Huánuco, Peru, is marked by complex relations of food, nutrition, and agrobiodiversity (Figure 2; Malice et al., 2010; Velásquez-Milla et al., 2011; Jones et al., 2018; Zimmerer et al., 2020). Climate, topography, soils, and environmental diversity of Huánuco are representative of valley-upland regions of the Andes Mountains in Peru (Pulgar Vidal, 1996, p. 225) and western South America. The Huánuco Andes extend to ecotones of the Upper Amazon. This “bright spot” of agrobiodiversity (sensu Gould et al., 2021) is also a “hotspot” subject to dynamic agri-food changes including widespread food and nutrition insecurity (Zimmerer et al., 2020) where policy legacies and land privatization (Mayer, 2009) contribute to the urgent need to address SDG 2. Social-ecological and political-ecological drivers of agrobiodiversity change in Huánuco are characteristic of the Andes, Amazon, and global trends (de Haan et al., 2010; Oyarzun et al., 2013; Skarbø, 2014; Arce et al., 2019b; de Haan, 2021; Zimmerer et al., 2021).

This study was sited in three Huánuco landscapes (Quishqui, Amarilis-Malconga, and Molinos-Umari; Figure 2) that are environmentally and socially distinct. Census and municipal-level data guided the structured-random selection of 10 communities with similar elevation-range characteristics in each landscape. 20 households were then randomly selected to participate out of the 25–40 households in study communities. Eligible households met the following inclusion criteria: (1) members were permanent residents of the household, (2) a woman aged 15–49 years was a household member, and (3) field and/or garden crops had been cultivated by one or more members in 2016–17. For the household survey, we sampled 20 households per community. The selected communities had 25–40 households while extremely small communities were excluded. Institutional human subject approvals (IIN in Lima, Peru, and University of Michigan) guided informed consent and research ethics protocols.

We administered a multi-module household survey to the 600 participating households (April–June 2017) that collected data on potential co-variates with agrobiodiversity. It included modules on sociodemographic characteristics, livelihood assets, food security, dietary intake, and livelihood activities, among other topics.

In addition, a quantitative 24-h recall of food intake of the young or medium-age woman used the multiple-pass method (Gibson, 2005). One hundred women from this sample were randomly selected for a second food intake recall interview after the first interview. From recall data, a 10-food group diet diversity score was used to calculate the Minimum Dietary Diversity for Women (MDD-W) indicator, defined as 1 if the respondent consumed five or more food groups in the previous 24 h and 0 otherwise (Martin-Prevel et al., 2015; FAO, 2016). A 15 g minimum cut-off defined consumption of a given food group. Information on co-variates obtained through above methods is marked with “S” in Table 1 (1st column).

One half of surveyed households (n = 300) were randomly selected to participate in sampling agrobiodiversity. Most households cultivated fields (n = 268) and about one half produced gardens (n = 159) as sites for combined market production and home food consumption. The household's principal food-producing spaces were visited with members that participated in sampling (Agrobiodiversity Sampling, AS) using local-name identification and spatial sub-areas. Species-level scientific identification of AS taxa was overseen by local agrobiodiversity experts at the Universidad Nacional Hermilio Valdizan in Huánuco. AS incorporated information on the major food-generating spaces of each household (e.g., rotation history, ownership, and type such as field or garden. These co-variate data are marked “AS” in Table 1 (1st column).

Household-level diversity estimates derived from AS data for fields (n = 268 households) and gardens (n = 159 households) subsequently were used in regression models described below. Our diversity-estimation approach cites specific works since focus on the species level requires general justification (Colwell, 2009: 258; Magurran, 2013; Williams, 2016; Jones, 2017; Zimmerer et al., 2020; Goldberg et al., 2021) and is used specifically to distinguish key food spaces in this study. This species-level focus complements existing agrobiodiversity estimation of cultivars, varieties, and landraces (Smale et al., 2001; Obayelu et al., 2015; Wale and Holm-Mueller, 2017; Dedeurwaerdere and Hannachi, 2019; Gatto et al., 2021), genetic populations (de Haan et al., 2010; Perronne et al., 2017; Arce et al., 2019a,b), and landscapes (Zimmerer et al., 2020), as well as agroecological functional diversity addressed below.

Richness (a count statistic of the number of cultivated and managed food-producing species) and the Shannon diversity index (H = - ∑si=1pi ln p_i) (Colwell, 2009, p. 260)—the latter is an estimate of the relative abundances of species referred to as evenness—were chosen as twin statistical estimates of agrobiodiversity [for statistical formulas, symbols, and rationales see Colwell (2009), Magurran (2013), Smale (2006)]. These diversity measures are widely used individually and in combination for agrobiodiversity research (Benin et al., 2004; Jarvis et al., 2008; Oyarzun et al., 2013; Williams, 2016; Jones et al., 2018; Goslee, 2020; Goldberg et al., 2021) as well as “wild” biodiversity (Colwell, 2009; Hayek and Buzas, 2010). Biodiversity estimates of richness and evenness were visualized on maps and graphs (Figure 5 in Results).

We calculated additional agrobiodiversity estimates using information statistics of the Margalef and Menhinick indices (D_Mg = (S-1)/(ln N) and D_Mn = S/√N, respectively) and alternative evenness measures (modified Shannon diversity index, H' = e^H and two forms of the Simpson diversity index, D = 1 – ∑i=1spi2 and D' = (∑i=1spi2)⁻¹ (Smale, 2006; Colwell, 2009, p. 260; Magurran, 2013). These additional biodiversity estimates (Supplementary Tables 1, 2) were important though less central to this study and less common in existing agrobiodiversity research. Functional diversity, defined as the diversity of species niches or functions (Villéger et al., 2008; Finney and Kaye, 2017; Blesh, 2018), shown elsewhere as complexly related to agroecological multi-functionality (Blesh, 2018), is treated in this study as important general information. Preliminary distinctions of plant functional groups and nutrient management, described methodologically below, suggests a promising area where future research can more fully integrate functional diversity and agroecology as outlined in the Discussion.

Descriptive statistics assessed the hypothesized predictive factors, with calculations of mean, standard deviation and range of each hypothesized determinant and dependent variable (Table 2 in Results). We then applied multiple regression analysis using Stata statistical software package, version 15.1 (2018; StataCorp) to determine associations of AKF-hypothesized determinants with the four dependent variables (crop species richness of crop fields and home gardens, and Shannon diversity index of crop fields and home gardens).

Poisson regressions were fit to main models regressing covariates on cultivated species richness of crop fields and home gardens (Tables 3, 4 in Results). Values are reported as incidence rate ratios (IRR) where a one-unit increase in the independent variable is associated with a percentage increase in the dependent variable based on the IRR (e.g., an IRR of 1.08 equates to an 8% dependent-variable increase). Table 1 describes each independent variable in these models including its hypothesized process of influence on the dependent variables.

Ordinary least-squares regressions were fit to main models regressing covariates on the Shannon diversity index (Tables 3, 4 in Results) in addition to supplementary models using the additional biodiversity indices (Supplementary Tables 1, 2). “Distance” and “current participation in programs” were assessed as AKF-guided independent variables although subsequently omitted since they were found to be statistically insignificant and, due to data limitations, would reduce the utilizable sample of households in regression models. Independent variables used in the garden models differed slightly from the models of crop-field agrobiodiversity, with extent of total garden cultivated area substitute for extent of total cultivated field. The variables “presence of a garden” and “fields with chemical fertilizer use” were omitted in the garden-agrobiodiversity regressions.

Associations of predictor factors and diversity indices were considered consistent with random variability at P > 0.05 (Fisher, 1950, p. 80), with coefficients, Standard Error (SE) and P-values reported in Tables 3, 4 in Results. In addition, a supplement of 16 regression sub-models for grouped AKF factors (farm characteristics and agroecology; diets and nutrition; governance; global change) were estimated for crop fields (Supplementary Tables 3–6) and home gardens (Supplementary Tables 7–10). Breaking out hypothesized factors into sub-models was used to check for possible over-parameterization in the main models. Only factors determined statistically significant in both the main models and supplementary sub-models are reported and discussed below.

Ecological nutrient management and livelihood roles were estimated for key landscape spaces of agrobiodiversity and food production. Nutrition-focused analysis (Jones et al., 2018; Zimmerer et al., 2020) has signaled the importance of both crop fields (locally chakras or parcelas) and home gardens (kitchen or dooryard gardens, locally huertos). Initial field research involving visits with Huánuco food-growers undertaken in 2017 indicated the potential importance of multi-species maize fields (locally maizales) as an additional distinct type specified further in regression results and fieldwork (Results, Tables 2–4, Figure 8).

Preliminary characterization of ecological nutrient management utilized the AS data to estimate uncultivated fallow (2011–2017), crop rotation (2013–2017), and multi-species plantings (2017). Additionally, AS data enabled categorizing plants in relation to management of soil nutrients and general soil maintenance: (1) nitrogen-fixing legume crops for food and forage; (2) woody vegetation and perennials with generally more extensive root zones; and (3) maize that functions as an extensively rooted annual species. Identification of plant-group categories reflects agroecological, agronomic, and soils research on ecological nutrient management (Smil, 1997; Schipanski and Drinkwater, 2012; Meena and Lal, 2018) as well as research on these plant groups utilized in soil and nutrient management in the Andes (Fonte et al., 2012; Vanek et al., 2020). The total of six above-mentioned agroecological management techniques and plant groups were determined as preliminary estimates feasible using the AS data that had been collected primarily for taxonomic biodiversity estimates. Potential expansion of future social-ecological and political-ecology research on agrobiodiversity to include agroecological methods is described in the Discussion.

Additionally, food and income, which serve major livelihood roles (Arce et al., 2019b; Zimmerer et al., 2020), were characterized based on proportional inputs relative to overall self-produced food and overall farm income, respectively. A group of five key informants knowledgeable about local food, agriculture, and livelihoods rated each farm space from “1 = very important” to “5 = notably unimportant.”

Agrobiodiversity sampling (AS) and identification resulted in a total of 92 cultivated species in crop fields (Figure 3). Most frequent among households were maize (Zea mays, maíz; 65.02%), Andean common beans (Phaseolus vulgaris, frejol; 34.98%), potatoes (Solanum tuberosum, papa; 29.37%), Andean squashes (Cucurbita maxima, zapallo, and Cucurbita ficifolia, calabaza; 29.04%), and fava beans (Vicia faba, habas; 18.81%). Home gardens showed 146 species of agriculturally managed plants (Figure 4). Most frequent were onion (Allium cepa, cebolla; 46.47%), oregano (Origanum vulgare, orégano; 40.59%), cilantro (Coriandrum sativum, culantro; 38.24%), peach (Prunus persica, durazno; 34.71%), and chincho (Tagetes elliptica; 32.94%). The species-frequency curves of both crop fields and home gardens were inverse exponential relationships (Figures 3, 4).

Richness of field crops based on agrobiodiversity sampling (AS) with 268 households (Table 2) averaged 3.7 cultivated species/household with the range of 1–24 species. Mean richness of agriculturally management plants in home gardens was 10.2 species per household with a range of 1–33 species (Table 2). Results showed the mean of 7.9 species for all households. This estimated total agrobiodiversity richness, as well as field- and garden-level estimates, did not vary significantly among the three study landscapes. Mapping the species-level richness of agrobiodiversity (combined fields and gardens) illustrated the notable occurrence of household-level variation and the lack of geographic patterning or clustering either within or among the study landscapes (Figure 5). Mean values of the Shannon diversity index were estimated as 0.70 and 1.83 for the cultivated and managed species of crop fields and home gardens, respectively.

Results showed the small extents of total cultivated crop-field and home-garden areas (mean values of 0.361 hectares and 0.0925 hectares, respectively) among sampled households. Similarly, low values were estimated regarding food security (i.e., high food insecurity; 58.7%), income, education, and elevation range (Table 2). Estimated household-level capabilities for food production included field numbers (2.5/household), self-produced food in the diet (36.5%), traditional foods in the diet (34.2%), and degree of agricultural commercialization (42.4%) (Table 2). These values reflected limited resource access and hybrid traditional-modern food customs including combined non-market/market linkages.

Certain estimated conditions showed large standard deviations. This high variation occurred in legume crop rotations that were defined as leguminous food and forage crops (mean index value 0.60; standard deviation 0.62) and numbers of fields with chemical fertilizer use (mean value 0.64; standard deviation 0.77). Mean climate change adaptations and elements of agrobiodiversity-loss awareness were low-moderate (2.67 and 1.22, respectively). Sample sizes for results estimated in this paragraph and the preceding varied from 100 to 300 households with most estimates statistically robust. Detailed definitions, data sources, and statistical estimates are given in Tables 1, 2.

Social-ecological and political-ecological factors incorporated in this study's AKF-guided model explained ~16% of variability in the species richness of crop fields (RIRR, Pseudo R² = 0.16, Table 3). These factors accounted for 45% of the variability in the species evenness of these fields (OLS, R² = 0.45, Table 3). In the case of home gardens, social-ecological and political-ecological factors of the AKF-guided model explained ~16% of variability in the species richness of these spaces (RIRR, Pseudo R² = 0.15, Table 4). These factors accounted for 24% of the variability in the species evenness of these fields (OLS, R² = 0.45, Table 4). Social- and political-ecological factors of the AKF-guided models produced similar results to Tables 3, 4 using other biodiversity measures as the dependent variables (Materials and Methods; Supplementary Tables 1, 2).

Both the species-level richness and evenness of field crops were significantly associated with a set of five factors (Table 3). Field number and presence of a multi-species maize field (maizal) were most significant in the main model of crop fields (P < 0.001; Table 3). These highly significant statistical associations occurred across the range of models using the additional agrobiodiversity measures (Supplementary Tables 1, 2) as well as the AKF sub-model of farm characteristics (Supplementary Table 3).

Field-level richness and Shannon diversity measures showed that legume crop rotation was strongly associated though less significantly (P < 0.05; Table 3). Additional factors, types of association, and significance levels in the main model of crop fields (Table 3) varied among the positive associations of species richness to garden presence (P < 0.001) and post-secondary education (P < 0.01). Field evenness was negatively associated with the presence of a garden (P < 0.05) and the sub-area of the Molinos landscape (P < 0.05).

In home gardens, species-level richness was significantly associated with a set of nine factors and the Shannon evenness measure with one (Table 4, 2nd and 3rd columns). Age of household head was positively associated at highly significant levels with the species richness (P < 0.001) and species evenness of gardens (P < 0.01) (Table 4). Other factors showing high and positive associations with the species richness of home gardens were garden cultivated area (P < 0.001), field number (P < 0.01), traditional food in diet (P < 0.01), Quechua language (P < 0.01), household income (P < 0.001), and extent of post-secondary education (P < 0.01). Additional factors showed significant positive associations albeit at lower statistical levels. These included the sub-area of the Molinos landscape (P < 0.05) and degree of product commercialization (P < 0.05).

Individual social- and political-ecological factors identified as statistically significant represented each the four AKF themes (Figures 6, 7) and offered comparisons to previous agrobiodiversity models (last column of Table 1) as detailed in the Discussion.

Agroecological and livelihood characteristics were compared among keystone agrobiodiversity-and-food spaces comprised of multi-species maize fields, other crop-field types, and home gardens. Identification of multi-species maize fields (maizales) as an additional keystone space was based on above regression results and fieldwork information (Figure 8), including conversations with local people explaining importance. Following above results, further focused analysis of multi-species maize fields as a key space (Supplementary Table 11 and Figure 6) showed the highly positive effect of legume crop rotation (P < 0.001), moderate-high correlation with overall field area (P < 0.01), and negative association with agricultural commercialization (P < 0.01).

Fallow and crop rotation were moderately and highly common in multi-species maize fields (35.2 and 86.2%, respectively) and other field types (32.7 and 76.1%, respectively) (Table 5). By contrast, these two practices were applied at low levels in home gardens (1.6 and 4.9%, respectively) (Table 5). Intercropped polycultures (planting areas with more than one food species), crop rotation, and legume crops occurred more commonly multi-species maize fields than in other fields. Multi-species mixtures were also moderately common in other crop fields (48.3%). Plant groups with distinct functional roles and traits—legume food and forage crops; the suite of trees, shrubs, and perennials managed in cultivation; and maize—occurred at different levels in the three key agrobiodiversity-and-food spaces (Table 5).

Estimation of the frequency of legume crops and edible managed plants was distinctive among multi-species maize fields, crop fields, and home gardens (71.4, 37.1, 34.6%, respectively; Table 5). Leguminous taxa provisioning important foods in the sample consisted of Phaseolus vulgaris (frijol, vainita, or numia), Inga feuilleei (pacay), Caesalpinia spinosa (tara), Pisum sativum (arveja), Vicia faba (haba), Lupinus mutabilis (tarwi), Arachis hypogaea (mani), Inga edulis (guaba), Lens culinaris (lenteja), and (Medicago sativa alfalfa; important as an animal food) as well the widespread ground-covering clover Trifolium spp. (trebol). A total of 51 tree species were managed and cultivated, as well as 20 species of perennial herbs.¹

Estimated relative levels of dietary and income importance varied substantially among the key agrobiodiversity-and-food spaces (Table 5). Multi-species maize fields (maizales) typically rated intermediate between crop fields (highest levels) and gardens (lower levels yet still important to diet).

This study's use of the interdisciplinary Agrobiodiversity Knowledge Framework (AKF) generates reflection on and comparisons with other SDG 2-relevant research. Overall, the AKF framework was thematically comprehensive, theoretically cohesive, and methodologically innovative in predicting agrobiodiversity that is linked to SDG 2-related nutrition security in our overarching design that integrates the agrobiodiversity analysis here with focus on food, nutrition, and diet research (Figure 1B; Jones et al., 2018; Zimmerer et al., 2020).

Results showed significant associations across each of the four main themes of AKF predictive factors (Figure 1A; agroecology-food/nutrition-governance including sociocultural factors-global change impacts) that provide insights to address SDG 2. First, the AKF-guided models offer a demonstration of agroecological factors (e.g., legume crop rotation; Table 3) positively linked to agrobiodiversity. This linkage combines with results on the importance of leguminous food crops to nutrition wellbeing (Jones et al., 2018). Farm-level, resource-access factors were similarly important. For example, access to a garden both determined agrobiodiversity outcomes as shown in this study while it was also associated with favorable nutrition outcomes (Jones et al., 2018). This study's AKF-guided demonstration of well-proven linkages to nutrition build on initial calls (Jarvis et al., 2007; Hajjar et al., 2008) to provide concrete evidence of agrobiodiversity functions that can address SDG 2.

Second, the AKF shows that food and nutrition can operate as significant predictors (Table 4 and Supplementary Table 8). This AKF result is novel since the predominant view is to treat nutrition, food, and other SDG 2-related conditions solely as outcome variables. Instead, it highlights influential bi-directional interactions in which food and diet can serve as model inputs to explain agrobiodiversity, We urge this insight be built into SDG 2 approaches.

Third, AKF-guided analysis of governance predictors, including socioeconomic factors, were significant in the results (e.g., income; Table 4). Other significant governance factors were sociocultural factors in the changing spatial strategies of food-growing (multi-species maize fields), agrobiodiversity knowledge (Quechua language), and institutional influences (e.g., school-based education; Tables 3, 4). Each of these factors potentially serves as a strong linkage in addressing SDG 2, including other governance factors (such as the potential major influence of seed systems Arce et al., 2019a) and public-good policies (Graddy-Lovelace, 2021).

Fourth, results showed AKF-SDG linkages involving global-change factors in the significant association of the number of fields with agrobiodiversity (Table 3) since the dispersion of fields provides adaptations to weather variation propelled by climate change. At the same time, the resulting higher level of crop diversity has been shown in our related research to predict favorable nutrition outcomes (Jones et al., 2018).

Comparability of model results is needed to advance the use of agrobiodiversity and agroecology to address SDG 2. The AKF supported a much-needed, cohesive approach to interdisciplinary research to enable rigorous cross-case comparisons and policy dimensions ranging from incentives and capacity-building to new initiative such as participatory varietal selection using agrobiodiversity to address SDG 2 (Scurrah et al., 2019). Here we briefly use AKF-guided results to discuss comparisons with other agrobiodiversity studies relevant to SDG 2 approaches and conclude the section by synopsizing this study's model results.

This study's model results show various similarities when compared to related agrobiodiversity studies holding promise for SDG 2 approaches (e.g., Pseudo R² = 0.1883 of species-level agrobiodiversity in important research linked to food consumption and security in the Northern Ecuadorian Andes; Skarbø, 2014, p. 723). In addition, our results found similar factors to be significance (e.g., field number and education level). Comparisons are partly limited, however, since the other research did not model determinants of evenness or garden-specific agrobiodiversity [see also Velásquez-Milla et al. (2011) on Huánuco agrobiodiversity custodian farmers that identifies similar statistically significant factors but does not report model-level results; Skarbø, 2014]. As a result, this study recommends comparisons to thematically related social- and political-ecological modeling of agrobiodiversity evenness, such as species-level agrobiodiversity in coastal Nicaragua (R² = 0.34; Williams, 2016, p. 234) whose general similarity suggests SDG 2 promise.

Our model-level results on the social- and political-ecological determination of home garden agrobiodiversity in Huánuco suggest extended geographic comparison to SDG 2-relevant research in the neighboring Amazon region (e.g., Ban and Coomes, 2004, p. 353). Individual factors identified as highly significant in this study (e.g., garden area and gardener experience) were similar (Ban and Coomes, 2004, p. 353), though different statistical techniques and lack of model-level estimation and evenness estimates limit further direct comparison.

Finally, this study's results from regression models account for species richness to a similar degree in both crop-field and home-garden analysis (Pseudo R² values of 0.16 and 0.15, respectively) while evenness, as measured by the Shannon diversity index, differs substantially among crop fields (R² = 0.45) and home gardens (R² = 0.24). Relative species abundance in home gardens is known to vary depending on the complex characteristics of individual households (Ban and Coomes, 2004; Coomes and Ban, 2004; Wezel and Ohl, 2005; Perrault-Archambault and Coomes, 2008; Whitney et al., 2018), which may contribute to the lower results of the evenness model.

This study's estimations of agrobiodiversity levels and specific social- and political-ecological factors offer valuable specific comparisons to SDG 2-related research as well as general research advances. Our result on the species-level richness of crop fields (mean 3.7 species/household) was similar though less than findings in the Ecuadorian Andes (7.22 species/household; Skarbø, 2014, p. 714) whose fields traversed a significantly larger elevation gradient (1,000 meters). This study's results on specifies-level richness of home gardens (mean 10.2 species/household) resembled the richness of cultivated species of fruits, vegetables, and herbs (10.2 species/household) that were grown the Ecuadorian Andes (Skarbø, 2014, p. 714). Furthermore, this study's results on total agrobiodiversity species richness in Huánuco, when summed across the households' crop fields and home gardens, resembled an additional study of the Ecuadorian Andes (Oyarzun et al., 2013, p. 525).

Another useful comparison is to local high-agrobiodiversity custodian farmers in Huánuco (Velásquez-Milla et al., 2011) and high-level agrobiodiversity hotspots elsewhere in the Peruvian Andes (Arce et al., 2019a). The current study complements these others, while our sample design and methods were distinct since we integrated focus on agrobiodiversity's roles in SDG-related diets and nutrition that included the randomized sampling of households (see also Jones et al., 2018; Zimmerer et al., 2020).

Important to highlight as a research advance is this study's demonstration that combined biodiversity richness and evenness estimates are needed to understand agrobiodiversity's nexus role connecting to both food-nutrition, such as SDG 2, and agroecology.

Finally, individual factors determined to be significantly associated with agrobiodiversity in this study (see previous sub-section) conform hypotheses in Table 1 and need to information expanding SDG 2 research globally. Overcoming barriers to link or couple agrobiodiversity access and related SDG 2 benefits for food sovereignty (Zimmerer et al., 2020) will require promoting agroecological management, garden cultivation, and livelihood improvement in addition to income and education as parts of the agendas of social justice and food and nutrition security.

Select factors not showing statistical significance in this study's model results, such as field size and elevation range (Tables 3, 4), help to explain contextual influences. Extremely limited size of cultivated areas in this study (0.404 hectares/household; Table 2) is typical of many places in the Andes (e.g., 0.5-hectare farms are common in the central highlands of Ecuador; Oyarzun et al., 2013, p. 523) and smallholder farming globally (van Vliet et al., 2015). Likewise, this study's result on the limited elevation range among fields (Table 2) did not result in a significant model result though it is consistent with fieldwork observations of the clustering of small-size fields and gardens near residences. Additionally, this study's findings showed the high frequency of global change factors among surveyed households (e.g., climate change adjustments and elements of agrobiodiversity-loss awareness; Table 2), though this study's regression results did not reveal specific statistical associations to these variables.

This study's focus on new descriptive, statistical, and conceptual understandings of the role of multi-species maize fields (maizales) as a Keystone agrobiodiversity-and-food space is designed to offer a novel contribution and to build on previously hypothesized functions (see multi-species maize fields, Table 1). Model results demonstrate that multi-species maize fields, which are distinct due to the types and extent of intercropping (see Results, Agroecological and Livelihood Characterization of Key Spaces), hold a high level of significance in agrobiodiversity relationships (Tables 3, 4). Multi-species maize fields comprise a key space of broad significance to the nexus of agrobiodiversity-agroecology-SDG 2.

We propose the concept of “Keystone agrobiodiversity-and-food space” to describe multi-species maize fields (maizales) owing to multi-faceted functions. These fields of Huánuco incorporate widespread inter-plantings of Andean common beans (frejol, Phaseolus vulgaris), Andean squash (most commonly zapallo, Cucurbita maxima), and arracacha (giant Andean carrot or parsnip, Aracacia xanthorrhiza), among other species. Agrobiodiverse maizal assemblages, which are extensive in western South America, are distinct while partly resembling the well-known milpa system of Mexico and Central America that is agrobiodiverse and nutritionally important (Toledo and Barrera-Bassols, 2017; Novotny et al., 2021).

Notable nutrient management techniques characterize the agroecological functions of the multi-species maize field as a “keystone agrobiodiversity-and-food space” with extensive rotation of crops (86.2% of maizales incorporate crop rotation; Table 5). Various agrobiodiverse species depend on seed flows and rotated planting sites that link the maizal system to other fields. For example, Huánuco households undertaking the non-maizal field cropping of Andean maize, Andean common beans and Andean squash species frequently depend on maizal-based seed sources and rotated planting sites, and vice versa. A second major function is continued utilization and development of agrobiodiversity management knowledge. This role is crucial since cultivated area and field number are extremely limited among Huánuco households (Table 2). In this context, the multi-species maize field (maizal) enables the crucial continuation of knowledge and practices of production as well as food processing and consumption that are vital to agrobiodiversity-SDG 2 linkages.

Another key agrobiodiversity-support function of the Andean multi-species maize field is the concentration of livestock grazing on above-ground plant residues following harvest. This emphasis is crucial to the functioning of ecological nutrient management of agrobiodiversity-containing maizales as well as nearby fields. Home gardens (locally huertos), which are distinct from multi-species maize fields, are also “keystone agrobiodiversity-and-food spaces” though they contain lesser degrees of the linkage functions described above.

This study's concept of “keystone agrobiodiversity-and-food spaces” is supported by advances in ecological theory extending the keystone species idea from an original food-web focus to other connectivity (e.g., Davic, 2003). An analogous development has occurred in the concept of “cultural keystone species” (Coe and Gaoue, 2020) being expanded to agroecological applications (Nabhan, 2018; Zapico et al., 2020). Use of “keystone” here signifies that addition or loss leads to major changes in occurrence or abundance of other species.

The “keystone agrobiodiversity-and-food spaces” concept is designed to rework the single-taxon definition of a keystone species to one centered on the vital and distinct roles of spatially, culturally, and agroecologically distinct suites of interacting species. Spatial dynamics and influence of food-generating units is traditional in farming and land use (Brookfield, 2001; Mayer, 2018). Changing spatial dynamics of agrobiodiversity pose new challenges and opportunities regarding agroecological sustainability (De Molina et al., 2019). Supporting the positive factors enabling multi-species maize fields (Supplementary Table 11 and Figure 6), such as legume crop rotation, will need to overcome access barriers (e.g., currently multi-species maize fields are linked to larger, less resource-poor farms among indigenous Huánuco smallholders; Supplementary Table 11; Zimmerer et al., 2020).

In sum, the multi-species fields or maizales of Huánuco function as a distinct, locally recognized, and valued agrobiodiversity-and-food space with widespread intercropping, associated agrobiodiversity, and nutrient management. The latter's characterization (Table 5) is preliminary in scope and is intended to stimulate research. We anticipate other “keystone agrobiodiversity-and-food spaces” to include the multi-species, high-agrobiodiversity fields of Andean tuber crops in sectoral fallow management (also known as common field agriculture; Arce et al., 2019b; Vanek et al., 2020). Milpa agriculture of Mexico and Central America is another vital space occurring in highly dynamic contexts (Tamariz, 2022) with key linkages and agroecological functions of agrobiodiversity anticipated to be changing rapidly.

Finally, this study suggests future research avenues centered on AKF-guided approaches to addressing SDG 2 through agrobiodiversity models integrating one or more emphases of agroecology, biodiversity science, and comparisons with existing studies. It indicates the promising role of “keystone agrobiodiversity-and-food spaces” well-suited to expanded integration with these areas of emphasis. Further new research is needed on functional diversity and multi-functionality in agroecology (Blesh, 2018) suited to SDG 2 goals. New AKF-guided modeling approaches related to SDG 2 goals could involve alternative sampling and statistical methods such as Principal Components Analysis, path analysis, constrained ordination, permanova, systems modeling, and reduced-variable parsimonious models.