Designing resilient farming systems for a turbulent world: learning from communities at the frontline

In today’s increasingly turbulent world, farming systems are confronted by unpredictable and compounding shocks such as extreme weather events, pest outbreaks, market volatility, and socio-political conflicts that collectively redefine the conditions under which they operate^1,2. These disruptions often exceed the adaptive capacity of existing systems, thereby exposing structural weaknesses. The COVID-19 pandemic, for example, revealed the fragility of global food systems by disrupting supply chains, markets, and livelihoods worldwide³. Such experiences collectively underscore the urgent need for farming systems that can not only withstand shocks but also adapt and evolve in response to them^4,5.

Considering these growing uncertainties, resilience as a concept has thus gained prominence among policymakers, sustainability scientists, and development practitioners⁶. Unlike the contemporary efficiency-driven approaches that prioritize short-term productivity and minimize redundancy, resilience emphasizes the long-term functionality of systems through the capacity to absorb disturbances, adapt to change, and transform under uncertainty^7,8,9. This shift in thinking reflects a move from optimization toward flexibility, redundancy, and adaptive learning—attributes essential for sustaining agricultural systems under volatile conditions.

Resilience, however, is not a uniform property. It operates across multiple levels and dimensions that together determine how systems respond to stress. At the individual level, resilience reflects personal traits such as optimism and coping strategies, as well as external support networks^10,11,12. At the collective level, it emerges from shared resources, governance mechanisms, and a collective identity that fosters mutual support^13,14. Moreover, resilience can be specific, addressing known disturbances such as drought or pest outbreaks, or general, encompassing the ability to respond to unforeseen and complex challenges¹⁵.

To better understand these dynamics, the social–ecological systems (SES) framework provides a useful conceptual foundation. It defines resilience through three interrelated capacities: (i) absorptive capacity, the ability to absorb disturbances and reorganize while maintaining core functions and feedbacks; (ii) adaptive capacity, the ability of actors to influence resilience within existing system configurations; and (iii) transformative capacity, the ability to create fundamentally new systems when current ones become untenable^15,16,17. More recent refinements extend this view by framing resilience as the capacity to ensure the continued provision of system functions amid increasingly complex economic, social, environmental, and institutional stresses, through robustness, adaptability, and transformability^9,18.

While the Social-Ecological Systems (SES) framework provides a strong theoretical foundation, translating its concepts into measurable indicators remains challenging. This challenge is compounded by the inherently latent nature of resilience in farming systems, which can only be observed retrospectively after one or multiple shocks have occurred (e.g., ref. ¹⁹). Such retrospective assessment limits the ability to predict a system’s capacity to absorb, adapt, and transform in response to unforeseen disturbances²⁰. To overcome this limitation⁹, proposed an integrated framework that combines ex post and ex ante assessments of resilience. This approach emphasizes both the attributes of resilience and the enabling—or constraining—conditions that strengthen individual and collective competences within food systems. These competences enhance one or more of the three resilience capacities—robustness, adaptability, and transformability—and, in turn, contribute to overall system resilience^3,21. further categorized the resilience attributes of food systems into four dimensions, collectively known as ABCD: Agency, the capacity of actors to make decisions and act; Buffering, the ability to absorb disturbances while minimizing damage; Connectivity, the physical and virtual linkages among system components; and Diversity, the variety of components and processes that promote adaptability and flexibility.

These ABCD attributes are interdependent and operate collectively to shape the resilience capacity of farming systems^{7,9,22,23,24,25}. The resilience capacities of farming systems are further enhanced or constrained by the enabling environment (E), which encompasses policies, regulations, and institutional actions at the public, private, or local levels²⁶. In the African smallholder contexts, the enabling environment plays a crucial role^27,28,29. For instance, public investments in rural roads enhance connectivity, while policy support for credit access and warehouse development strengthens the buffering capacity.

Recent literature indicates that the manifestation and relative importance of the ABCD (and E) attributes are not static; rather, they vary across time, space, and scale^20,30,31. For example, study³⁰ demonstrated that transitions in European farming systems toward alternative models alter the relative contribution of resilience attributes, underscoring the dynamic nature of resilience under changing conditions.

Despite these significant advancements in resilience science, two critical knowledge gaps remain. First, it is still unclear which resilience attributes are most pivotal to the collective resilience of smallholder farming systems. Second, the ways in which the contributions of these attributes change under multiple, concurrent disruptions are not well understood. Addressing these gaps is essential for designing farming systems that can better withstand shocks, particularly in regions dominated by smallholder agriculture.

To address these gaps, we examined the resilience of communities that have experienced diverse abiotic, biotic, and socio-political shocks in fragile environments that are vulnerable to land degradation. Globally, land degradation—characterized by the depletion of soil, water, and vegetation resources—is a serious challenge for farming systems that threatens land productivity and, consequently, food security^32,33. Human-induced degradation affects more than 1.6 billion hectares of land worldwide, directly impacting the livelihoods of roughly 3.2 billion people³⁴. In Ethiopia, the problem is particularly acute, posing a severe threat to both rural livelihoods and the natural environment^35,36,37.

Recognizing the magnitude of this challenge, the Ethiopian government has launched large-scale land restoration initiatives, including the national Sustainable Land Management Programme (SLMP)^38,39, a national flagship initiative implemented across more than 3000 community watersheds, each covering approximately 500 to 1000 ha (40, http://nrdsmis.moa.gov.et/app/landing). Through this program, the government has provided policy, technical, and financial support to communities for implementing SLM measures that improve key land functions—such as soil health, water retention, and biomass productivity—which are essential for ecosystem recovery and agricultural resilience^40,41,42,43.

We studied the resilience of communities participating in the SLMP. While these communities received similar programmatic support, their restoration outcomes have varied considerably. After nearly a decade of implementation (2012–2021), our initial assessment revealed stark contrasts in the restoration of land resources, as measured by the proportion of watershed area under sustainable land management⁴⁴. While some communities achieved remarkable restoration success, others lagged. Our previous study identified that highly successful communities—often referred to as bright spots⁴⁵—were characterized by strong self-organization, visionary local leadership, and high community participation in planning and implementation. Yet, how these communities respond to agricultural shocks—and how the disruptions that occurred during program implementation have shaped restoration performance—remains unclear.

These disruptions may have influenced restoration outcomes by diverting resources toward immediate needs or depleting them altogether. Conversely, such challenges might also have stimulated innovation, prompting communities to adopt adaptive strategies and strengthen their commitment to sustainable practices^46,47. Figure 1 illustrates the conceptual framework describing the dynamic interplay between land management interventions, disruptions, and restoration outcomes.

Against this backdrop, this study sought to: (i) explore the trajectories of the land restoration outcomes among communities facing multiple shocks; (ii) examine the relative importance of the ABCD resilience attributes of the farming system in coping with the impacts of multiple shocks; and (iii) identify key attributes required for the general resilience of smallholder farming systems.

Trend of land restoration outcomes

Trend analyses of soil moisture content, green soil cover fraction, and above-ground biomass at the study sites revealed divergent responses to biotic, abiotic, and socio-political shocks across the watersheds (Fig. 2). During the severe 2015 drought, soil moisture (Fig. 2E) and green soil cover fraction (Fig. 2C) declined, whereas above-ground biomass remained relatively stable (Fig. 2A). Extending this comparison across agroclimatic zones further revealed pronounced differences in community resilience under multiple, overlapping shocks.

In the semi-arid Tigray region, which experienced compounded disturbances—including the 2015 drought, desert locust invasions (2019–2021), the COVID-19 pandemic (2020–2022), and civil war (2020–2022)—the resilience capacity varied substantially among performance groups (Fig. 2A, C, E). Both the high and low performance groups exhibited a declining trend in soil moisture and green soil cover fraction during the short-term drought of 2015. However, high-performing communities subsequently demonstrated sustained recovery with continuous improvement in soil moisture content, green soil cover fraction, and above-ground biomass between 2019 and 2022, despite these concurrent shocks. In contrast, these indicators stagnated or declined in low-performing communities.

In the humid southern regions, where major disturbances were largely absent except for the COVID-19 outbreak in late 2020, soil moisture content and the fraction of green soil cover remained relatively stable across both performance groups (Fig. 2B, D, F). However, above-ground biomass increased steadily in high-performing communities during the COVID-19 outbreak (2020–2022), whereas low-performing communities experienced a consistent decline over the same period.

Resilience attributes of land restoration performance groups to multiple shocks

The analysis results of the FGDs (Fig. 3) revealed that communities assigned varying levels of importance to resilience attributes depending on the type of disturbance, the level of land restoration performance, and the agroclimatic context. Across both the semiarid and humid regions, the high-performance groups in both agroclimatic regions consistently emphasized the importance of the following attributes: (1) agency attributes, demonstrated through community leadership and self-organization; (2) buffer attributes, such as access to communal resources (e.g., fodder from protected areas and community seed banks) formal and informal safety nets (e.g., access to cash/food for work, local credits, and lending), food and feed preserves, and access to savings; and (3) connectivity attributes, including access to communication infrastructure (e.g., phone lines). While the low-performing communities also emphasized the role of communication infrastructure, they did not emphasize agency or buffer attributes, and instead emphasized the connectivity attribute of remittance flows, as a key mechanism for coping with the impacts of the COVID-19 pandemic (Fig. 3A, B).

During droughts and desert locust invasions—shocks specific to the semiarid region—both high- and low-performing groups reported medium to very high contributions of buffer, connectivity, and diversity attributes (Fig. 3C, D). During the civil war (Fig. 3E), however, the communities indicated little or no contribution from connectivity and diversity attributes. High-performing communities attributed medium importance to access to information, crop diversification, and alternative household energy, and high importance to integrated pest and soil fertility management in coping with the impacts of the conflict. Low-performing communities, however, perceived these same attributes as having minimal or no contribution.

Across all disturbance types—a combination of severe drought, desert locust invasions, civil war, and the COVID-19 pandemic (Fig. 3F)—clear differences emerged between the performance groups, in which the high-performing groups consistently recognized agency as having a high to very high contribution to community resilience. Furthermore, these groups identified a wide range of buffering attributes as important in managing multiple stresses, whereas the low-performing communities tended to rely more narrowly on specific buffering and connectivity attributes to cope with these complex challenges.

Unlike earlier discourses that characterized resilience attributes as stable and systemic^7,9,22,48, the findings of this study demonstrate that the relative importance of community resilience attributes is inherently context-specific and dynamic. Specifically, their contributions vary with the type and intensity of shocks, as well as with the prevailing socio-ecological conditions. For instance, access to road networks emerged as a positive and enabling factor during droughts and desert locust invasions, as it facilitated government and development support to mitigate the impacts of shocks. By contrast, the same attribute contributed little—or even negatively—during periods of civil conflict, when mobility and access were constrained. Taken together, these findings underscore the need for a more nuanced, context-sensitive understanding of how resilience attributes interact dynamically, enabling more targeted and anticipatory resilience-building strategies.

Building on this insight, Fig. 4 synthesizes the key attributes that enable communities to cope with biotic, abiotic, and socio-political shocks. Importantly, these insights offer practical guidance for anticipating and preparing for future shocks, even when their timing or nature remains uncertain. The results indicate that community leadership, locally available resources, formal and informal safety nets, food and feed buffers, and access to information are consistently critical across all types of shocks. Additionally, attributes such as access to technical advice, ambition for change, self-organization, water conservation practices, communal labor, and labor-saving crops play a complementary yet significant role in strengthening overall community resilience.

Amid increasing environmental and sociopolitical uncertainty, strengthening the capacity of smallholder farming systems to cope with, adapt to, and, where necessary, transform in response to change has become a central priority on agricultural development agendas. This study demonstrates that multiple, compounding shocks—spanning drought, desert locust invasions, COVID-19, and armed conflicts—place substantial pressure on smallholder systems, thereby challenging communities’ land restoration trajectories. For example, desert locust invasions and drought have reduced green cover and biomass⁴⁹, while the COVID-19 pandemic and armed conflict have disrupted labor, markets, and restoration efforts^24,50.

Despite these compounding disruptions, the land restoration bright spots demonstrated remarkable resilience, exhibiting consistent improvements in aboveground biomass, green soil cover fraction, and soil moisture. These sustained gains signify increased soil carbon accumulation and the restoration of soil health within farming systems^51,52—both of which are critical for maintaining and enhancing land productivity.

These unprecedented trends are likely attributable to strong local agency, manifested in effective leadership, the capacity to mobilize resources, clearly defined governance rules, local innovation, and the use of indigenous knowledge⁴⁴. Together, these attributes foster resource conservation and efficient use, enable the adoption of transformative yet pragmatic local solutions, and support timely responses to shocks⁵³. In turn, such conditions motivate communities to invest sustainably in land management and to strengthen their resilience capacities by building buffers, reinforcing connectivity, and diversifying livelihood strategies^{6,54,55,56,57}.

Consistent with our findings, recent studies^58,59 on the impacts of war in Tigray also highlight the link between the resilience of land systems and the stewardship and institutional capacity of local communities in land management. Qualitative insights from focus group discussions further reinforce this evidence. Participants emphasized that sustained investments in land, water, and vegetation conservation—combined with continuous local learning and innovations—have significantly enhanced adaptive capacity. These capacities, in turn, enhanced productive use of the water and land resources and enabled the achievement of multiple agricultural production cycles annually, even during periods of drought and armed conflict.

Within this context, land restoration “bright spot” communities tend to perceive disturbances not merely as threats, but as opportunities for innovation. For example, participants from the Tigray region described how floods—once viewed as destructive—have been transformed into a productive resource in the aftermath of drought, as communities have learned to store and utilize floodwater more effectively.

These insights underscore the interrelated and mutually reinforcing nature of agency, buffers, connectivity, and diversification as key resilience attributes. Moreover, the observed variation in the importance of these attributes across different shock types and intensities (Fig. 3A–F) highlights the need for farming systems—particularly in low-performing communities—to invest simultaneously in strengthening agency, buffering capacity, connectivity, and diversification in order to better cope with multiple and unpredictable shocks.

Extending these insights, strengthening the ex-ante and ex-post resilience of smallholders to cope with unknown disturbances requires both proactive and reactive adaptation strategies. As evidenced in this study and in the broader organizational resilience literature^60,61, proactive measures—such as water harvesting, flood protection, savings mechanisms, and access to timely information—enhance preparedness and enable early responses. Complementarily, reactive measures, including social safety nets, labor mobilization, remittances, and cash reserves, support adaptation and transformative responses during periods of crisis.

However, the extent to which these “resilience attributes” are specific to farming communities in the Ethiopian Highlands—or can be generalized to smallholder farming systems with similar socio-ecological contexts—requires further investigation. In this regard, Wageningen University & Research is currently undertaking a large-scale study involving “bright spots” across the Global South that are part of the Global Network of Lighthouse Farms⁶².

In summary, our findings indicated that:

• The trajectories of land restoration outcomes among smallholder farming communities vary across locations, shaped by their collective capacity for resilience in response to the cumulative effects of multiple, compounding shocks.

• Agency and buffers are critically important for the collective resilience of smallholder farming communities facing multiple compounding shocks. These attributes are mutually reinforcing buffers create time and space for agency to mobilize resources and implement adaptive or transformative strategies.

• The roles of connectivity and diversity are context-specific. Improved connectivity facilitates rapid responses to crises, particularly drought and desert locust infestations, while diversification enhances long-term adaptive capacity. However, connectivity—particularly through access roads—may also increase vulnerability during periods of civil conflict, as roads can facilitate the movement of hostile groups that threaten established assets.

These findings underscore the importance of designing land restoration initiatives that extend beyond purely biophysical interventions and deliberately strengthen multiple dimensions of smallholders’ resilience attributes, including agency, buffers, connectivity, and diversity. Achieving this requires targeted policy and institutional support that fosters local innovation, empowers communities, and expands access to essential resources, knowledge, and infrastructure. By systematically embedding these capacities within restoration initiatives, such programs can enhance both the resilience and sustainability of smallholder farming systems, thereby ensuring long-term adaptive capacity in fragile socio-ecological contexts.

Study approaches

In this study, we employed a transdisciplinary research framework that integrates Earth Observation (EO) data with participatory community engagement to investigate restoration trajectories and resilience dynamics. Building on this framework, we analyzed independent EO datasets to quantify selected key performance indicators (KPIs) of land restoration, while simultaneously using participatory approaches—particularly focus group discussions with watershed members—to elicit experiential and context-specific insights.

We engaged community members as active research partners rather than passive respondents throughout the study, contributing to the identification and prioritization of explanatory variables associated with the ABCD resilience attributes. Such active engagement of community members and the embedding of their knowledge and lived experiences into the analytical framework enhanced both the contextual robustness and empirical validity of the study.

This integrative approach enabled a comprehensive understanding of both the biophysical and social dimensions of the farming systems. While the observational data provided objective information on trends in land restoration outcomes, the participatory approach offered nuanced insights into how diverse communities responded to shocks and the relative importance of resilience attributes in mediating the impact of individual and multiple shocks on restoration outcomes in the face of multiple disturbances. By doing so, the research ensured that local knowledge and lived experiences were systematically incorporated into the analytical framework, thereby strengthening the contextual relevance and validity of the findings.

Furthermore, the study sites were deliberately selected to represent contrasting Agroclimatic conditions—semi-arid and humid zones—as well as varying levels of restoration performance, including highly successful and less successful cases. These contrasting contexts all reflect smallholder mixed crop–livestock systems that have experienced recurrent agricultural shocks over the past two decades. This design facilitated a nuanced interpretation of the interactions between biophysical and socio-economic factors influencing restoration success and resilience trajectories. Consequently, it provided a robust basis for examining how ecological conditions and restoration histories jointly shape community resilience and adaptive capacity.

The subsequent sections provide detailed accounts of the case study sites, including the types and magnitudes of shocks encountered, the methodological procedures used to generate spatial and temporal data for selected KPIs of land restoration trajectories, and the participatory processes employed to define the explanatory variables underpinning the resilience attributes.

Description of case study sites

This study was conducted in the Ethiopian Highlands, which are characterized by a mixed crop–livestock farming system, where rainfed cereals such as teff (Eragrostis tef), barley, wheat, and maize are cultivated alongside pulses and fodder crops, while livestock provide draught power, manure, and income⁶³. This integration enhances nutrient cycling and soil fertility but also places significant pressure on the limited land resources. High population density and smallholder dependence contribute to the vulnerability of the system, making it highly sensitive to environmental stress.

Farmers in the highlands frequently encounter agricultural shocks and stresses, including recurrent droughts, floods, locust invasions, pest outbreaks, market fluctuations, labor shortages, and conflicts. These shocks can sharply reduce crop yields, livestock productivity, and household food security^64,65. The region is also affected by significant land degradation, driven by deforestation, overgrazing, nutrient depletion, and cultivation on steep slopes, which diminishes soil moisture retention, carbon sequestration, agricultural productivity, and the capacity to adapt to climate change^66,67,68,69.

In addressing these land degradation challenges, extension workers facilitated the self-organization of communities into watershed user cooperatives, which collectively implement sustainable measures for the management and use of watershed resources^70,71. Through the support of extension workers, the watershed user cooperatives elect watershed development executive committees—hereafter referred to as executive teams—which provide oversight of the development, use, and management of watershed resources.

The executive teams comprise diverse social groups, including women, landless youth, farmers, and religious leaders. They are mandated to identify and prioritize land management intervention areas within the watersheds, formulate community rules, mediate conflicts, and monitor implementation^44,72.

Despite initially similar levels of land degradation and comparable financial and technical support from the SLM program, restoration performance after more than a decade (2012–2021) of program implementation varied among the communities. In this study, we selected twelve community watersheds—six highly successful and six less successful. This classification was based on the extent of sustainable land management (SLM) implementation in their watershed over the ten-year (2012 to 2021) program intervention period. The highly successful communities had implemented SLM practices across more than 82% of their watershed areas, whereas the less successful communities managed implementation on less than 50%⁴⁴. The selected sites represent two contrasting agroecological contexts: six watersheds (three high-performing and three low-performing) in the semi-arid Tigray region, and six (three high-performing and three low-performing) in the humid Southern region (Fig. 5).

The community watersheds in the semi-arid region receive about 600 mm of annual rainfall concentrated between mid-July and mid-August, whereas those in the humid watersheds receive over 1200 mm, distributed across 6–8 months. Despite these climatic differences, both regions experience similar degradation pressures. Anthropogenic drivers such as deforestation, overgrazing, and continuous cultivation, combined with natural stressors including steep slopes, intense rainfall, and erodible soils, continue to threaten landscape resilience and the sustainability of restoration outcomes.

Exposure to agricultural shocks

The farming systems in the case study sites have experienced multiple agricultural shocks and disturbances over the past two decades. Drawing on the literature and input from key informants, we found that major disturbances include: severe droughts in 2002/2003 and 2015/16⁷³; desert locust outbreaks during the 2019/20, 2020/21, and 2021/22 cropping seasons^74,75; the civil war from late 2020 to 2022⁵⁸; and the COVID-19 pandemic from 2020 to 2021⁷⁶ across different locations in the country. The case study sites in the Tigray region faced compounded challenges, including severe droughts, the COVID-19 pandemic, desert locust outbreaks, COVID-19 pandemic and civil conflict, all of which disrupted farming systems and restoration efforts. In contrast, communities in southern Ethiopia experienced COVID-19 as a major disturbance during the same period (Fig. 5). In this study, we define major disturbance as an event that significantly challenges a community’s regular farming practices, thereby affecting the stability and functioning of its farming system.

Assessment of land restoration outcomes

To assess the restoration outcomes of the selected communities, we focused on three KPIs: green soil cover fraction (the proportion of months in a year during which the soil remains covered by vegetation), soil moisture content, and annually aggregated biomass (including both agricultural and forest biomass), used as proxies for land productivity, farming system health, soil carbon sequestration and climate change adaptation^77,78. Given the absence of site-specific data on soil conditions, land productivity and soil erosion trends, we utilized satellite-derived data from multi-source independent EO data to track land restoration trajectories^66,79. These satellite-based assessments were complemented by field observations and focus group discussions to capture local insights and validate patterns observed in remote-sensing data. Using these parameters, we analyzed these metrics on a yearly basis for the period 2000 to 2023 across the twelve case study communities. The analysis was conducted using the following approaches:

Green cover fraction was derived from Landsat Analysis Ready Data produced by the Global Land Analysis and Discovery (GLAD) team at the University of Maryland⁸⁰. The dataset harmonizes Landsat 5 TM, Landsat 7 ETM+, and Landsat 8 OLI/TIRS. Landsat is the only high-resolution source that has acquired high-quality and consistent ~30 m resolution Earth observation data from 2000 to 2022 worldwide⁸¹. For this analysis, satellite imagery of the study sites for 2023 was not available; therefore, our analysis covers the period 2000–2022, for which data were available.

A bimonthly product (i.e., one image every two months) was derived for the study watershed for each year from 2000 to 2022 using weighted temporal aggregation, where weights were assigned based on the clear-sky fraction. This approach helped remove noise and capture a stable seasonal pattern for each pixel^81,82.

The green soil cover fraction was derived using the concept of the bare soil fraction index. The bare soil fraction was calculated by dividing the number of pixels classified as bare surface within a year’s time series (identified by NDVI values below 0.35) by the total number of pixels analyzed in that year⁸⁰. Subsequently, the green soil cover fraction was defined as:

Where:

• N = Total number of pixels analyzed in the watershed for a given year i

• X(NDVI ≥ 0.35) = The number of pixels with NDVI values below 0.35 for the given year

• n = Number of observations each year i

We utilized the Global Land Surface Satellite (GLASS) soil moisture product (available for 2000–2020) to monitor changes in soil moisture content in the uppermost soil layer (0–5 cm) of the study watershed. During the assessment period (conducted in October 2024), GLASS images for 2021–2023 were unavailable; therefore, our analysis covers the period 2000–2020, for which data were available.

The soil moisture data were generated using an ensemble machine learning approach that integrated multiple datasets, including surface reflectance, in-situ soil moisture observations, European reanalysis (ERA5-Land) soil moisture products, and auxiliary data such as DEM and soil information from Soil-Grids⁸³. Although the data were not derived from direct observations, they provided extensive spatial coverage and exhibited high spatiotemporal consistency⁸⁴. Finally, the soil moisture data generated using machine learning were resampled to a 30-m resolution to align with the pixel size of Landsat-derived products.

We applied a method developed by ref. ⁸⁵ to estimate aboveground biomass (AGB) using satellite-derived passive microwave instruments. According to ref. ⁸⁵, Vegetation Optical Depth (VOD) has a nonlinear correlation with aboveground biomass. Enhanced Vegetation Index (EVI) images from 2000 to 2023, which were available during the assessment period in October 2024, were obtained for the study sites. These data, derived from the MODIS MOD13Q1 product via Google Earth Engine, provide 16-day composite EVI observations.

To ensure data quality, we extracted stable signals and filtered noise on a yearly basis using a pixel-wise weighted Savitzky–Golay smoothing filter from Phenofit, an R package⁸⁶. Subsequently, we fitted a non-linear model by resampling 250-m EVI data onto a 0.025-degree (~10 km) grid, which was averaged to align with the VOD data. The relationship between EVI and aboveground biomass was modeled as follows:

Where:

• AGB: Above ground biomass

• a, b, c, d: Model parameters

• EVIy: Enhanced vegetation Index for year y

The model is calibrated within Ethiopia’s national boundaries. We were unable to calibrate it for the selected agroclimatic zones due to insufficient data for zone-specific calibration. The fitting results show an R² of 0.87 and demonstrate the model’s ability to simulate the steep increase in biomass with higher EVI values, which occurs concurrently with high-biomass forests. EVI values do not exhibit the same high-biomass saturation as observed in NDVI-based metrics (Fig. 6). Once the model parameters were determined, it was applied to the 250-m yearly EVI sum at the watershed level to produce a 250-m resolution time series of aboveground biomass for the years 2000 to 2023.

Finally, the annual median values of each KPI were used to analyze and compare trends among community watersheds, categorized into high- and low-performance groups based on the proportion of watershed area under SLM. These comparisons were conducted within the semi-arid Tigray region and the humid southern region of Ethiopia, both of which have experienced varying shocks.

Assessment of resilience attributes

Building on previous approaches for evaluating resilience in farming systems^{9,25,87,88,89}, we employed Focus Group Discussions (FGDs) with members of the executive teams. Accordingly, a total of 12 FGDs were conducted, each comprising 8 to 10 members of the executive team, selected based on their willingness and availability during the data collection period (January to March 2024). Participants were informed that their responses would be kept confidential and that they could withdraw from the discussion or interview at any time. In accordance with the German Agency for International Cooperation (GIZ)—the employer of the first author—each participant was also asked to provide voluntary consent before taking part in the study.

To capture the influence of the disruptions identified at the study sites on farming systems, each FGD rated the impact of these shocks in their respective communities using a scale from 1 (very low) to 5 (very high). Accordingly, we analyzed the median value of FGD participants’ responses based on the median ratings of respondents in Tigray for drought, desert locust invasion, and civil war, as well as the median ratings across both regions for the impact of the COVID-19 pandemic.

Following the assessment of the impacts of the disruptions, we the engaged executive committee members from each watershed to identify the contribution of resilience attributes in coping with the observed shocks. To achieve this, we conducted one-hour, in-depth discussions with each group, focusing on their conceptual understanding of resilience. During these discussions, it became evident that resilience was a well-established concept across all communities. Participants consistently described it in their local languages as the capacity to withstand or cope with shocks, reflecting a shared and practical understanding of the term. Subsequently, we asked each FGD, “Which resilience attributes enabled your community to cope with these disruptions?” To capture context-specific perspectives, participants were not constrained by a predefined set of ABCD attributes but were instead encouraged to reflect on their lived experiences and local realities. Through extensive brainstorming, the FGDs initially identified 35 resilience features, reaching response saturation. These features were subsequently refined by each community to ensure relevance to their specific watersheds. Through this iterative process, at least seven of the twelve communities (over 50%) consistently recognized 28 features.

To provide an analytical structure, we subsequently organized the twenty-eight features under the four key resilience attributes—agency, buffering, connectivity, and diversity—based on their characteristics. Specifically, six features were categorized under agency, ten under buffering, seven under connectivity, and five under diversity (Figs. 3 and 4). Each attribute was further elaborated and contextualized with input from FGD participants. Finally, participants rated the contribution of each attribute to coping with disruptions over the past decade using a Likert scale ranging from 0 (no contribution) to 5 (very high contribution), thereby reaching a consensus on the relative importance of each attribute.