (a–c) and their clustering (d–f). Species diversity (SD) (a), dietary functional diversity (DFD) (b), and nutritional functional diversity (NFD) (c), scored from 0 (yellow, least desirable) to 1 (dark green, most desirable). Optimized Hotspot Analysis output for SD (d), DFD (e), and NFD (f) shows hotspots (light red: 90%, orange: 95%, red: 99%), coldspots (bluish gray: 90%, light blue: 95%, dark blue: 99%), and not significant (white), excluded (dark gray).

Fig. 5: ABDIS and ABDPS distributions and hotspot analysis.
Fig. 5: ABDIS and ABDPS distributions and hotspot analysis.
Full size image

a–d Maps of the district-level agrobiodiversity index score (ABDIS) distribution (a), scored from 0 (yellow, low overall taxonomic–functional diversity) to 1 (dark green, high overall taxonomic–functional diversity), and agrobiodiversity potential score (ABDPS) distribution (b), scored from 0 (taxonomic–functional parity, yellow) to 1 (high taxonomic–functional disparity, dark blue), with balance (0.4–0.6, light green). Hotspot maps for ABDIS hotspots (c) and ABDPS (d) show hotspots (light red: 90%, orange: 95%, red: 99%) and coldspots (bluish gray: 90%, light blue: 95%, dark blue: 99%); not significant (white), excluded (dark gray).

Stunting and agrobiodiversity overlap

We found that of the 350 districts with very high child stunting, 44% (154 districts) were ABDPS hotspots, indicating significant potential for nutritional diversification, while 14% (50 districts) were coldspots (Supplementary Fig. 7). ABDIS hotspots covered 146 districts (41.7%), with 46 (13.1%) as coldspots, indicating areas with aligned or deficient diversity. SD showed the greatest overlap, with 196 districts (56%) as hotspots and 62 (17.7%) as coldspots. NFD had 89 hotspot districts (25.4%) and 141 coldspots (40.3%), whereas DFD had the lowest overlap, with 43 hotspots (12.3%) and 20 coldspots (5.7%). These overlaps, driven by the priority score (ABDPS), identify taxonomic–functional (nutritional outcome) gaps.

Crop composition and agrobiodiversity potential

Crop composition across ABDPS typologies shapes functional diversity outcomes (Supplementary Table 1, Supplementary Figs. 4–6). Hotspots—districts with significantly above-average agrobiodiversity indicators—show higher shares of cereals (0.74) and pulses (0.09) but lower proportions of sugar crops (0.08) than coldspots (cereals: 0.62; sugar crops: 0.15). Coldspots also have moderate levels of oilseeds (0.10) and non-food crops (0.06), whereas areas with no significant hotspot–coldspot classification display more balanced crop shares (cereals: 0.60) but retain cereal dominance. These patterns highlight that high-potential areas gain diversity through pulses and vegetables, whereas coldspots are constrained by sugar crops concentration and limited crop mixing. Non-food crops, pulses, and fruits show pronounced regional specialization, whereas vegetables, roots, and spices are more spatially dispersed. Policies should promote pulses and vegetables in cereal-heavy coldspots (e.g., eastern Bangladesh, Chhattisgarh) and support market linkages for nutrient-rich crops in hotspots to enhance dietary outcomes.

Correlations and trade-offs

Agrobiodiversity declines with increasing shares of cereals (r = –0.50) and sugar crops (r = –0.27), reflecting trade-offs in cereal-dominated systems that constrain diversification (Supplementary Fig. 8). Conversely, pulses (r = 0.44), oilseeds (r = 0.30), spices (r = 0.24), non-food crops (r = 0.23), and vegetables (r = 0.20) are positively correlated with greater diversity, underscoring their role in broadening crop portfolios. The cereal area share is strongly negatively correlated with sugar crops (r = –0.63), oilseeds (r = –0.44), non-food crops (r = –0.42), fruits (r = –0.38), pulses (r = –0.36), roots and tubers (r = –0.21), and vegetables (r = –0.20), highlighting land use constraints. Post-hoc Dunn’s tests confirmed significant differences in agrobiodiversity between low and high area–production typologies (p < 0.001 for ABDIS and NFD; p < 0.05 for SD), with DFD showing distinct separation between the low- and medium-area groups (p < 0.001) but convergence among the high-area groups (p > 0.8). Only 4% of districts are simultaneous hotspots for SD and DFD, indicating limited overlap between crop diversity and dietary outcomes, whereas 11% overlapping coldspots. Twenty-four percent of districts are simultaneous hotspots for SD and ABDPS, while 21% are simultaneous hotspots for SD and ABDIS, suggesting alignment between species richness and broader agrobiodiversity benefits. The gap between ABDIS and ABDPS highlights underutilized potential for nutritional gains, particularly in cereal-dominated regions. Data limitations may affect the precision of these correlations. Policies should incentivize the integration of pulses, oilseeds, and vegetables into cereal-based systems and support market linkages to translate production diversity into improved diets. The consistent gap between the current status and potential indicates underutilized functional capacity across the region. This gap underscores the need for policies and practices that translate species diversity into tangible dietary and nutritional gains, for example, by promoting underused nutrient-rich crops within dominant cereal-based systems.

Discussion

This study provides a spatially explicit analysis of agrobiodiversity across 906 districts in Bangladesh, India, Nepal, and Pakistan, using two novel indicators: the Agrobiodiversity Index Score (ABDIS; a composite index of current species, dietary, and nutritional functional diversity) and the Agrobiodiversity Potential Score (ABDPS; potential for diversification to enhance nutritional outcomes). Building on global frameworks21,43, we provide high-resolution, district-level monitoring that integrate ten dietary food groups and eight nutrients, setting locally relevant benchmarks. By combining species diversity (SD), dietary functional diversity (DFD), and nutritional functional diversity (NFD) into composite indices—ABDIS and ABDPS—we provide a comprehensive framework to identify where diversity is realized versus underutilized. These measures enable benchmarking of agrobiodiversity at the district level and reveal opportunities to optimize production diversity for nutrition-sensitive outcomes.

A key finding is the taxonomic–functional gap, where high species diversity (SD; median 0.56 regionally) does not consistently translate into dietary (DFD; median 0.44) or nutritional (NFD; median 0.49) diversity, reflecting broader global concerns about the limitations of diversification when not functionally aligned44. As shown in the results, only 4% of districts are simultaneous hotspots for SD and DFD, while a larger share (24%) of SD hotspots align with ABDPS hotspots, indicating considerable underutilized potential for improving nutritional outcomes through better crop alignment, suggesting systemic misalignment between crop species diversity and functional diversity across South Asia. For example, in central and eastern India (e.g., Bihar, Uttar Pradesh) and parts of Bangladesh, high SDs are primarily concentrated in cereals (74% of production), with limited representations of pulses, vegetables, and fruits (Supplementary Fig. 4). This mismatch undermines the capacity of diverse production systems to deliver micronutrient-rich diets, as cereals provide calories but lack micronutrients critical for addressing stunting. The results of the crop composition analysis revealed hotspots with higher pulse shares (0.09) than coldspots (0.62 cereals, 0.15 sugar crops), suggesting that strategic diversification into nutrient-rich crops could bridge this gap.

While cereals dominate production in many districts, it is important to recognize that the nutritional potential of staple crops can be enhanced through targeted breeding and biofortification. Programs led by CGIAR centers, including CIMMYT’s protein-enriched maize and micronutrient-enhanced wheat and rice, demonstrate that staple crops can contribute meaningfully to dietary quality when genetic diversity is harnessed45. Furthermore, global genebanks maintain vast repositories of crop biodiversity that, although not currently widely grown in farmers’ fields, represent an untapped resource for increasing the nutritional richness of existing agricultural systems46. Incorporating these genetic resources can complement efforts to diversify into pulses, vegetables, and fruits, reinforcing functional agrobiodiversity without disregarding the role of staples.

The ABDIS–ABDPS comparison further reveals underutilized potential. While ABDIS captures realized diversity, ABDPS identifies districts where functional diversity could be enhanced without additional land use, revealing opportunities to target interventions. For instance, 44% of districts with very high child stunting overlap with ABDPS hotspots, although this overlap is descriptive and does not imply causality. This suggests that regions with unrealized agrobiodiversity potential coincide with areas of nutritional vulnerability, providing a spatially explicit guide for prioritizing nutrition-sensitive agricultural interventions.

ABDPS coldspots, often in economically and ecologically marginalized regions (e.g., eastern Bangladesh, Chhattisgarh, Odisha), show better alignment between taxonomic and functional diversity but lower overall diversity (SD 0.43–0.42). These areas face structural barriers, including poor market access, limited credit, and weak infrastructure, which constrain diversification, all of which restrict farmers’ ability to diversify their crops47. For example, the results highlight the reliance of coldspots on sugar crops (0.15) and moderate oilseed shares (0.10), constraining their nutritional contributions. By contrast, non-classified (not significant) areas with balanced crop shares (cereals: 0.60) suggest potential for incremental diversification. Smallholder systems, especially in Nepal (SD: 0.75 and ABDPS hotspot), demonstrate resilience but require support to scale nutrient-rich crops. Addressing these barriers through targeted interventions, including public procurement for local pulses, support for women-led vegetable production, and community seed banks, could safeguard agrobiodiversity, improve equity, and strengthen functional outcomes. Aligning such measures with existing programmes (e.g., integrating local pulses into school meal schemes) would help protect emerging agrobiodiversity and improve nutrition in marginalized regions48,49.

The spatial clustering and crop composition results (e.g., hotspots in the Indo-Gangetic Plain, coldspots in western India) provide a roadmap for targeted interventions. In ABDPS hotspots, where high SD does not translate into functional gains, policies should focus on three fronts: (1) incentivizing nutrient-dense crops (e.g., subsidies for pulses, intercropping vegetables with cereals in Uttar Pradesh and Bihar) and training on intercropping vegetables with cereals can diversify production; (2) Market incentives, such as price support for fruits and vegetables, can make these crops financially viable, whereas local procurement for nutrition programs ensures accessibility50,51,52; and (3) procurement for school meals to secure demand and nutrition education campaigns can create demand for diverse diets, encouraging households to retain nutrient-rich crops for consumption rather than exporting them49,51.

In ABDPS coldspots, where diversity is low but alignment is better, strengthening local food systems is key. Short value chains and community markets can keep diverse foods affordable51,53, as shown in the results’ localized distribution of vegetables and roots. For example, in eastern Bangladesh, supporting smallholder vegetable production through extension services and storage facilities can increase dietary diversity. In low-diversity ABDIS coldspots (e.g., parts of Pakistan), overcoming structural barriers requires integrated approaches: climate-smart seeds, input access, and public procurement can diversify farming systems54,55. Although farm-based nutrition programs are common, the link between production and improved diets is not always direct, highlighting the need to address household production and market access together30,56. To lift low-diversity regions out of these traps, policies must go beyond production to strengthen market systems, storage, and private sector value chains for new crops such as fruits and vegetables. Correlations support these priorities: pulses (r = 0.44), oilseeds (r = 0.30), and vegetables (r = 0.20) positively associate with higher diversity, whereas cereals (r = –0.50) and sugar crops (r = –0.27) constrain agrobiodiversity. These findings demonstrate that crop composition explains observed trade-offs between species diversity and functional outcomes. Integrated strategies that combine resource management, locally adapted seeds, input access, climate-smart practices, and public procurement—alongside raising nutrition awareness among rural households and developing nutrition-sensitive value chains—can help translate increased on-farm diversity into better local diets57,58.

Finally, sustaining high-performing ABDIS hotspots (districts with high species and functional diversity) demonstrates resilient, nutrition-sensitive production but faces risks from market concentration and climate stressors59. Policies should integrate agriculture with the health and education sectors, using ABDIS/ABDPS to guide foodshed (that is, areas that are self-sufficient and rely mainly on internal resources) specific planning60. For example, linking high-diversity hotspots (e.g., Nepal’s smallholder systems) to resilient foodsheds via crop insurance and local markets can sustain agrobiodiversity amid climate risks49. Our findings support frameworks for resilient, equitable food systems, reinforcing recent guidance on multi-criteria decision analysis to design nutrition-sensitive agriculture interventions that maximize both biodiversity and nutrition61. Interventions should not only promote diverse crop portfolios but also leverage existing staple crops enhanced for nutritional quality through biofortification and improved germplasm from global genebanks62. Integrating biofortified cereals alongside legumes, vegetables, and fruits can provide immediate gains in nutrient availability while structural diversification strategies are scaled up, particularly in cereal-dominated landscapes.

However, above inferences must be interpreted with caution. First, reliance on crop-based data, may underrepresent small-scale, underreported crops (e.g., kitchen garden species), and exclude non-crop food sources, potentially underestimating functional diversity. Second, single-year data may miss long term seasonal variations, affecting long-term nutritional insights. Third, data gaps, such as missing varietal or underutilized crop data, may underestimate diversity in these regions, and the focus on crops excludes livestock and wild foods critical to rural diets. Fourth, limitations, such as the exclusion of livestock and fisheries, may overlook their dietary contributions, and descriptive stunting overlays do not capture causal factors such as healthcare or income.

While this spatial framework is novel, reliance on administrative crop data likely underrepresents underutilized species (e.g., kitchen garden species), livestock, fisheries and wild foods, which may potentially underestimate agrobiodiversity status. Combining remote sensing and on-farm surveys could capture varietal diversity and enable fine-scale foodshed level mapping43,63,64. Future work should also test context-specific nutrient thresholds to improve ABDPS precision and use multivariate spatial analysis to explore causal links between diversity and stunting. Expanding multi-year data and connecting farm diversity with household dietary adequacy (meeting nutrient intake levels or healthy food plate models) or nutritional yield (nutrients produced per unit area) would clarify food security pathways18,65,66. Integrating ABDIS/ABDPS into decision-support systems requires co-design with communities and policymakers to reflect local realities and adapt to climatic and economic changes, strengthening nutrition-sensitive policy actions. Finally, while this study focuses on crops currently cultivated within districts, future research could explicitly integrate the untapped potential of genebank-held crop diversity and biofortified staple varieties. Doing so would allow a more comprehensive assessment of how agrobiodiversity can be translated into nutrition-sensitive outcomes across South Asia, highlighting pathways to maximize both current and potential dietary contributions.

Methods

Study design and scope

We conducted an agrobiodiversity assessment by providing high-resolution, district-level monitoring of species and functional diversity across 906 districts in Bangladesh, India, Nepal, and Pakistan, building on prior global frameworks9,21,43 and a regional agrobiodiversity database. We incorporated eight nutrients from country-specific food composition tables to construct a detailed agricultural biodiversity database.

Two composite indices—the agrobiodiversity status index (ABDIS; current state of crop diversity and its dietary/nutritional contributions) and the agrobiodiversity potential score (ABDPS; potential for improving diversity to enhance nutritional outcomes)—were developed via three sub-indicators: crop species diversity (SD; the diversity of crop species grown), dietary functional diversity (DFD; crop contributions to dietary needs based on food groups), and nutritional functional diversity (NFD; crop contributions to nutritional outcomes, such as micronutrient availability). District-level agrobiodiversity pattern was assessed by linking crop area fractions and area–production typologies with child stunting data, and analyzing spatial patterns through hotspots and coldspots (Fig. 6). The resulting indicators of realized and potential agrobiodiversity can guide the geographic prioritization of nutrition-sensitive agricultural policies, such as crop diversification or value-chain support, without evaluating specific interventions.

Fig. 6: Methodological framework for district-level spatial assessment of species and functional crop diversity pattern and their linkage.
Fig. 6: Methodological framework for district-level spatial assessment of species and functional crop diversity pattern and their linkage.
Full size image

The workflow integrates subnational crop production with dietary and nutrient data to compute species diversity (SD), dietary functional diversity (DFD), and nutritional functional diversity (NFD). These are combined into two composite indices—Agrobiodiversity Index Score (ABDIS) and Agrobiodiversity Potential Score (ABDPS)—and analyzed spatially to identify hotspots, coldspots, and regions where agrobiodiversity potential aligns with child stunting prevalence area.

Data sources

We sourced recent, publicly available district-level agrobiodiversity data (Table 1) meeting five criteria: (1) public availability, (2) relevance (containing crop area/production), (3) comparability across countries, (4) recency, and (5) online accessibility. The datasets included crop statistics from national censuses (2020–21 for Bangladesh, Nepal, Pakistan; 2016–21 for India, reflecting the latest comprehensive coverage), covering 149, 117, 112, and 115 unique crop entries for Bangladesh, India, Nepal, and Pakistan, respectively.

Table 1 Agrobiodiversity data sources
Full size table

Crop species were classified into ten food groups based on MDDW67 and HDDS68 guidelines, supplemented with custom categories (e.g., edible cash crops, non-food crops). The food group classifications are (1) grains, white roots, tubers, and plantains; (2) pulses (beans, peas, and lentils); (3) nuts and seeds; (4) other vegetables; (5) other fruits; (6) other vitamin A-rich fruits and vegetables; (7) dark green leafy vegetables; (8) condiments and seasonings; (9) edible cash crops; and (10) non-food crops. Nutritional content was extracted from respective national food composition table of each country (Supplementary Table 1), on eight key nutrients: carbohydrates, protein, fat, fiber (macronutrients), and calcium, iron, vitamin C, and β-carotene (micronutrients).

Data harmonization and classification

Crop data were standardized into comparable units (hectares for area, metric tons for production) using country-specific conversion factors to ensure consistency across datasets. To address district size disparities, the crop area was transformed into a fraction of the total crop area per district (Supplementary Fig. 1), enabling comparisons of agricultural efficiency across regions.

Dietary diversity indicators group foods together based on their nutritional similarity and role in the diets65. The ten groups are associated with stronger correlations to micronutrient adequacy than other candidate indicators with different groupings69. Note that these food groups follow culinary rather than botanical definitions, so items such as onions, tomatoes, and peppers are categorized as vegetables rather than fruits. Additionally, the “nuts and seeds” group includes only those seeds commonly referred to in cuisines, such as sesame and pumpkin seeds. However, to ensure consistency across diversity dimensions, we constructed our indices via a harmonized classification of crop types. Our primary model used the strictest definition, excluding non-food crops (e.g., jute, cotton, and tobacco) and marginal dietary categories such as ‘Other beverages and foods’ and ‘Condiments and seasonings’, which fall outside the core ten MDDW food groups and use.

We classify the unique crops according to FAO’s indicative crop classification (ICC)70 to filter to ten major crop classes. We then collected nutritional food composition tables for each country to gather the relevant nutritional data, adhering closely to the FAO/INFOODS guidelines for verifying food composition data. This step was essential for accurate food identification, which involves naming, describing, classifying, and coding crops and foods.

To ensure the accuracy of our data, we manually matched local names, spellings, English names, and scientific names of crops to the crop type data and extracted the corresponding nutritional values. Similar to our dietary classification, we excluded non-food crops and certain edible cash crops. We adopted a stepwise approach, initially seeking the highest quality food matches and subsequently using lower quality matches derived from neighboring countries’ food composition tables (FCTs) or other sources (e.g., USDA and secondary articles). Ultimately, we identified 149 unique crops species in Bangladesh, 114 in India, 108 in Nepal, and 111 in Pakistan, with their respective values for eight key macro and micronutrients: carbohydrates, protein, fat, fiber (macronutrients), calcium, iron, vitamin C (water-soluble vitamins), and β-carotene (related carotenoids) (micronutrients). We consider only the edible portions of these foods, match them with crops, extract the nutrient values, and calculate the total crop-specific nutrient production at district level.

To match district-level census data with spatial data, we used district boundary data from the freely available Global Administrative Areas (GADM). We followed a systematic approach to ensure high-quality matching, which included identifying district names in both the census dataset and the spatial district data, correcting discrepancies in the agricultural census data names, and searching for old names or changes due to district formation and reorganization. As a result, we successfully matched 906 districts out of a total of 956 across the four countries.

Finally child stunting prevalence data (2015–16 India, 2024 Bangladesh, 2025 Pakistan, 2016 Nepal71,72,73,74) in geospatial shapefile format, used as a proxy for dietary outcomes due to the absent consumption data, were joined with districts using unique IDs and analyzed via crosstab to assess overlap with SDs, DFDs, NFDs, ABDISs, and ABDPSs, identifying districts where a very high prevalence of child stunting aligns with agrobiodiversity indicators and sub-indicators. All the data preprocessing steps were performed on the python platform via customized codes.

Calculation of diversity indices

Once the dataset was ready, sub-indicators (SD, DFD, and NFD) were calculated via Shannon’s diversity index using Eq. (1), a metric that quantifies the richness and evenness of species or functional units75. SD measures crop species diversity, DFD assesses diversity across dietary food groups, and NFD evaluates diversity of nutrient profiles, treating crops with identical nutritional functional traits as single units.

Shannon’s diversity metric (H)

$${\rm{H}}={\sum }_{{\rm{i}}=1}^{{\rm{n}}}-({{\rm{P}}}_{i}* \mathrm{ln}{{\rm{P}}}_{i})$$
(1)

where: H = the Shannon diversity index, \({{\rm{P}}}_{i}\) = fraction of the entire population made up of species I and S = number of species encountered

Scores were scaled to a 0–1 range for comparability, with 0 indicating no diversity (e.g., one crop or identical compositions) and 1 representing the highest diversity among districts. This scaling was achieved via a linear transformation76 as shown in Eq. (2).

$${indicator\; score}=\frac{X-\min (X)}{\max \left(X\right)-\min (X)}$$
(2)

where X is the raw indicator value, and min(X) and max(X) are the minimum and maximum thresholds set for each sub-indicator, respectively.

ABDIS was calculated by summing normalized SD, DFD, and NFD scores via a straightforward, unweighted additive approach76, as outlined in Eq. (3) without weighting. ABDPS was derived by combining SD with the average of NFD and DFD, emphasizing the potential for nutritional diversification.

$${ABDIS}=({SD}+{DFD}+{NFD})$$
(3)

Finally, the priority score, which is used to identify districts for agrobiodiversity management, was calculated by subtracting the average NFD and DFD from the SD (outlined in Eq. (4)), highlighting areas where high species diversity does not translate into functional benefits.

—using the following arithmetic expression:

$${ABDPS}=\left({SD}\right)-({NFD}+{DFD})/2$$
(4)

Spatial statistics: global Moran’s I and optimized hotspot analysis

Spatial autocorrelation was assessed via the global Moran’s I statistic77 (see equations in Supplementary analysis: Supplementary Eqs. (6)–(9)), which tests whether the crop classes and the agrobiodiversity indicators and sub indicators are clustered, dispersed, or random on the basis of district locations and indicator values. A positive Moran’s I indicates clustering, with significance determined by z-scores and p-values.

Optimized hot spot analysis (OHSA) uses the Getis-Ord Gi* statistic78,79 (Supplementary Methods) to identify hotspots (districts with high indicator values surrounded by similar districts) and coldspots (low-value clusters), applying false discovery rate (FDR) correction to adjust for multiple testing and spatial dependence. OHSA aggregated incident data and optimized the analysis scale in ArcGIS Pro 3.1.0, producing maps of significant clusters. Spatial statistics extend traditional (nonspatial) statistics to address the issue of independence violations among samples. These methods explicitly model spatial autocorrelation and adjust the degrees of freedom in significance tests to account for the lack of complete independence among closely spaced points. First, we assess the null hypothesis of no spatial pattern among the features via the global Moran’s I statistic. OHSA identified spatial clusters using the Getis-Ord Gi* statistic, with scale optimization based on input features (details in Supplementary Method). Further processing involved exporting data from ArcGIS to CSV, dbf, Excel, and Python to perform further statistical analysis.

Correlation and statistical analysis

We applied a quantile classification method to jointly classify the relative shares of area and production across districts, generating nine area–production typologies ranging from low area–low production to high area–high production (Fig. 1). This approach enabled systematic comparison of agrobiodiversity indicators across distinct production contexts. Pearson’s correlation coefficient was used to examine the relationships between the crop area fractions and ABDIS scores at the 95% confidence level. To compare agricultural typologies with SD, DFD, NFD, ABDIS, and ABDPS, we tested for normality using the Shapiro-Wilk test as well as for sensitivity to different crop inclusion scenarios (e.g., excluding non-food crops and marginal foods in the primary model, including marginal foods in a secondary model, or including all crops for species diversity in a tertiary model), as detailed the in supplementary information (Supplementary Table 1). Non-normal distributions led to the use of the non-parametric Kruskal–Wallis H-test to assess median differences across typologies, followed by Dunn’s post-hoc tests with Holm correction for pairwise comparisons (p < 0.05). Analyses were conducted in Python using the scipy and scikit-posthocs packages.

Policy-relevant analytical framework

This study adopts a diagnostic and decision-support perspective rather than evaluating or implementing policy interventions. The analytical framework is designed to generate spatially explicit indicators of crop agrobiodiversity that are relevant for nutrition-sensitive agricultural planning. Specifically, the constructed indices and hotspot analyses aim to identify (i) districts with high realized agrobiodiversity, where existing production systems already support dietary and nutritional diversity, and (ii) districts with high latent agrobiodiversity potential, where crop species diversity is not yet reflected in dietary or nutritional functional diversity.

By distinguishing between realized agrobiodiversity status and unrealized functional potential, the methods provide an evidence base that can inform the geographic prioritization of future policy actions such as crop diversification strategies, value-chain support, or extension services without assessing the effectiveness of any specific intervention. All policy relevance discussed in this study is therefore inferential and derived from the spatial patterns observed in the data.