Determination of Typologies of Andean Suburban Agroecosystems in Southern Ecuador

Abstract

The identification of producer typologies is a crucial tool for understanding the heterogeneity of agroecosystems and designing targeted policies. Andean agroecosystems, particularly those in rapidly suburbanizing areas, have been understudied in this regard, creating a critical knowledge gap. This study addressed this void by determining the typologies of smallholder agroecosystems in the suburban periphery of Cuenca, Ecuador, by applying an unsupervised machine learning technique, Partitioning Around Medoids (PAM) Clustering, to survey data from 293 farmers. Our analysis revealed three distinct typologies, highlighting a socio-economic and productive gradient defined by income sources, market access, and agrochemical use. The typologies range from economically vulnerable households to more commercially oriented and environmentally sustainable ones, underscoring the complex interplay between livelihoods strategies and environmental management. This research provides one of the first empirical typologies of suburban Andean agroecosystems, demonstrating the value of unsupervised learning for capturing farm heterogeneity in data-scarce contexts. The findings offer a robust evidence base for moving beyond one-size-fits-all approaches, enabling the design of differentiated agricultural and territorial policies that enhance sustainability, equity, and resilience at the rural–urban interface.

1. Introduction

Suburbanization is a dynamic, multidimensional process driving profound social, economic, and environmental transformations, which in turn reshape the functional and spatial structure of urban areas [1]. This process has spurred residential expansion into formerly agricultural areas at the urban fringe [2]. This zone, often termed the rural–urban interface, constitutes a complex landscape molded by a multitude of socioeconomic forces [3]. Therefore, suburban spaces have presented both opportunities and challenges for farmers. For instance, the emergence of a new customer base—typically urban residents—has provided farmers with new opportunities for higher-value crops. Consequently, farmers have demonstrated remarkable adaptability by adjusting their production systems to capitalize on emerging economic opportunities in the urban periphery [4]. This rural dependency on urban areas has been emphasized by Bock [5], who argues that urban dynamics will define rural spaces, and that rural areas will increasingly depend on their ability to meet urban needs. Suburbanization is a global phenomenon observed across diverse geographical contexts, including Latin American cities. Here, demographic explosions have led to farmers concentrating around urban peripheries seeking subsistence [6].

Traditionally, agroecosystems have been regarded as complex ecological systems modified by humans to produce food, fiber, and other agricultural products. Their complexity stems primarily from the interaction between socioeconomic and ecological processes [7]. Various definitions of agroecosystems exist, depending on their composition and purpose [8]. However, a general and widely accepted definition characterizes them as systems of biological and natural resources managed by humans, primarily for food production but also for other socially valuable non-food goods and environmental services [9]. Gliessman [10] maintains that an agroecosystem is created when humans manipulate and alter an ecosystem to establish agricultural production. The diversity of agroecosystems exhibits complexity and stability similar to that of natural systems. Therefore, understanding this diversity—characterized by a mosaic of interconnected elements shaped by resource flows interacting with local rural cultural practices—forms the foundation for sustainable agroecosystem management [11]. The structure and function of agroecosystems are largely determined by local context, including the interaction between ecological conditions (encompassing biological, geological, and chemical factors) and social factors such as farmers’ economic needs, cultural and spiritual values, social structure, and technology [12]. Humans play a crucial role within agroecosystems. Consequently, to fully characterize an agroecosystem or identify potential alternative management options, any agroecosystem analysis must incorporate indicators of the economic, cultural, and social forces that shape human activities [13].

Approximately 28% to 30% of the Earth’s land surface is occupied by agricultural land (cropland and pastures). This cultivated area extends across all continents, covering roughly 1.3 billion hectares and encompassing diverse agroecosystems—from large-scale systems to small farm plots and fragmented cultivation zones [8]. Small agroecosystems, typically represented by smallholdings or family farms, dominate rural landscapes across the Global South while persisting in the Global North [14].

Rural households employ heterogeneous livelihood strategies conditioned by environment-specific constraints and opportunities. Agroecological systems, market structures, and cultural contexts thus generate regionally distinct land-use configurations and farm management approaches [15]. Significant intra-community variation exists among households due to differential resource endowments, production orientations, ethnic backgrounds, educational attainment, experiential knowledge, and managerial capacities [16], yielding diverse natural resource governance approaches. Policy frameworks aiming to assist resource-poor farmers must therefore prioritize understanding this multi-scale variability—occurring within farms, between farm systems, and across geographic locations [15]. In this context, capturing agroecosystem heterogeneity through farm typology analysis is crucial for designing sustainable policies and enhancing technology adoption [17]. Particularly, understanding the variability of Andean agroecosystems—characterized by smallholder farms—is essential for accurately representing their reality.

A farm or agricultural holding constitutes the basic unit in typology analysis [18]. Farm typologies serve as the foundation for analyzing farm performance and rural livelihoods [19]. However, agroecosystem typologies have been employed for diverse purposes. Sarker et al. [20] summarize several applications, including: studying greenhouse and climate-smart technology adoption, assessing food security, evaluating resource efficiency, identifying potential adopters of alternative farming methods, and classifying farm categories. These authors emphasize that typologies form the basis for developing appropriate interventions and policies. Nevertheless, farm-level typological analyses are essential for developing tailored management strategies for these agricultural production systems. Research on smallholder farmers remains notably scarce in Ecuador’s Andean highlands. Vanegas et al. [21], in their examination of peasant agriculture in southern Ecuador’s Andes, demonstrate that understanding Andean smallholder systems requires deeper, more inclusive analytical frameworks.

The development of typologies serves as a methodological tool to understand agroecosystem complexity by providing a simplified representation of their diversity through classification into homogeneous groups based on multiple characteristic combinations [19]. Various approaches exist for developing typologies [22]. At a general level, however, Maton et al. [23] distinguish two classes of farm typologies: those employing positivist methods based on statistical data analysis, and those using constructivist approaches grounded in expert knowledge. Most studies related to farm typology construction have utilized positivist methodologies [24].

The criteria defining agroecosystem or farm types may be derived from local stakeholders’ knowledge or directly from farmer survey data [24]. Traditional data analysis has undergone a significant transition, with automated statistical techniques largely replacing manual analysis. Currently, multivariate statistical techniques are most frequently employed for typology construction. However, typologies remain constrained by study objectives, data availability, and sample size [19]. Regarding statistical techniques for typology development, unsupervised machine learning methods—particularly clustering approaches—are widely used. Graskemper et al. [25] provide a synthesis of studies applying clustering techniques to develop farmer typologies across different world regions. Unlike supervised learning, unsupervised learning does not predict outcomes but rather identifies previously unrecognized patterns in data. These exploratory methods can subsequently inform predictive model development [26].

Therefore, the present study addresses a critical gap in both the scientific and applied literature by characterizing the diversity of Andean suburban agroecosystems and identifying farm typologies that reflect distinct socioeconomic, productive, and environmental profiles. While most typology research in Ecuador has focused on rural or commercial farming systems, the suburban scale remains largely undocumented despite its growing importance for urban food security and territorial planning. This study contributes to filling that void by providing an empirical basis for understanding how suburban households organize their production, manage natural resources, and integrate into urban markets. Accordingly, this study aimed to (i) characterize production agroecosystems in suburban areas near Cuenca city in the Southern Ecuadorian Andes, and (ii) determine the principal smallholder farm typologies based on livelihood systems using an unsupervised machine learning method.

2. Materials and Methods

2.1. Study Area

The city of Cuenca (officially named ‘Santa Ana de los Ríos de Cuenca’) is Ecuador’s third largest city and the second largest Andean city after the national capital, Quito. Located in southern Ecuador, Cuenca serves as the capital of Azuay province. The city lies within an inter-Andean valley at 2550 m above sea level, characterized by cool to cold temperate climates [27]. Administratively, Cuenca belongs to the canton of the same name. Its urban area comprises the cantonal seat (formed by urban parishes), surrounded by rural parishes [28]. The rural parishes nearest to Cuenca’s cantonal seat (Cuenca city) constitute what this study considers the suburban zone (Figure 1). Five parishes—hereafter referred to as districts—were selected from this area as they represent Cuenca’s primary agricultural peripheries. The study districts are (a) San Joaquín, (b) Ricaurte, (c) Sinincay, (d) Baños, and (e) Sayausí. All operate under decentralized autonomous parish-level governments (GADs).

2.2. Sampling and Data Collection

The data were obtained from farmer surveys conducted as part of the research project ‘Re-territorialization of Local Agri-food Practices in Semi-urban Areas of Cuenca Canton,’ funded by the Universidad de Cuenca (Ecuador). The study population consisted of all agricultural households registered in the five districts, totaling 1090 households. These registries were provided by each district’s local government (GAD). Data were collected through a questionnaire administered to a sample of 293 farmers during the second half of 2021. A proportional stratified sampling design [29] was applied to ensure representation across districts.

The questionnaire was administered as a digital form developed using KoBoToolbox (KoBoToolbox, Harvard Humanitarian Initiative, Cambridge, MA, USA; available at: https://www.kobotoolbox.org, accessed on 15 January 2022). The questionnaires were administrated by a trained team of researchers, technicians and students from the University of Cuenca. Questionaries were applied in person. During fieldwork, several challenges were encountered, including reluctance of some farmers to disclose information, and logistic restrictions associated with the COVID-19 pandemic. These issues were addressed through collaboration with local parish councils, which facilitated access to communities and helped established trust with respondents. Anonymity and confidentially were guaranteed to all participants.

While the original survey contained 146 questions covering multiple producer characteristics, 117 questions focused on ethnographic and organizational aspects (e.g., customs, beliefs, language, myths, ideologies) and organizational management themes (SWOT analysis of producer associations), which are part of a separate study. Consequently, the current analysis considers only 29 questions addressing farmers’ socioeconomic characteristics, agricultural practices, and marketing strategies (

Table 1. Variables of small farmers used for the construction of typologies.

	Variable Code	Variable Definition	Scale/Measure
Socio-economics	tipo	How the person is considered	producer/intermediary/producer-intermediary
	tam_hogar	People who make up the household	1–2/3–4/5–6/7–9/>9
	adult_trab	Adults who work in agriculture	1–2/3–4/5–6/>6
	ingresos_agri	Income from agriculture	yes/no
	fuente_ingreso	Sources of economic income	agriculture/raising and commercializing animals/food preparation and sale/sale of labor/others
	remesas	Remittances	yes/no
	Ingresos	Monthly income (USD)	<400/401–715/>715/prefer not to answer
	ingreso_aliment	Income for food	0–25%/26–50%/51–75%/76–100%
	ayud_econon	Receives economic assistance from institutions	yes/no
	pro_ser_cuenca	Products/services from Cuenca	yes/no
Agricultural aspects	terreno	Land ownership	rented/sharecropped/communal/family patrimony/owned/loaned/inherited
	cultivo1	Main crop	vegetables/fruit trees/pastures/corn/roots and tubers/other
	cultivo2	Secondary crop	vegetables/fruit trees/pastures/corn/roots and tubers/other
	maiz_asoc	Associated corn	yes/no
	frutal	Fruit trees	yes/no
	pasto	Pastures	yes/no
	terreno_prep	Land preparation	manual/tractor/animal traction/mixed/other
	fertilizacion	Fertilization	yes/no
	tipo_fertilizante	Type of fertilizer	natural/chemical/mixed
	fumigacion	Fumigation	yes/no
	rotacion_cult	Crop rotation	yes/no
	semillas	Seed acquisition	own/purchased/exchanged
	agua_suficienc	Perception of water sufficiency	yes/no/depends on the time of year
	riego	Irrigation	yes/no
	asisten_tec	Technical assistance	yes/no
Marketing	dest_pro	Destination of agricultural production	markets and fairs in the parish/markets and fairs in Cuenca/markets and fairs in other cities/intermediaries/family consumption/supermarkets
	comercializacion	Virtual commercialization	yes/no
	asociacion	Producers’ association	yes/no
	transporte	Transportation of production	public transport/rental trucks/assistance from a producers’ association/taxi/GAD assistance/own vehicle

and Appendix A 

Table A1. Survey questions considered for the present study.

Nº	VIP
1	How do you consider yourself? Producer/Intermediary/Producer-Intermediary
2	How many people make up your household?
3	How many adults >18 years old in your household work or help in agriculture?
4	Does your household’s economic income come solely from agriculture? YES/NO
5	From what other activities do those economic incomes come? Agriculture/Sale of labor/Raising and commercializing animals/Food preparation and sale/Others
6	Do you receive remittances? YES/NO
7	What percentage of your economic income is allocated to household food?
8	Do you receive any social assistance from the Government or from another public or private institution? YES/NO
9	Do you bring products or services from Cuenca? YES/NO
10	What is your approximate monthly economic income entering the household?
11	The lands you have for cultivation are: rented/sharecropped/communal/family patrimony/owned/loaned/inherited
12	What is the main crop you produce? vegetables/fruit trees/pastures/corn/roots and tubers/other
13	What is the secondary crop you produce? vegetables/fruit trees/pastures/corn/roots and tubers/other
14	Do you plant corn in association with other crops? YES/NO
15	Do you have fruit trees? YES/NO
16	Do you plant pastures? YES/NO
17	How do you prepare your lands for planting? manual/tractor/animal traction/mixed/other
18	Do you fertilize your lands? YES/NO
19	What type of fertilizers do you use to fertilize your lands? natural/chemical/mixed
20	Do you fumigate your crops to control pests and diseases? YES/NO
21	When you plant, do you rotate crops? YES/NO
22	Where do you get the seeds to plant your crops? own/purchased/exchanged
23	Do you believe you have enough water to cover the needs of your crops? YES/NO/Depends on the time of year
24	Do you have an irrigation system for your crops? YES/NO
25	Have you received technical assistance or help from public or private institutions? YES/NO
26	What is the destination of your agricultural products? Markets and fairs in the parish/Markets and fairs in Cuenca/Markets and fairs in other cities/Intermediaries/Family consumption/Supermarkets
27	Do you sell your products through virtual means? YES/NO
28	Are you a member of a Producers’ Association? YES/NO
29	How do you transport your products to the points of sale? Public transport/Rental trucks/Taxi/Own vehicle

). This study is grounded in the foundational theory of the marketing mix, commonly conceptualized as the 4Ps framework (Product, Price, Place, and Promotion) [30]. However, our analysis focuses on a set of specific, operational marketing variables that are of critical importance within the context of our study. The assessment of marketing strategies was based on key marketing variables for short food supply chains and direct-to-consumer sales [31], such as destination of production, commercialization, association memberships, and product transportation. At the national level, these marketing dimensions are aligned with the Ecuadorian agricultural commercialization policies established by the Minister of Agriculture and Livestock of Ecuador, which aim to strengthen local markets and reduce dependency on intermediaries.

2.3. Data Analysis

To address the first study objective (to characterize production agroecosystems in suburban areas near Cuenca city in the Southern Ecuadorian Andes), descriptive analytics were applied to survey-derived data. The quantitative variables—‘tam_hogar’ (household size), ‘adul_trab’ (working adults), ‘ingreso_aliment’ (food income), and ‘Ingreso’ (total income)—were converted into categorical variables (see variable definitions in Table 1) for statistical processing. Analytical procedures included contingency table construction and chi-square tests (α = 0.05) to examine variable–district relationships.

For the second objective (to determine the principal smallholder farm typologies based on livelihood systems using an unsupervised machine learning method), we implemented an unsupervised learning approach using the PAM (Partitioning Around Medoids) clustering algorithm [32]. The analysis was conducted in R [33] using the “pam” function from the “cluster” package [34].

PAM algorithm was selected due to its particular suitability for datasets containing categorical, binary, or mixed data types, which is the dominant characteristic of our survey data (Table 1). Furthermore the use of medoids enhances the robustness of the algorithm to outliers and, critically, provides interpretable and representative ‘typical cases’ for each cluster. This is essential for constructing a meaningful and actionable typology of agroecosystems, as it allows the resulting agroecosystem types to be directly understood and related to real-world decision-making contexts, thereby aligning the methodology directly with the study’s objective. The quantitative variables ‘tam_hogar’ (household size), ‘adul_trab’ (working adults), ‘ingreso_aliment’ (food income), and ‘Ingreso’ (total income) were converted to categorical format for this analysis. Cluster generation employed Gower’s dissimilarity coefficient [35], which calculates pairwise dissimilarities between observations (e.g., i and j) relative to their nearest medoid in PAM.

The optimal number of clusters was determined based on a combination of statistical indicators using the Silhouette method [26] and interpretative relevance, balancing quantitative rigor and practical meaning. This approach evaluates clustering quality by calculating the Silhouette Index (SI), where values approaching +1 indicate well-separated, distinct clusters, values near −1 suggest poor clustering, and values around zero reflect ambiguous group assignments [36]. The optimal k corresponds to the value maximizing the average SI across potential k values—higher average silhouette widths denote superior clustering solutions. Following Alvarez et al. [37], we constrained our cluster solutions to between 3 and 7 groups, consistent with typical typology studies for capturing heterogeneity while avoiding oversimplification. To assess statistical differences between the selected clusters, we analyzed the Silhouette Index (SI) values of individual observations within each cluster using the non-parametric Kruskal–Wallis test (α = 0.05), followed by Benjamini–Hochberg post hoc procedures to control for false discovery rates. For two-dimensional cluster visualization, we employed t-distributed Stochastic Neighbor Embedding (t-SNE), a nonlinear dimensionality reduction technique [38], implemented through the Rtsne package [39]. For cluster interpretation, clusters were characterized by descriptive statistics (frequency counts) and chi-square tests (α = 0.05) to assess variable–cluster associations.

3. Results

3.1. Characterization of Agroecosystems

The characterization of Andean suburban agroecosystems was conducted at the study area level, examining three fundamental aspects: (a) socioeconomic conditions, (b) production system structure and management, and (c) agricultural output destinations.

3.1.1. Socioeconomic Characteristics of Farm Households

Table 2. Socio-economic variables (relative frequencies) of farm households by district, (n = 293), * (significant) and NS (not significant) indicate whether or not a significant relationship exists between the variables and the districts as a result of the Chi-square test (α = 0.05).

	Baños (%)	Ricaurte (%)	San Joaquín (%)	Sayausí (%)	Sinincay (%)	Total	X2	p-Value
How the person is considered
Producer	4.65	0	3.91	4.44	5.45	3.69	10.51	0.2309
Intermediary	58.14	90.91	69.53	71.11	78.18	73.58		NS
Producer-intermediary	37.21	9.09	26.56	24.44	16.36	22.73
						100.00
Number of people who make up the household
1–2	11.63	18.18	14.06	24.44	21.82	18.03	14.78	0.5405
3–4	55.81	40.91	43.75	44.44	41.82	45.35		NS
5–6	27.91	27.27	32.03	26.67	23.64	27.50
7–9	4.65	9.09	6.25	2.22	12.73	6.99
>9	0	4.55	3.91	2.22	0	2.14
						100.00
Number of adults working in agriculture
1–2	83.72	81.82	85.94	82.22	76.36	82.01	10.87	0.5402
3–4	16.28	18.18	11.72	15.56	18.18	15.98		NS
5–6	0	0	0.78	2.22	5.45	1.69
>6	0	0	1.56	0	0	0.31
						100.00
Income from agriculture
Yes	58.14	72.73	39.84	64.44	87.27	64.48	39.3	6.04 × 10−8
No	41.86	27.27	60.16	35.56	12.73	35.52		*
						100.00
Sources of economic income
Agriculture	41.86	27.27	60.16	35.56	12.73	35.52	72.1	4.27 × 10−9
Raising and commercializing animals	11.63	13.64	7.03	8.89	34.55	15.15		*
Food preparation and sale	13.95	40.91	12.5	8.89	25.45	20.34
Sale of labor	4.65	0	0.78	4.44	0	1.97
Other	27.91	18.18	19.53	42.22	27.27	27.02
						100.00
Remittances
Yes	90.7	72.73	85.94	60	76.36	77.15	18.23	0.00111
No	9.3	27.27	14.06	40	23.64	22.85		*
						100.00
Monthly income
<400 USD	58.14	27.27	63.28	44.44	54.55	49.54	29.58	0.00323
401–715 USD	0	4.55	5.47	2.22	10.91	4.63		*
>715 USD	30.23	63.64	26.56	51.11	32.73	40.85
Prefer not to answer	11.63	4.55	4.69	2.22	1.82	4.98
						100.00
Income for food
0–25%	41.86	50	67.19	48.89	45.45	50.68	19.03	0.0877
26–50%	41.86	50	25	37.78	45.45	40.02		NS
51–75%	13.95	0	7.03	11.11	7.27	7.87
76–100%	2.33	0	0.78	2.22	1.82	1.43
						100.00
Receives economic assistance from institutions
Yes	79.07	95.45	90.62	86.67	70.91	84.54	14.9	0.00492
No	20.93	4.55	9.38	13.33	29.09	15.46		*
						100.00
Products/services from Cuenca
Yes	34.88	54.55	53.91	37.78	43.64	44.95	7.272	0.1222
No	65.12	45.45	46.09	62.22	56.36	55.05		NS
						100.00

presents the percentage values of socioeconomic variables for surveyed farm households by district, which form the basis of the socioeconomic characterization described below.

At the general level, most household heads (73.6%) self-identified as intermediary farmers. Family units typically consisted of 3–4 members (45.3% of respondents), with 1–2 household members primarily engaged in agricultural work (82% of cases). While agriculture served as the main income source (35.5% of households), earnings were frequently supplemented through other activities—most notably remittances from migrant family members. This contributed to approximately 41% of households earning above $715 monthly. However, the majority (~50%) reported incomes below $400 (equivalent to Ecuador’s 2021 unified basic monthly wage). Consequently, 84.5% of households received economic assistance, primarily through the government’s Human Development Bonus program. A significant proportion (45%) relied on products and services from Cuenca city. As shown in Table 2, household economic variables (agricultural income, income sources, remittances, monthly earnings, and economic assistance receipt) demonstrated statistically significant district-level variation (Chi-square test; p < 0.05).

3.1.2. Structure and Management of the Production Systems

Table 3. Agricultural variables (relative frequencies) of farms by district, (n = 293). * (significant) and NS (not significant) indicate whether or not a significant relationship exists between the variables and the districts as a result of the Chi-square test (α = 0.05).

	Baños (%)	Ricaurte (%)	San Joaquín (%)	Sayausí (%)	Sinincay (%)	Total	X2	p-Value
Land ownership
Rented	0	0	0.78	2.22	5.45	1.69	64.452	1.43 × 10−6
Sharecropped	9.3	9.09	31.25	17.78	1.82	13.85		*
Inherited	6.98	4.55	2.34	20	5.45	7.86
Family patrimony	18.6	4.55	3.12	11.11	16.36	10.75
Loaned	9.3	9.09	10.16	0	9.09	7.53
Owned	55.81	72.73	52.34	48.89	61.82	58.32
						100.00
Main crop
Fruit trees	0	0	0	0	3.64	0.73	89.862	7.84 × 10−11
Vegetables	53.49	90.91	97.66	66.67	49.09	71.56		*
Corn	16.28	0	0.78	17.78	21.82	11.33
Pastures	11.63	4.55	0	11.11	5.45	6.55
Roots and tubers	13.95	4.55	0.78	2.22	14.55	7.21
Other	4.65	0	0.78	2.22	5.45	2.62
						100.00
Secondary crop
Fruit trees	2.33	9.09	6.25	6.67	7.27	6.32	36.459	1.36 × 10−2
Vegetables	34.88	13.64	21.88	26.67	25.45	24.50		*
Corn	27.91	22.73	23.44	37.78	43.64	31.10
Pastures	6.98	4.55	14.06	4.44	1.82	6.37
Roots and tubers	4.65	36.36	21.88	11.11	14.55	17.71
Other	23.26	13.64	12.5	13.33	7.27	14.00
						100.00
Associated corn
Yes	23.26	22.73	56.25	17.78	18.18	27.64	41.639	1.98 × 10−8
No	76.74	77.27	43.75	82.22	81.82	72.36		*
						100.00
Fruit trees
Yes	30.23	27.27	53.91	26.67	36.36	34.89	16.905	2.02 × 10−3
No	69.77	72.73	46.09	73.33	63.64	65.11		*
						100.00
Pastures
Yes	18.6	18.18	53.91	31.11	21.82	28.72	31.531	2.39 × 10−6
No	81.4	81.82	46.09	68.89	78.18	71.28		*
						100.00
Land preparation
Manual	86.05	72.73	47.66	66.67	65.45	67.71	72.527	3.59 × 10−9
Tractor	11.63	27.27	27.34	20	12.73	19.79		*
Animal traction	0	0	0.78	0	0	0.16
Mixed	0	0	22.66	2.22	1.82	5.34
Other	2.33	0	1.56	11.11	20	7.00
						100.00
Fertilization
Yes	9.3	4.55	7.81	11.11	18.18	10.19	5.4498	2.44 × 10−1
No	90.7	95.45	92.19	88.89	81.82	89.81		NS
						100.00
Type of fertilizer
Natural	2.33	13.64	38.28	11.11	5.45	14.16	51.622	7.23 × 10−7
Chemical	88.37	81.82	51.56	75.56	76.36	74.74		*
Mixed	9.3	4.55	7.81	11.11	18.18	10.19
None	0	0	2.34	2.22	0	0.91
						100.00
Fumigation
Yes	74.42	50	24.22	57.78	76.36	56.56	60.447	2.34 × 10−12
No	25.58	50	75.78	42.22	23.64	43.44		*
						100.00
Crop rotation
Yes	11.63	4.55	10.16	13.33	30.91	14.12	16.134	2.84 × 10−3
No	88.37	95.45	89.84	86.67	69.09	85.88		*
						100.00
Seed acquisition
Own	76.74	86.36	97.66	73.33	67.27	80.27	40.937	2.14 × 10−6
Purchased	2.33	0	0	0	0	0.47		*
Exchanged	20.93	13.64	2.34	26.67	32.73	19.26
						100.00
Perception of water sufficiency
Yes	13.95	13.64	10.94	15.56	21.82	15.18	100.35	2.20 × 10−16
No	69.77	27.27	5.47	42.22	50.91	39.13		*
Depends on the time of year	16.28	59.09	83.59	42.22	27.27	45.69
						100.00
Irrigation
Yes	74.42	13.64	21.88	46.67	63.64	44.05	58.417	6.24 × 10−12
No	25.58	86.36	78.12	53.33	36.36	55.95		*
						100.00
Technical assistance
Yes	30.23	59.09	71.88	35.56	60	51.35	32.809	1.31 × 10−6
No	69.77	40.91	28.12	64.44	40	48.65		*
						100.00

presents the percentage values for the variables related to the structure and management of the production systems.

The farmers are, for the most part (58.3% of those surveyed), the owners of the land used for their agricultural activities. The main crops are vegetables (71.6%), with corn as a secondary crop. The production systems are characterized by being based on manual tillage for soil preparation, fumigation of crops for pest and disease control, and without fertilization; they do not practice crop rotation. The seeds used by the farmers are typically their own. Approximately, 56% of farms lack water for their crops, and approximately 48% of respondents state that water availability is dependent on the time of year. Approximately 51.4% report that they have not received any form of technical assistance or support from public or private institutions for production.

As can be observed in Table 3, all variables, except for “Fertilization,” show a significant relationship (Chi-square test; p < 0.05, Table 3) with the districts, which would indicate a high diversity in the aspects of structure and management of the production systems among the districts.

3.1.3. Destination of Agricultural Production

The majority of respondents (approximately 75%) stated that they sell their products, marketing them at parish and city of Cuenca markets and fairs (

Table 4. Marketing variables (relative frequencies) of farms by district, (n = 293). * (significant) indicates a significant relationship between the variables and the districts, based on the Chi-square test (α = 0.05).

	Baños (%)	Ricaurte (%)	San Joaquín (%)	Sayausí (%)	Sinincay (%)	Total	X2	p-Value
Destination of agricultural production
Family consumption	30.23	13.64	15.62	13.33	30.91	20.746	84.85	1.01 × 10−8
Family consumption and markets	0	0	0	2.22	0	0.444		*
Intermediaries	0	0	10.16	0	5.45	3.122
Markets and fairs in Cuenca	46.51	0	54.69	35.56	20	31.352
Markets and fairs in the parish	23.26	86.36	17.19	46.67	43.64	43.424
Markets and fairs in other cities	0	0	1.56	2.22	0	0.756
Supermarkets	0	0	0.78	0	0	0.156
						100
Virtual commercialization
Yes	69.77	72.73	89.84	82.22	83.64	79.64	11.4	2.24 × 10−2
No	30.23	27.27	10.16	17.78	16.36	20.36		*
						100
Producers’ association
Yes	0	90.91	68.75	11.11	43.64	42.882	103	2.20 × 10−16
No	100	9.09	31.25	88.89	56.36	57.118		*
						100
Transportation of production
Rental trucks	48.84	40.91	39.84	46.67	20	39.252	26.44	9.31 × 10−3
Taxi	4.65	0	1.56	4.44	0	2.13		*
Public transport	20.93	9.09	11.72	22.22	27.27	18.246
Own vehicle	25.58	50	46.88	26.67	52.73	40.372
						100

). Because this study was conducted during the COVID-19 pandemic, products were also marketed virtually according to the majority of respondents (79.6%). Transportation of production is primarily carried out using their own vehicles (40.4% of respondents) and by rental pickup trucks (39.3% of respondents). Some 57.1% of producers report being members of a producers’ association, and the advantages of this membership are primarily having more opportunities to sell their products and also receiving technical assistance. All variables corresponding to the aspect discussed in this section (“Destination of agricultural production”) show a significant relationship (Chi-square test; p < 0.05, Table 4) with the districts, which would also indicate that these variables are responsible for the diversity of the production systems among the districts.

3.2. Agroecosystems Typologies

The determination of the optimal number of typologies balanced statistical metrics with establishing methodological conventions. The average Silhouette width [26] indicated that a two-cluster solution provided the statistically optimal partition of the data (Figure 2). However, following established practices in typological development, where the number of clusters commonly ranges from three to seven to ensure sufficient discrimination without excessive complexity [37], a three-cluster solution was selected for this study. This decision was substantively driven by the finding that the three-cluster model, while resulting in a marginally lower average Silhouette width, yielded a more nuanced and theoretically meaningful segmentation. Three clusters captured typologies that were conceptually distinct and critical to the research objectives. Retaining only two clusters would have consolidated differences, resulting in an oversimplification that masked important heterogeneity within the data. Consequently, the three-cluster solution was adopted for its superior interpretative utility and alignment with typological standards, despite the marginal statistical trade-off.

The typologies generated based on the three clusters can be visualized in Figure 3. Considering the SI values of the observations belonging to each cluster and comparing them between clusters, statistical differences were evidenced between the clusters (Kruskal–Wallis, p < 0.01054). Subsequently, using the BH procedure as a post hoc analysis, differences were detected between cluster 1 and clusters 2 and 3 (

Table 5. Mean and standard deviation values of the Silhouette Index (SI) for each cluster. Statistical differences (α = 0.05) were determined by the Kruskal–Wallis test (K-W) with the Benjamini–Hochberg procedure (BH) as a post hoc analysis (different letters indicate significant differences between clusters).

Cluster	Mean SI	p-Value (K-S)	Differences (BH)
1	0.108 ± 0.091	0.01054	a
2	0.074 ± 0.105		b
3	0.069 ± 0.097		b

). Although statistically similar, clusters 2 and 3 showed a difference in their mean SI value; thus, the minimum number of typologies is three.

The characteristics of each are described in

Table 6. Typologies generated based on PAM (Partitioning Around Medoids) clustering. * (significant) and NS (not significant) indicate whether or not a significant relationship exists between the variables and the clusters as a result of the Chi-square test (α = 0.05).

Characteristics		Typologies			p-Value (Chi-Square Test)	Significance
		Cluster 1	Cluster 2	Cluster 3
Socio-economics	How the person is considered	Intermediary	Intermediary	Intermediary	0.5308	NS
	People who make up the household	3–4	3–4	5–6	0.0005469	*
	Adults who work in agriculture	1–2	1–2	1–2	0.576	NS
	Income from agriculture	Yes	No	Yes	1.179 × 10−15	*
	Sources of economic income	Agriculture	Agriculture	Others	2.003 × 10−14	*
	Remittances	Yes	Yes	Yes	0.004309	*
	Monthly income	>715	<400	>715	2.77 × 10−12	*
	Income for food	0–25%	0–25%	0–25%	0.05395	NS
	Receives economic assistance	No	No	No	0.1981	NS
	Products/services from Cuenca	Yes	Yes	No	0.7788	NS
Agricultural aspects	Land ownership	Own	Own	Own	0.05825	NS
	Main crop	Vegetables	Vegetables	Vegetables	0.009639	*
	Secondary crop	Corn	Corn	Roots and tubers	6.039 × 10−6	*
	Associated corn	No	Yes	Yes	4.955 × 10−13	*
	Fruit trees	No	Yes	No	6.531 × 10−6	*
	Pastures	No	Yes	No	2.2 × 10−16	*
	Land preparation	Manual	Manual	Tractor	2.2 × 10−16	*
	Fertilization	No	No	No	0.00408	*
	Type of fertilizer	Chemical	Chemical	Natural	2.2 × 10−16	*
	Fumigation	Yes	Yes	No	1.496 × 10−7	*
	Crop rotation	No	No	No	0.006687	*
	Seed acquisition	Own	Own	Own	0.5232	NS
	Water sufficiency	Depends on the time of year	Depends on the time of year	Depends on the time of year	0.00004217	*
	Irrigation	No	Yes	No	2.462 × 10−8	*
	Technical assistance	No	Yes	Yes	1.147 × 10−14	*
Marketing	Destination of agricultural production	Markets and fairs in the parish	Markets and fairs in Cuenca	Markets and fairs in Cuenca	0.00006811	*
	Virtual commercialization	Yes	Yes	Yes	0.00699	*
	Producers’ association	No	Yes	No	9.502 × 10−10	*
	Transportation of production	Rental trucks	Rental trucks	Own vehicle	1.244 × 10−11	*

, taking into account only the values of the highest frequencies found for each of the variables (socioeconomic, agricultural aspect, and marketing variables) within each cluster. Although this approach has its limitations by considering only the highest frequencies, it has the advantage of synthesizing the most representative information for each typology (the values for all response frequencies of the variables can be found as Supplementary Material: Tables S1–S3). However, the statistical relationship between the variables and the clusters was also evaluated using the Chi-square test (α = 0.05), which evidenced that the agricultural and marketing aspect variables are the most influential in determining the clusters (17 of the 19 variables show a significant relationship with the clusters, Table 6). A description of each cluster is provided below.

3.2.1. Typology 1

Cluster 1 corresponds to a typology of farmers with a high economic level, with incomes exceeding 715 USD (the unified basic monthly wage (SBU) in Ecuador for 2021 was 400 USD). This income originates predominantly from their activities as intermediary farmers; this cluster presented the highest percentage of intermediary farmers (approximately 74% of farmers). They allocate up to approximately 25% of their income to food. They are characterized by owning their own land (land ownership is a common characteristic across all three typologies), where vegetables and corn are the dominant crops. Their production systems are based on practices that are predominantly reduced in the use of fertilizers (approximately 94% of its farmers), although a low percentage (approximately 6%) use chemical fertilizers. Some 55% of farmers resort to fumigation for pest and disease control in their crops. They prepare the land manually, practice crop rotation, the majority (approximately 65%) lack water for irrigation, and also the majority (approximately 68%) do not receive technical advice from institutions, whether governmental or non-governmental. The destination of production is sale at parish and city markets and fairs, and for transporting production they use rental pickup trucks. In this typology, the majority of farmers (approximately 70%) do not belong to producer associations.

3.2.2. Typology 2

Cluster 2 corresponds to a typology of producers with the lowest incomes among the clusters (Table 6). They earn an income below the Ecuadorian monthly SBU (<400 USD) for the sustenance of a typical household of 3 to 4 members. This income does not originate from agricultural activities (74.29%), and it is the cluster with the highest percentage of farmers who sell their labor—particularly to other cultivated fields, as the majority of their income comes from agriculture and they also act as intermediaries (Table 6). They allocate up to 25% of their income to food, and since their income is below the monthly SBU, this affects the quality of life of households in this typology, particularly in terms of nutrition.

Their cultivated land is their own, and they maintain a greater diversity of crops (predominantly cultivating vegetables, corn associated with legumes—commonly with beans) than farmers in Typology 1 (which lacks fruit trees) and Typology 3 (which lacks both fruit trees and pasture)—see Table S2. Their land management is similar to that of Typology 1, meaning they prepare their land manually, rotate crops, do not use fertilizers, and use pesticides. Unlike Typology 1, the majority of farmers in this typology have irrigation and receive technical assistance.

The destination of their production is the markets and fairs of the city of Cuenca; they belong to producer associations and use rental pickup trucks for product transport. Economically, this typology is the least sustainable due to having the lowest income, which results in a lower quality of life.

3.2.3. Typology 3

Cluster 3 corresponds to a typology of farmers who earn incomes above the unified basic wage (>715 USD), but it is a typology with the highest number of household members (5 to 6 people, approximately 51%). Although agriculture remains an income source for this cluster, it can be observed (Table 6) that there are other main sources of economic income, which, being equally high (>715 USD), can be attributed to money coming particularly from remittances.

Regarding income allocated to food, receiving financial aid, and bringing products and services from the city, and land tenure, their trends are similar to the other typologies. On their land, crop diversity is lower than the other typologies since they predominantly only grow vegetables (they do not sow pastures nor have fruit trees, as is the case with the other typologies). The majority prepare their soil with a tractor (approximately 66%). They do not use fertilizers (approximately 95%), and those who do use natural fertilizers; also, the majority (85%) do not fumigate their crops. These aspects would indicate that these are production systems characterized by the lowest use of agrochemicals, making it the most environmentally friendly typology.

Similarly to the other typologies, they use their own seeds and do not rotate crops, but the majority have irrigation (85%) and also receive technical assistance (approximately 78%). The destination of their production is the markets and fairs of Cuenca, transporting their production in their own vehicles, and the majority are also not associated (60%).

4. Discussion

4.1. Socioeconomic Heterogeneity and Livelihood Diversification

The characterization of Andean suburban agroecosystems in southern Ecuador revealed a notable heterogeneity in both the socioeconomic aspects of households and the structure and management of production systems, as well as in the destination of agricultural production. This heterogeneity successfully fulfills the first specific objective of the study, which aimed to describe the conditions of small farmers in suburban areas near the city of Cuenca. A key finding that emerged from this characterization is that, despite of its variability, agricultural activity is often an insufficient sole source of income for these households. Many farming families complement agricultural earnings with off-farm employment and remittances from relatives living abroad, a pattern that reflects both the vulnerability and adaptability of these agroecosystems. Similar livelihoods diversification strategies have been reported in other Andean territories by Toledo et al. [40] and Haro et al. [41], where migration and wage labor function as mechanisms to cope with unstable agricultural incomes and increasing production costs. This dependence on external income sources highlight the fragility of local agricultural economies and the urgent need for strategies that enhance the profitability and resilience of small-scale farming.

4.2. Production Systems and Environmental Sustainability

From a production standpoint, vegetables dominate the cropping systems, which suggests a strong integration with urban markets. Nevertheless the widespread use of synthetic fertilizers and pesticides, combined with limited crop rotation, raises concern regarding the long-term environmental sustainability of these agroecosystems. This result coincides with previous studies in Andean agroecosystems that warn of the increasing pressure towards more input-intensive agriculture, even on small scales [42,43,44,45,46]. The intensive management practices observed may compromise soil health, reduce biodiversity, and increase production risks associated with pest resistance and input price volatility. Despite the evident short-term economic benefits, such reliance on chemical inputs reflects a path dependency that is difficult to reverse without targeted interventions. Therefore, promoting ecologically based management practices, such as diversified rotations, integrated pest management, and the use of organic amendments, should be prioritized to reduce environmental pressures while maintaining productivity [47]. Strengthening technical assistance programs, access to sustainable inputs, and farmer-to-farmer training could help transition suburban agriculture toward more sustainable and resilient production systems.

It is important to acknowledge that the results presented here capture a single temporal snapshot of suburban agroecosystems, corresponding to the period of fieldwork conducted during the second half of 2021. Consequently, seasonality variability in crop management and income composition was not fully captured. In the Andean highlands, marked climatic seasonality strongly influences crop calendars, labor allocation, and market prices, leading to fluctuations that can substantially modify farmers’ decisions through the year [48]. Likewise, long-term processes—such as demographic changes, migration patterns, and shifts in agricultural policies—can reshape the socioeconomic conditions and sustainability trajectories of suburban households over time. Future studies should therefore incorporate longitudinal and multi-seasonal approaches that enable the identification of dynamic changes in production practices and livelihood strategies.

4.3. Theoretical Framing of Typologies: Livelihood Strategies and Sustainability Transitions

The pursuit of our second objective—determining farmer typologies using data mining techniques– revealed that the observed heterogeneity is not random but can meaningfully categorized into three distinct typologies. These typologies represent distinct livelihood strategies shaped by differential access to resources, market integration, and adaptation to the suburban context.

The significant heterogeneity of agroecosystems has also been highlighted in similar studies, such as the one conducted in southern Ecuador by Vanegas et al. [21]. Our findings can be further interpreted through the lens of sustainability transition theory [49,50], which provides a dynamic perspective on how these different farm types interact within a broader system. In this framework: Typology 2 represents the most vulnerable group, whose practices are often marginalized by the dominant socio-economic “landscape”. Their low incomes and high vulnerability highlight systemic pressures and exclusions. Typology 1 embodies the dominant “socio-technical regime” of commercial, input-intensive agriculture that is well integrated into existing market structures but faces growing environmental sustainability challenges. Typology 3 can be viewed as an emerging “niche” characterized by practices that align more closely with sustainability principles (e.g., lower agrochemical use), potentially supported by alternative income sources like remittances. This framing moves beyond a static classification and helps identify leverage points for policy. The goal becomes not just to support each type in isolation, but to foster interactions that can empower niche innovations (Typology 3), reshape the prevailing regime (Typology 1) towards greater sustainability, and create protective buffers for the most vulnerable (Typology 2).

4.4. Concrete Policy Implications for Differentiated Interventions

The typologies identified in this study provide a valuable empirical foundation for designing differentiated policies that move beyond one-size-fits-all approaches. The following concrete, actionable recommendations are linked directly to the specific profiles of each typology:

For Typology 2 (vulnerable, low-income households): Policy should focus on poverty alleviation and risk reduction. This includes facilitating access to social safety nets, microcredit for basic inputs, and programs that promote food security and subsistence production. Training in low-cost, sustainable practices (e.g., homemade organic pesticides) and support for forming commercial alliances could help improve their market position and resilience.

For Typology 1 (commercial, input-intensive farms): Interventions should aim at “sustainable intensification”. This group would benefit from technical assistance and economic incentives for adopting integrated pest management and precision fertilization to reduce environmental externalities without sacrificing yields. Policies that strengthen their access to stable, value-added markets (e.g., through contracts or branding) and provide green subsidies for environmentally friendly technologies are key.

For Typology 3 (lower-input, remittance-supported households): This group represents a potential model for transition. Policy should aim to “empower the niche” by providing support for organic or agroecological certification, creating market linkages for sustainably produced goods, and offering training in advanced agroecological techniques like crop diversification and agroforestry, as shown by Córdova et al. [51], such strategies can reduce dependence on off-farm activities and income. This could help solidify their environmentally friendly practices and enhance their economic viability, making them a demonstration model for others.

For all typologies, the high dependence on remittances underscores the need for policies that strengthen the autonomy and profitability of the agricultural sector itself. National initiatives of the Ecuadorian government such as the “Plan Nacional para el Buen Vivir” and the “Estrategia Nacional para la Soberanía Alimentaria” emphasize the dual need to improve rural livelihoods while safeguarding environmental integrity, they could be actively implemented in these suburban contexts, which often occupy an ambiguous policy space between urban and rural frameworks [52,53].

The patterns identified in this study are closely aligned with broader debates on agricultural policy and territorial development in Ecuador. Our findings suggest that suburban agriculture in Cuenca occupies an ambiguous policy space, where it is neither fully integrated into urban planning frameworks nor adequately supported by rural development programs. The typologies identified provide a practical diagnostic tool that can inform local governments, development agencies, and non-governmental organizations in designing evidence-based interventions.

Future research should build upon this foundation by integrating longitudinal monitoring and participatory approaches. The inclusion of time-series data or repeated surveys would help to assess the stability of the typologies identified and to determine whether households move between types in response to climatic shocks, market opportunities, or policy interventions. Such an approach would also strengthen the capacity of typology-based research to serve as an early warning tool for assessing the resilience and adaptability of suburban agroecosystems under conditions of environmental and socioeconomic uncertainty. All this work will be essential to foster transitions toward more equitable and sustainable suburban food systems in Ecuador and across the broader Andean region.

5. Conclusions

This study characterized the heterogeneity of Andean suburban agroecosystems in southern Ecuador and identified three distinct smallholder farm typologies using an unsupervised clustering approach (PAM with Gower’s distance). The typologies reveal that suburban agriculture at the rural–urban interface is not a monolithic entity but a complex landscape where households employ diverse livelihood strategies. These range from economically vulnerable, subsistence-oriented systems to more commercially integrated and environmental conscious operations. A critical cross-cutting finding is the heavy reliance on remittances and off-farm income, underscoring both the economic vulnerability and the adaptive capacity of these agroecosystems in the face of limited agricultural profitability.

From a conceptual perspective, this research makes a significant contribution by demonstrating that farm typologies, when interpreted through frameworks like sustainability transition theory, provide a powerful lens for understanding the dynamics of agricultural systems. The identified typologies—representing vulnerable, regime, and niche profiles—offer a structured way to analyze interactions, constrains, and potential pathways for change within the suburban agricultural landscape.

From a practical and policy standpoint, the findings provide a robust, empirical basis for designing differentiated interventions. The one-size-fits-all approach is demonstrably inadequate for the diverse realities captured in this study. Instead, policymakers and local governments can use this typology to tailor support: (i) targeting economic strengthening and safety nets for vulnerable households (Typology 2); (ii) promoting sustainable intensification and market integration for commercial farmers (Typology 1), and (iii) empowering and scaling agroecological practices among the more environmentally sustainable producers (Typology 3).

The novelty of this research lies in the integration of machine learning techniques with territorial analysis in a suburban Andean context—a region where empirical typologies are scarce despite the growing importance of peri-urban agriculture for urban food security and territorial resilience.

Finally, we acknowledge the limitations of a cross-sectional design and the conversion of some quantitative variables into categorical data. Future research should employ clustering on mixed data, with continuous variables retained, to test sensitivity. Furthermore, longitudinal and multi-seasonal are needed to explore how household typologies evolve in response to environmental change and policy changes. Socializing and validating these typologies with the producers themselves will be a critical next step. Undoubtedly, this will be a task for future work, which can continue to use the findings of this study as a starting point. Such efforts are essential to advance our understanding of these dynamic agroecosystems and to support their transition toward greater sustainability, equity, and resilience in the face of rapid urbanization in the Andean region.