This study was conducted in Aba-Bora watershed, Halaba, southern Ethiopia (Fig. ). Selected characteristics of the watershed are summarized in Table  (Mekuria et al., 2023). Households in the Aba-Bora watershed make their livelihood mainly from a subsistence mixed crop–livestock farming system (Mekuria et al., 2023). A number of households also engage in on-farm and non-farming activities, but agriculture and crop sale incomes account for over 80 % of household incomes (Mekuria et al., 2023).

Declining soil fertility, severe soil erosion, reduced access to surface and groundwater, and poor water quality are the main socio-economic and environmental challenges in the studied watershed (Sinore and Umer, 2021). In particular, gully erosion (Fig. ) is one of the main causes of land degradation, manifested by deep gullies exacerbated by increased deforestation in the upstream areas (Mekuria et al., 2023), soils susceptible to soil erosion (Yakob et al., 2022), and a shift from livestock- to crop-based agriculture (Byg et al., 2017). The lack of low-cost gully rehabilitation technologies and the awareness required from local communities to address gullies at the early stage of gully initiation and formation have also contributed to the expansion of gullies (Addisie et al., 2017).

Nine paired gullies (i.e. treated and untreated) were selected (Table ) from the mid-slope position (Fig. ). They all had depths of less than 3 m to control the influence of gully size and location on the low-cost gully rehabilitation measures investigated. Moreover, paired plots had similar soil, topography, and land use and land cover, with spatial distributions of less than 2 ha of land area so that climatic variables, such as rainfall and temperature, were similar. The study considered three low-cost gully rehabilitation measures as treatments (Fig. ), each replicated three times. The treatments were (1) regrading the gully head (45°) by establishing stone riprap (riprap treatment); (2) regrading the gully head and gully bank by establishing stone riprap at the gully head and planting grasses on the gully banks (riprap–grass-planting integration treatment); and (3) regrading the gully head and gully bank by establishing stone riprap at the gully head, planting grasses on the gully banks, and constructing small sandbag check-dams (riprap–grass-planting–check-dam integration treatment). The on-farm field experiment was established in 2021. In this study, low-cost gully rehabilitation measures refer to interventions that can be implemented within farmers' capacities or with limited external support.

In the established field experiments, the upward expansion of gully heads in the treated and control gullies was monitored at different intervals, after 4, 9, 12, 17, and 26 months. The upward expansion of gully heads was measured against pegs installed at 1 and 2 m from the gully head. The distance of a permanent reference point (i.e. big trees on the farm and close to the investigated gullies) from the 2 m peg was also measured in case it was removed by the upward expansion of the gully head with time. To estimate the volume and amount of soil loss during the period of investigation, the depth, width, and length of the upward-expanded part of gullies were measured at the beginning and end of the project period at 26 months. The soil loss from the upward expansion of gullies was calculated by multiplying the volume of soil by an approximate bulk density of 1.3 gcm-3. Erosion further down the gully was beyond the scope of our study.

According to Yakob et al. (unpublished), the soil in the study area is made up of Luvisols, with an argic subsoil horizon containing more clay than in the topsoil and a high base saturation at 50–100 cm depth. Selected characteristics of the soil in the study area are summarized in Table .

Four different forms of stakeholder engagement and qualitative data collection were conducted: (i) field days, (ii) key-informant interviews involving experts, (iii) focus group discussions, and (iv) household surveys. Details of these activities are summarized below, with details provided in the Supplementary.

Field days. To promote knowledge and understanding of the experiments, a field day was organized in the kebeles where the on-farm trials were established. This was held in November 2022 in collaboration with the Southern Agricultural Research Institute (SARI) and district and zonal agricultural offices. Kebeles are typically localities containing around 500 households, with an administrative centre around which agricultural extension is typically organized. The field day focused on the demonstration of gully head treatments and gully bank stabilization, as well as incentives used to support the sustainable management of land resources. More than 55 participants, comprising farmers; extension agents; experts from district-, zonal-, and regional-level agricultural offices; and local administrative bodies, attended the field day to share ideas.

Key-informant interviews. Key-informant interviews were conducted to understand the system and to assess the drivers, pressures, states, and impacts of and responses to (DPSIR) gully erosion and to develop the conceptual DPSIR framework of the interventions for gully erosion and formation. The DPSIR approach was adopted as it has been widely deployed to provide a causal framework to link environmental and social processes (Kristensen, 2004). A total of 25 key informants consisting of 8 local practitioners (district and zonal agricultural offices), 10 knowledgeable farmers, 2 NGOs, 2 cooperatives, 2 civil societies or community-based organizations, and 1 local administrative body were selected. We used semi-structured questionnaires and checklists to gather data. The discussion points with key informants included the causes and severity of gully erosion, existing gully rehabilitation measures, the effectiveness of responses to address gully erosion, the management and allocation of natural resources, and the key changes following the implementation of gully rehabilitation measures. Knowledgeable farmers in this study were experienced in organizing and leading watershed management practices, members of community watershed teams, and early adopters.

Focus group discussions. Two focus group discussions (men and women separately) assessed the costs and benefits of five gully rehabilitation measures from the perspective of local communities. The assessed gully rehabilitation measures included both those in our field experiments and others commonly implemented locally, i.e. gully head treatment, gully reshaping and planting of grasses, loose-rock check-dams, vegetation log check-dams, and gabion check-dams. The focus groups first discussed the selected gully rehabilitation measures (i.e. the practices were presented using understandable descriptions in the local language and were confirmed by the participants), specifically the advantages and disadvantages from the perspective of the local communities. Next, the participants rated the importance of each identified advantage and disadvantage, giving scores of 3 points for the most important, 2 for those of intermediate importance, and 1 for the least important. Finally, the participants graded each practice or intervention as positive, negative, or neutral.

The qualitative data collected through key-informant interviews and focus group discussions were analysed using content analysis (Hsieh and Sannon, 2005; Bernard, 2006; Elo and Kyngas, 2008). For qualitative analyses, we developed a coding scheme based on the main thematic areas, including items such as drivers of gully erosion and formation, pressure on the natural resources, response of gully erosion, impacts of gully rehabilitation measures, and decision-making processes. These were coded manually. Deductive manual coding, which starts with a predetermined set of codes derived from the conceptual framework used in this study (i.e. the DPSIR framework), was then employed using codes such as the severity of gully erosion, types of drivers and impacts, and types of interventions. During the coding process, more codes were added when new issues emerged from the textual data. In addition, codes were merged, removed, or modified to avoid repetition and to resolve disagreements between codes.

Household surveys. In addition, two household surveys were conducted in 2021 and 2023 in the Aba-Bora and nearby Guder watersheds in localities where land was judged to be highly degraded with large gullies present. This included the kebele where the on-farm field experiment and focus group discussions took place. In 2021, data were collected from 522 households (including 123 from the kebele where the on-farm field experiment took place). In 2023, a follow-up survey collected data from 500 households, of which 451 could be identified as being the same respondents as previously (including 114 where the on-farm field experiment took place). The sample represented a random sample of households drawn from lists provided by each kebele administration after stratification based on wealth and gender status. The data were collected during February and March in 2021 and 2023 using a team of enumerators employing computer tablets, with the questionnaires being available in both English and the local language Amharic (S1 in the Supplement).

The 2021 and 2023 household surveys included a series of questions exploring respondents' knowledge, attitudes, and behaviour with respect to land degradation and gullies. The respondents' data were compared in terms of “treated” and “non-treated” to explore the potential impact of the on-farm experiment. Where identical questions were asked in both the 2021 and 2023 surveys, the impact of the on-farm experiment was examined using a panel difference-in-difference (DD) approach, controlling for individual fixed effects and a common trend (Glewwe and Todd, 2022; see also S1 and S2 in the Supplement). The DD model compares changes in the indicator over time. Therefore, the DD model compares trends in the control group from before and after the project to trends in the treatment group. The double difference then refers to the difference over time (the first difference) and the difference between the control and treatment (the second difference). If the trends are statistically significantly greater for the treatment group (in a statistical sense), this suggests that the intervention had an impact. Thus, the DD estimator combines cross-sectional and over-time variation to correct for differences between groups when the treatment and controls start at different levels. Sections S1 and S2 provide further details of the questions asked and the DD method.

The 2023 survey also included several more specific questions (items) on gullies, drawing on the nature of the on-farm experiments, to explore respondents' knowledge, understanding, perceived capacity to act, and behaviour with respect to land degradation and gullies. Knowledge, understanding, and perceived capacity to act can be conceived as latent (unobservable) values but where different survey questions (scales or items) can be designed to capture these underlying unobserved values (see S1 and S2).

The specific approach applied here draws on the marketing literature to consider both the recall and recognition of different interventions (Boshoff and Gerber, 2008; Macdonald and Sharp, 2010; Hoyer and Brown, 1990). This approach has also been used to explore how knowledge and understanding may relate to attitudes about nature conservation (Pearson et al., 2022; Veríssimo et al., 2017; Schlegel and Rupf, 2010). In line with this approach, the respondents were first asked in an open-ended-question format to name three interventions or activities that they thought were important in reducing land degradation. The answers were then grouped and weighted to provide an awareness score ranging from 0 for completely unaware to 3 for very aware (Boshoff and Gerber, 2008). The second set of questions explored the extent to which the respondents have knowledge of specific interventions. Here, the respondents were shown pictures of different interventions, where the first two (gully head treatment and reshaping gully banks) could be linked directly to treatments one and two implemented in the on-farm experiments, while the third and fourth interventions shown could be more loosely related to treatment three (Table ). First, to explore respondents' recognition and understanding, they were asked about their view of the effectiveness of the measures. Second, to explore their perception of their capacity to implement the measures, they were asked whether they could undertake these measures on their own or if they would need the help of neighbours or full community mobilization (see S1 and S2 for the specific question formats used). The original five-point answers were aggregated into the three categories reported on (on their own, with neighbours, community mobilization). The reliability with which the various sets of questions (items) captured the same underlying latent variable (knowledge, understanding, and capacity to act) was tested using Cronbach's alpha. Cronbach's alpha is an internal-consistency method of reliability that shows a measure's homogeneity and that occurs when different attempts at measuring a concept converge on the same result (Churchill, 1979). The coefficient alpha ranges from 0 to 1, and a coefficient alpha of 0.7 or greater is considered to be an acceptable measure of reliability (Taber, 2018).

The more specific questions on gullies asked in 2023 only were analysed using a simple comparison between the treated and non-treated kebeles. This included analysing the respondents' capacities to implement the measures by estimating a simple multinomial logit to model the probability that a respondent felt interventions could be implemented by them on their own, with the help of neighbours, or by means of community mobilization (S2).

Regardless of the mitigation practice, all treated gullies had negligible upward expansion (Fig. ). This contrasted with untreated gullies that expanded by a few centimetres after 4 months, increasing up to several metres after 26 months, ranging from 19 (± 4.3) to 671 (± 354) cm. Assuming a bulk density of 1.2–1.5 gcm-3, the estimated soil loss due to the upward expansion of gully heads in the untreated gullies ranged from 2.8 to 20.2 t, with a mean value of 11.0 (± 5.9) t, over 26 months. This value was zero in the treated gullies.

Good knowledge of gully initiation processes was demonstrated by key informants. They indicated that gullies usually initiated in the upstream areas and took the form of rill erosion. Over time, they recognized that rills got longer and deeper in the mid- and foot-slope landscape positions to form gullies. They remarked that the density and depth of gullies usually increased from mid-slope to foot-slope positions but that gullies were unevenly distributed within the watershed, with a few locations eroding differently along the slope. The key informants identified multiple human-induced, natural, and climatic drivers of gully erosion and formation (Fig. ). About 80 % of the key informants characterized the severity of gully erosion as very severe, while the rest characterized gully erosion as severe, causing environmental, social, and economic impacts. Gully erosion was considered to cause both long- and short-term negative impacts that affect the livelihood of local communities (Fig. ). The key informants also mentioned multiple initiatives implemented in the watershed to address gully erosion. These initiatives covered multiple dimensions: capacity building, providing technical support, and implementing diverse interventions or measures (Fig. ). The commonly implemented gully rehabilitation measures included check-dams (loose-rock, sandbag, and gabion check-dams), biological measures such as planting multipurpose tree species and grasses on gully banks, and gully head treatments.

The majority (90 %) of key informants indicated that the implemented gully rehabilitation activities were successful but that they did not cover all affected areas and were inadequate compared with the severity of the problem. They further elaborated that the rehabilitation measures not only reduced soil erosion but could also convert degraded lands into productive lands. The restoration of degraded lands by gully rehabilitation included reduced expansion and upward expansion of gullies, reduced runoff and soil loss, and improved cultivated- and grazing-land management (Fig. ). Furthermore, key informants suggested that gully rehabilitation measures enabled farmers to diversify their crops to grow cereals, vegetables, and fruit. This was mainly attributed to reduced soil loss that resulted in enhanced soil nutrient retention and sometimes the accumulation of sediments behind check-dams. One of the key informants elaborated upon this as follows: “Farmers who migrated have started to come back and grow cash crops such as khat, vegetables, and maize due to the improvement of soil fertility. Particularly, this is observed at the downstream areas of each treated gullies.”

In relation to the contribution of gully rehabilitation measures to livestock production, one of the key informants stated the following: “Decades ago, households used to have more than 30 livestock on average; however, the expansion of overgrazing and shortage of livestock feed forced them to reduce the number of livestock. Following the recent community-based watershed development activities, including gully rehabilitation works, the number of livestock per household is increasing due to increased production and availability of livestock feed from restored areas and gullies.”

The ongoing rehabilitation activities were also key to establishing community awareness about the negative consequences of gully erosion and the need for collective action to control gully erosion. The key informants mentioned that effective implementation and sustainability of gully rehabilitation measures required continuous technical support, particularly if more challenging biological measures were adopted. However, a few (10 %) of the key informants stressed that it was hard to judge the success of interventions compared to the challenges of gully erosion. The majority (90 %) of key informants also indicated that the efforts were not adequate compared with the severity of the problem.

The key-informant interview results suggested that gully erosion can be addressed by individual or collective actions. Particularly, the respondents mentioned that small gullies (of less than 3 m depth), such as those assessed in our field research, can be rehabilitated using locally available materials and capacities. However, once the gullies get large and deep, it is beyond the capacity of local communities to rehabilitate them. The decisions on the planning, design, and implementation of gully rehabilitation measures combined both bottom-up and top-down approaches. At the local or kebele level, a community watershed team was the key decision-maker, dictating intervention areas, types of interventions, and the labour and material contributions of communities. However, there were also cases where the regional bureau and zonal departments allocated a quota (e.g. types and quantities of interventions) to each district and kebele. The regional and zonal departments and NGOs were also involved in providing technical, material, and financial support.

Community-level decisions in relation to gully rehabilitation measures usually depend on the contribution of free labour and locally available materials. Our survey found that this extended beyond building materials, such as stone and wood, to the equipment and tools required for construction. In line with this, the community watershed team worked closely with kebele- and district-level experts to decide on community mobilization. This determined who participated, the types of interventions, and the use of material and labour. There were also household-level decisions that could be implemented by individual households, such as taking measures to prevent runoff from entering their houses and farmlands. On and around their farmlands, these usually involved constructing wooden or vegetation check-dams, water-harvesting structures, and bunds. Some individuals identified by the key informants were aware of gully erosion and took their own actions around the homestead by planting trees or by means of gully head treatment.

Cooperative efforts had also been implemented to effectively manage gully rehabilitation measures, including informal platforms established by Idir (self-support groups) and Iqub (voluntary cooperatives). Some actions were instigated by bylaws and so worked closely with formal institutions to enforce laws and to protect interventions from free-riders (i.e. those who seek benefits while violating existing rules and regulations). The cooperative efforts had incentive mechanisms, such as engaging landless young people and women in income-generating activities through providing beehives, sheep, and oxen. Such incentive mechanisms strengthened the protection of land management practices, including gully rehabilitation measures, in a way that could be sustained and beneficial to farmers. Furthermore, the use of locally available materials and the involvement of elders and religious institutions in awareness creation campaigns were effective strategies to sustainably implement gully rehabilitation measures.

The key informants mentioned multiple sources of information. A majority (40 %–60 %) of the respondents indicated that the key sources of information were the ongoing gully rehabilitation activities, personal experience, and observation. The main limitation of gully rehabilitation measures constructed using locally available materials was the short lifespan of physical interventions, such as sandbag or log check-dams. However, when physical interventions were combined with vegetation, the longevity of the intervention improved. Our findings suggest that most of the gully rehabilitation measures were well received by the communities, with the only concern being a temporary loss of access to their land or resources during construction. Out-scaling and/or wider implementation of gully rehabilitation measures were constrained by a shortage of personal tools (mentioned by 30 % of the respondents), limited access to inputs and raw materials (30 %), increasing prices of inputs (e.g. gabion wires, 60 %), high labour demands (10 %), technical requirements (30 %), and limited sources of finance (10 %).

Participants of focus group discussions, representing men and women separately, identified 24 benefits (or advantages) and 17 costs (or disadvantages) of the five gully rehabilitation measures assessed in the Aba-Bora watershed (Tables S1 and S2 in the Supplement). Across both focus groups, the number of benefits identified for most of the gully rehabilitation measures outweighed the costs (Fig. ). There were gender differences in the perceived costs and benefits. While female groups identified equal numbers of costs and benefits for loose-rock check-dams, male groups felt the benefits were greater than the costs. An opposite trend between genders was observed for gabion check-dams (Fig. ), where men indicated higher costs than women. The groups of both women and men identified more benefits for gully reshaping and planting, indicating that there was a consensus among the two groups regarding the benefits of this intervention. Both groups identified comparable numbers of benefits and costs.

The benefits were diverse (Table S1) and covered environmental (e.g. conservation and increased access to water, improved soil fertility, and reduced runoff and soil erosion), economic (e.g. increased agricultural production, reduced farmland loss, and increased availability of livestock feed), and social (e.g. movement of people and livestock, reduced damage by floods, and knowledge exchange) categories. There were 24 identified benefit factors, but women were the only respondents to identify (i) the use of locally available materials, (ii) conserving and increasing access to water, and (iii) increased access to productive land. On the other hand, the men's groups listed eight benefits that were not identified by the women's groups. These were (i) increased income, (ii) increased access to diverse food, (iii) improved appearance of the landscape, (iv) a reduction in damage caused by floods, (v) regulation of micro-climate, (vi) ease of constructing the interventions, (vii) durability of the interventions, and (viii) serving as a learning site (Table S1). The participants also identified several costs of the assessed interventions or practices that were environmental (e.g. lack of effectiveness in reducing soil erosion and runoff), economic (e.g. high labour demand, cost, shortage of local available inputs), or social (e.g. demand for skilled manpower, technical support) (Table S2).

Across both groups, the number of environmental benefits identified by the participants was larger than the respective environmental costs (Fig. ). Whereas female groups identified more social costs than benefits, male groups had an opposite opinion (Fig. ). Across both groups, the benefits of the assessed gully rehabilitation measures mainly related to the categories of environmental and economic benefits, whereas the costs largely related to the categories of economic and social costs (Fig. ).

The local surveys suggested that the benefits outweighed the costs for most of the interventions (Fig. ). The ratings of the importance of benefits assigned by women's groups across all the interventions varied between 2.3 and 2.5 (i.e. the rating is out of 3, and 3 represents high importance, 2 represents medium importance, and 1 represents low importance), with a mean value of 2.4, while the ratings of the costs ranged from 1.8 to 2.5, with a mean value of 2.0. The ratings of benefits assigned by men's groups varied between 2.2 and 2.8, with a mean value of 2.5, while the ratings of the costs ranged from 2.4 to 2.9, with a mean value of 2.6.

Apart from gabion check-dams, all other gully rehabilitation measures assessed were evaluated as positive (Table ). The local communities considered environmental and economic factors more than social factors in their overall assessment of the gully rehabilitation measures (Table ).

The knowledge and awareness of gullies exhibited by the > 500 respondents in the household surveys are summarized in Table . The data describe typical attitudes about land degradation and the respondents' views on the extent to which gullies have advanced between experimental locations with and without interventions in place. As far as the authors are aware, this represents the first evidence of this type, which explores whether local demonstrations of gully treatment effectiveness have an impact on local farmers' knowledge, awareness, and behaviour. In response to a general question about the extent of land degradation and gully erosion, there was no evidence of a change in attitudes before and after the experiments took place. However, in the treatment areas, there was some evidence of changes in perceptions regarding the extent and most important types of soil erosion and where gully erosion occurs, with a greater perception that the main impacts are focussed on farmland.

Table  reports the estimates using questions specifically asked in the 2023 survey and represents the difference between the mean in the treated and non-treated areas. Here, the recall questions consider whether respondents were aware of and able to recollect different land degradation measures from a general cue (i.e. from the list of land restoration measures presented to the respondents). The differences in the scores between the treated and non-treated areas are reported in the first panel of Table . Some differences in the levels of recognition were apparent, such as greater recognition of mountain measures such as terracing, high recognition but with weak evidence (p value of 0.12) of gully measures, and lower recognition of grassland management measures.

Through these household surveys, we confirmed our hypothesis that demonstrating interventions through field studies influenced the knowledge and understanding of similar gully treatments among respondents. The average measures of effectiveness were all higher for those living where demonstration gullies were established, especially for the gully shaping and check-dams. Testing the set of questions (items) using Cronbach's alpha confirms that they can be interpreted as capturing a single underlying characteristic, i.e. knowledge of the gully treatments (alpha 0.74). Testing the averaging aggregate score also suggests that there was a statistically significant difference between the knowledge and understanding of these measures in the area where the on-field experiments took place.

The last panel in Table  reports the results from the multinomial logit estimation applied to the questions (items) regarding an individual's capacity to deal with different gully treatments (see S2 in the Supplement). The estimates reported represent the marginal effects of being in the treated area on the probability of selecting a particular option. For the gully head treatment, for example, respondents in the treatment area had a 0.085 lower probability of selecting “could do this alone” than those respondents in the areas where no experiments took place. Overall, there was evidence that the respondents in the treated area have a different view of the resources required to undertake the measures. The probability that the respondents would choose the option “could do it alone” or “need neighbours” was lower in the treatment areas in all cases, except for the reshaping of gully banks treatment, and the probability was higher for the option “need community mobilization”, with the latter underlying difference being statistically significant (at 5 %) in three out of the five different treatments. Hence, the respondents appeared to be generally more pessimistic about their ability to undertake this alone or with neighbours. This emphasizes that increased knowledge of gully treatments may mean farmers could become aware of the challenges required to implement treatments.

The final set of questions in the last panel for Table  assessed possible differences in behaviours related to gullies between the samples (i.e. household survey respondents) in the areas with and without the on-farm experiments. Here, there are statistically significant differences that are consistent with greater efforts being made to undertake work to reduce gullies in the treated area. The averages for all questions were statistically significantly higher (p < 0.05) for the respondents in the treatment area; e.g. the average was 0.397 points higher for questions involving individual work and work with neighbours and friends and 0.248 points higher for questions around farmers actively trying to decrease gully formation. This is consistent with the hypothesis that the on-farm experiments and field day demonstrations have positively affected farmers' behaviours (or intentions) in acting to help restore and prevent gullies (despite their apparent greater pessimism about their capacity to do so alone).