The methodology was applied between October 2023 and January 2024 to a sample of 40 farms distributed across five NUTS II regions of continental Portugal—North, Center, Lisbon Metropolitan Area (AML), Alentejo, and Algarve—capturing the diversity of the country’s agricultural landscapes and sociocultural contexts. The sample comprised 20 matched pairs of farms, each including one agroecological and one conventional farm, located within the same territorial area. The regional distribution was as follows: Norte (n = 14), Centro (n = 6), Lisbon Metropolitan Area—AML (n = 8), Alentejo (n = 6), and Algarve (n = 6; Figure 1). This paired and stratified sampling approach was intended to ensure comparability by minimizing variation in farm scale, crop types, local food system characteristics, and edaphoclimatic conditions.

Farms were selected according to three main criteria to ensure internal consistency. First, all farms engaged primarily in horticultural production—accompanied by other complementary agricultural activities—to maintain a consistent production focus. Second, agroecological farms were defined by the exclusion of synthetic inputs, such as fertilizers and pesticides. The non-use of synthetic fertilizers and pesticides was applied as an initial filter to identify farms potentially aligned with agroecological principles, within a broader, multidimensional framework that does not define agroecology solely through input-based criteria but rather seeks to capture systemic transitions.

Data collection was carried out through field visits and semi-structured interviews using a standardized questionnaire, developed according to the methodology proposed by Costa-Pereira et al. (2024). Farmers’ responses were codified in binary or scaled formats, depending on the variable type: binary variables were coded as 0 or 1, while scaled variables captured increasing levels of characteristics such as education level or household income.

Based on these responses, 26 indicators were constructed and scored on a scale from 0 to 4, with 0 indicating low performance and 4 indicating high agroecological performance (in Supplementary Materials). These indicators were grouped into 10 agroecological dimensions, encompassing both agroecosystem management and food system interactions. The eight dimensions related to the agroecosystem were: (1) plant production, (2) animal production, (3) soil regeneration, (4) water cycle regeneration, (5) pest, disease and weed management, (6) ecological synergies, (7) economic synergies, and (8) social synergies. The two dimensions relating to the local/global food system were: (9) interaction with the local food system, and (10) sharing agroecological knowledge.

Each dimension incorporated multiple indicators. For example, the soil regeneration dimension includes practices related to soil fertility management and soil–plant systems, while plant production includes seed management and the integration of trees into cropping systems. Economic synergies reflect internal input use, income stability, and renewable energy adoption. Ecological synergies assess the farm’s integration with the surrounding landscape. Social synergies capture working conditions, gender and youth empowerment, and food security. The sharing of agroecological knowledge includes access to information, traditional know-how, and farmer-to-farmer knowledge exchange. This multidimensional structure allowed for a comprehensive and nuanced assessment of each farm’s agroecological performance, serving as a basis for subsequent clustering and policy-relevant analysis (Table 1).

The methodology included a specific section on public policy preferences to identify the institutional levers that farmers perceive as most effective in supporting the agroecological transition. Farmers were asked to indicate their preferred types of policy measures across four categories: (1) regulatory measures, such as the prohibition or taxation of synthetic chemical pesticides; (2) certification controls to ensure credibility and accessibility of agroecological certification systems; (3) economic support, including subsidies and financial incentives tailored to agroecological farming; and (4) education and awareness-raising initiatives targeting farmers, consumers, and society more broadly.

The analytical process aimed to classify farms according to their level of agroecological adoption and identify key patterns of differentiation. For this purpose, hierarchical clustering was applied to the indicator matrix, using Euclidean distance to compute the distance matrix and Ward’s method for linkage. Ward’s method was selected for its ability to minimize within-cluster variance and enhance the interpretability of the resulting clusters (Johnson and Wichern, 2007).

To assess the robustness of the clustering solution, three additional approaches were tested. First, alternative hierarchical methods—centroid and complete linkage—were applied. Centroid linkage evaluates cluster similarity based on the distance between group means, while complete linkage calculates the maximum distance between observations across clusters, thereby reinforcing internal cohesion. Second, K-means clustering was conducted to minimize intra-cluster variability and explore potential alternative segmentations. This cross-validation framework allowed for the testing of consistency and stability in the cluster solution.

To validate and refine the clustering further, a sensitivity analysis was performed on the grouped questionnaire data. Each agroecological dimension was tested to evaluate its capacity to discriminate between clusters. Given the small sample size and the ordinal nature of much of the data, the non-parametric Kruskal–Wallis test was used to assess the statistical significance of differences across clusters. Following this, Principal Component Analysis (PCA) was conducted on each dimension to identify the most explanatory indicators contributing to overall variance and cluster differentiation (Peres-Neto et al., 2003). To deepen the analysis, individual significance tests were applied to the key variables identified through PCA within each dimension. These results were examined alongside the indicator-level scores to enrich the interpretation of intra- and inter-cluster similarities and differences.

To ensure statistical rigor in profiling the identified clusters, assumptions of normality and homogeneity of variance for multidimensional scaling (MDS) scores were first tested. Depending on the distributional properties, either parametric (ANOVA) or non-parametric (Kruskal–Wallis) tests were performed. All statistical analyses were carried out using IBM SPSS Statistics software (version 29), with a significance level set at 5% (p < 0.05).

Three clusters of farmers were identified based on their performance across the 26 agroecological indicators. To characterize the clusters, the mean and standard deviation of each of the 26 agroecological indicators were calculated, followed by the application of mean comparison tests to assess statistically significant differences among clusters. This analysis enabled the identification of patterns across clusters and the detection of indicators that contributed most to their differentiation (Table 3).

The analysis of the 26 agroecological indicators across the three clusters revealed statistically significant differences for the vast majority of variables, highlighting the heterogeneity in agroecological practice adoption among the sampled farms. Only three indicators—youth empowerment and emigration, producer networks and consumer relations, and relations with the community—did not show statistically significant variation across clusters. These dimensions, which are more closely linked to broader social dynamics and community-level interactions, appear to be less sensitive to the structural and operational differences captured through the clustering process, or may require larger sample sizes to detect significant variation.

Among the indicators that did differentiate the clusters, integration with trees, water saving, and biodiversity presented significant variation across all three clusters. These practices are central to ecological design and conservation, offering critical insight into the extent and nature of the agroecological transition processes underway on these farms.

Cluster 2 consistently achieved the highest scores across most indicators, reflecting a strong commitment to integrated, diversified, and ecologically grounded farming systems. Farmers in this cluster demonstrated widespread adoption of agroecological management practices, particularly in areas such as seed systems, pest and disease control, energy use, income stability, and biodiversity enhancement. This group also reported higher levels of social and institutional engagement, with especially strong scores in indicators related to working conditions, gender equity, and traditional knowledge. The overall profile of Cluster 2 represents farms that are not only adopting agroecological practices but are also consolidating their ecological and social strategies within a broader systems-based approach.

In contrast, Cluster 3 presented the lowest mean scores across most agroecological dimensions. This group lagged particularly in key indicators such as biodiversity, water saving, integration with trees, and pest and disease management, signaling narrow ecological integration. Nevertheless, Cluster 3 exhibited notable similarities with Cluster 1 in several areas, including animal species management, animal welfare, integration of crops and animals, use of internal inputs, and diversity of production activities and services. In the domain of social synergies—especially production and household needs, working conditions, women’s empowerment, and traditional knowledge—Cluster 3 also aligns closely with Cluster 1. These findings suggest that, despite more limited ecological integration, the two clusters exhibit relatively similar profiles in the social domain.

Cluster 1 occupies an intermediate or transitional position between Clusters 2 and 3, with moderate scores across most indicators. It demonstrated strong performance in indicators as soil fertility management, soil–plant management, landscape connectivity, and biomass recycling, showing close alignment with Cluster 2 in these areas. However, in other indicators—such as animal integration, production diversity, and social empowerment—Cluster 1 is closer to Cluster 3. This dual alignment positions Cluster 1 as a strategic leverage point for transition: these farms may already have a foundation of practices and knowledge that, with targeted support, could be deepened into more integrated agroecological systems.

In sum, the clustering analysis identified three distinct yet interrelated profiles that do not represent a linear hierarchy but rather diverse expressions of agroecological engagement. The most significant contrasts were observed between Clusters 2 and 3, where differences in several key indicators exceeded two points. These findings underscore the diversity of agroecological strategies among family farms in Portugal and highlight the importance of differentiated, typology-sensitive approaches to support the agroecological transition.

To further investigate which variables contributed most to cluster differentiation, sociodemographic, technical, and economic characteristics—organized according the 10 agroecological dimensions—are presented and used to compare the three identified clusters. This information was collected through the questions answered by the farmers during the survey (Table 4).

Based on the indicators and characteristics of the farmers included in each cluster, it is possible to define them as agroecological (Cluster 2), proto-agroecological (Cluster 1) and conventional (Cluster 3) farmers. These typologies are not merely statistical groupings, but reflect meaningful patterns in farming strategies, social organization, ecological integration, and cultural identity.

The agroecological farmers, represented by Cluster 2, are the most advanced in terms of agroecological integration and systemic thinking. These farmers are also relatively young, with an average age of 49, and report high levels of formal education, often at the university level. With an average of 16 years of experience, they demonstrate a coherent and deliberate adoption of agroecological principles across multiple dimensions. Their practices include the use of trees for both soil fertility enhancement and erosion prevention, biological pest control methods such as the use of ladybirds and multifunctional plants, and the complete exclusion of synthetic pesticides. Farms in this cluster tend to incorporate structured ecological elements, including flower strips and stone walls, to support biodiversity, although some features like nesting sites for birds and bats remain underutilized. These farmers also engage in composting and renewable energy production, with many partially supplying their own energy needs through solar panels. Socially, they are highly engaged and score significantly higher in indicators related to gender equity, with women frequently involved in decisions regarding animal production, economic activities, and major household assets. Economically, this group shows greater diversification of income sources and improved access to credit. Their practices align with the concept of “transformative agroecology” as described by Altieri and Nicholls (2020), in which farms function as ecologically embedded, socially empowered, and territorially anchored systems. These farmers exemplify a comprehensive, integrated model of agroecological farming that extends beyond agronomic practices to encompass cultural identity, political engagement, and systemic resilience.

The conventional farmers, represented by Cluster 3, are the most experienced group in the sample, with an average of 40 years of farming activity. They are generally older (average age of 59) and tend to have lower levels of formal education, typically having completed secondary education or less. Their farming systems are rooted in long-standing practices and intergenerational knowledge, reflecting deep familiarity with local crops and livestock conditions. While these farmers continue to rely on conventional practices, particularly the use of synthetic herbicides and animal-origin fertilizers, they play an essential role in the rural economy. Their agricultural approach is shaped by practical experience and continuity, often emphasizing productivity and land stewardship in the face of changing policy and market dynamics. As highlighted by Fonseca et al. (2024) and Williamson et al. (2024), these systems face significant ecological challenges, including soil degradation and biodiversity loss, which may compromise long-term resilience. With appropriate incentives and peer-based learning opportunities, conventional farmers could gradually adopt low-input or regenerative practices, paving the way for partial transitions.

The proto-agroecological farmers, represented by Cluster 1, occupy a transitional position between conventional and agroecological systems. They are generally younger, with an average age of 49, and possess relatively high levels of formal education, including a high proportion with university degrees. With an average of 20 years of farming experience, they blend conventional and agroecological practices. Their scores are intermediate across most agroecological dimensions, but they align more closely with agroecological farmers in practices such as intercropping, soil fertility management—including the use of trees for soil fertility—landscape connectivity, and biomass recycling. Although ecological synergies are present, these are often underdeveloped or unstructured, with limited use of features such as stone shelters or flower strips. Their waste management practices and approaches to animal sourcing also differ significantly from both of the other groups. As such, proto-agroecological farmers demonstrate partial ecological integration and growing engagement with agroecological principles but have yet to adopt a fully systemic approach. Their position within the agroecological transition continuum suggests strong potential for advancement, especially with appropriate institutional support and targeted technical assistance. As described by Horstink et al. (2023), Matthews (2022) and van der Ploeg et al. (2019), this group fits the definition of “proto-agroecological”—systems that are moving toward agroecology through incremental practice adoption, innovation, and community engagement. Overall, this group consists of experienced and technically informed farmers who balance conventional and agroecological practices but have yet to fully engage in structured agroecological approaches.

In relation to public policy preferences, four questions were included in the survey to assess farmers’ perspectives on institutional measures to support agroecological transitions. Among these, statistically significant differences between clusters emerged only for support for regulatory measures aimed at restricting harmful inputs, such as synthetic pesticides, including, for example, the creation of taxes on the application of synthetic pesticides. This preference was particularly prominent among agroecological farmers, although it was also mentioned by farmers in the proto-agroecological and conventional clusters, indicating a broader awareness of the environmental impacts of input-intensive practices.

While the other policy-related questions did not show statistically significant differences between clusters, several of these variables were retained in the PCA and offer relevant insights for policy design. Notably, financial support through subsidies was mentioned across all groups. However, some farmers expressed concerns about subsidy dependence, highlighting the need for support structures that promote autonomy and resilience. In addition, several respondents emphasized the importance of continuous and accessible technical assistance, particularly through public extension services or similar advisory mechanisms. The need for improved transparency and enforcement in agroecological certification processes was also raised, underscoring the relevance of effective oversight in building trust and credibility in certification processes.