2.1 Study area

The survey was conducted in 55 villages across four territories (Walungu, Idjwi, Kabare, and Kalehe) of the South-Kivu highlands (Table 1). These four territories cover a wide range of biophysical et socioeconomic contexts encountered in the highlands of South-Kivu. Two watersheds per territory were selected taking into account the watershed's representativity in terms of soil type and smallholder farming systems but also their accessibility (Table 1). As a matter of fact, bad road conditions and insecurity (presence of some armed groups) are major constraints to access some areas (Ulimwengu, Funes, Headey, & You, 2009).

The watersheds were delimited based on the 30-m resolution DTM available from the ASTER GDEM2 website (METI, NASA, 2011) (Figure 1). The soils derive from volcanic and basaltic parent materials (mollic Andosols, systic Nitisols, humic Nitosols and eutric Cambisols) for the four watersheds in the northern part of the province (Kabare and Kalehe) or from ancient soil materials, mostly basaltic and strongly weathered (orthic Cambisols, dystic Nitisols and humic Nitosols) for the Walungu territory and the Idjwi Island (FAO, 1998) (Table 1). The study area experiences a tropical highland climate with a bi-modal rainfall distribution resulting in two wet (September–January and February–May) and one marked dry (June–August) seasons. As a result, there are two main cropping seasons per year. The 10-year average annual rainfall recorded in this area varies between 1,300 and 1,600 mm (IPAPEL, 2010).

In each watershed, the villages were selected using two criteria: (a) the absence of insecurity constraints and (b) a majority of the population being engaged in agricultural activities within the village. Indeed, in some villages, mining and brick-making activities dominate. Based on a list of villages bearing, these criteria for each watershed, a guided choice was made to cover as much as possible the cultivated areas in each watershed.

2.2 Data collection

In order to have a comprehensive understanding of farmers' opinion on soil erosion and SWC practices, we collected information through individual semi-structured questionnaire interviews. A total of 720 randomly selected farmers were interviewed, 90 per watershed. The survey aimed at assessing (a) the severity of land degradation by soil erosion and the soil fertility status, (b) farmer's knowledge regarding erosion and in particular indigenous indicators used by farmers to assess the occurrence and severity of erosion, (c) causes of soil erosion, with special emphasis on farmer's perception regarding the impact of their crops and cropping practices on soil erosion and (d) farmer's knowledge about SWC techniques and perceived constraints or reasons for non-adoption.

For this survey, respondents were heads of farming households (males or females). In case of absence, they were represented by their partners or grown-up children living within the household and having sufficient knowledge of the different activities of the household. The surveys were conducted in local languages, and, when needed, questions were explained by a member from the local development committee. The survey was done either at the farmer's home (when the farmer field was close to the house) or in his field so as to conform to their working programs, given that the period during which the survey was done coincided with the beginning of the cropping season. Being in or close to the field also allowed easy clarification or verification of the farmer's assertions.

An erosion severity index was calculated as follows. First, farmers were asked to list all signs of erosion observed in their fields. For each sign, farmers were asked to rate its severity on a scale from 1 (weak) to 4 (very high) and to justify the score based on what could be seen in their fields. Because farmers were not asked to rank the relative importance of each sign, a weight was calculated for each observed sign by dividing its frequency by the total frequency of all signs, that is, more weight is given to signs commonly reported by the surveyed farmers. Then, the erosion severity index of a given field was obtained by adding up the weighted severity indexes of all signs observed in that field.

Based on the actual erosion signs reported by each farmer, farmers were asked to list the main causes of water erosion in their fields and to rank the importance of each cited cause on a scale from 1 to 4. Whether or not crops or farming practices were cited as causes of erosion, farmers were requested to express their opinion regarding the possible impact of these two factors on erosion (promote, neutral, restrict).

Farmers were also asked about the SWC measures implemented in their fields and to rank the effectiveness of each SWC measure on a scale from 1 (low) to 4 (very high). Based on these ratings, an average effectiveness index was calculated for each technique. Finally, all farmers were asked to provide the reasons why they did not implement a range of conservation measures.

2.3 Data analysis

Data were processed and analysed with XLSTAT (version February 2, 2016). Analyses involved descriptive statistics, including frequency distributions and graphs. Chi-squared tests were used to test for differences across watersheds. Simple logistic regression was performed to analyse the relationship between the erosion severity index and the presence/absence of SWC techniques.

3.1 Erosion indicators

Nine main erosion indicators were reported by the farmers, of which eight were listed by more than 50% of the respondents (Table 2). Two indicators relate to forms of erosion (rills and gullies), while the remainder refer to the impacts. Impact indicators relate mostly to the consequences of erosion or sedimentation on soils or crops, though one indicator also relates to the impact on labour. Whereas the presence of gullies was reported by 80% of farmers as a sign of erosion, rills were only mentioned by 50% of farmers. The frequency of observation of erosion indicators varied significantly across watersheds (p < .0001). On average over all nine indicators (excluding 'others'), erosion signs were most frequently reported in Kal2 and Idj1 (> 80% of farmers on average) and least in Wal1 and Wal2 (< 45% of farmers on average). Although gullies were commonly cited in all watersheds (>60% of farmers), some other indicators were extremely variable. This was for instance the case for 'hampering of tillage' (4–93%) and 'stoniness/rock outcrop' (1–94%) but also for 'rills' (13–88%).

Not all farmers reported all erosion signs (Figure 2). Nevertheless, the severity of the signs was, on average, perceived by 21% of the respondents who mentioned a given erosion sign as high to very high. The severity was highest for root exposure (high to very high for 38% of respondents) and lowest for rills (high to very high for 9% of respondents).

Based on the severity of the observed erosion signs, a severity index was calculated for each farmer's field. Across the watersheds, erosion severity was 2.3 ± 1.1 (mean ± SD; Figure 3). On average, 14.4% of fields are subject to severe to very severe erosion (index ≥3) (Figure 3).

3.2 Causes of water erosion

The most commonly quoted causes of soil erosion (>50% respondents) in all the eight watersheds were high rainfall, steep slopes, occurrence of runoff and nature of the soil (Table 3). Overall, less than 20% of farmers associated erosion to their choice of crops and farming practices. Nevertheless, the perception of the main causes of soil erosion varied significantly across watersheds (Table 3). On average, causes were most frequently listed in Kal2 and Idj1 (the six main causes were listed by >70% of farmers on average) and least in both Walungu watersheds (the six main causes were listed by <40% of farmers on average). High rainfall was perceived as a cause of erosion in all watersheds (> 69% of respondents). Slope was viewed as a cause by a majority (>70%) of farmers in all watersheds except in Wal2 where only 50% of farmers mentioned this factor. Regarding crops and farming practices, their contribution to erosion was nearly fully ignored in Kal1, Wal1 and Wal2 (<5%) compared to the other watersheds where 21–36% of the farmers considered that these factors contribute to erosion.

The contribution of rainfall, slope steepness, runoff and soil properties to soil erosion was perceived as being of moderate to very high importance by 86, 61, 51 and 33%, respectively, of those farmers who identified the causal factor (Figure 4). In addition to the fact that few producers identified cropping practices and crops as causal factors, more than half of those who listed these two factors considered them to be of low importance.

3.3 SWC measures and adoption constraints

As mentioned above, common cropping practices such as mulch or compost application, tillage or intercropping were seldom perceived as beneficial for erosion control [Table 4(b)]. Consistent with this, their average effectiveness was perceived as low on average (1.1–1.4 on a scale from 1 to 4; not shown). Besides the cropping practices, farmers reported seven specific soil erosion control measures (Table 5), of which drainage ditches (69% of farmers) and agroforestry systems (42% of farmers) were most commonly applied. Additionally, 15% of farmers knew about drainage ditches and 13% knew about agroforestry for erosion control but did not implement these techniques. Planting a hedge along the contour was a known anti-erosive measure for 39% of farmers, but only half of them declared planting such hedges. On the contrary, all farmers who knew about anti-erosive bunds for erosion control also implemented them (15%). Only very few farmers (<4%) declared knowing or maintaining a permanent vegetation cover, ploughing along the contour or placing brushwood fences. Drainage ditches and permanent vegetation cover were deemed most effective for erosion control, closely followed by contour hedges and contour bunds (Table 5). The effectiveness of all four measures was rated 2.3 to 2.6, that is, between moderate and high. Agroforestry was deemed moderately effective. Brushwood fences were reported as being of low effectiveness, but only seven farmers had practical experience with this technique, and among those, only four reported on the effectiveness. The rating of contour plowing and permanent vegetation cover may similarly be questioned given the low number of evaluators.

Although there were significant differences across watersheds regarding the implementation of SWC measures, no systematic pattern could be identified (not shown). Drainage ditches and agroforestry were rather less common in Wal2. Earthen bunds were more common in Kal1 and Idj1, whereas permanent vegetation cover was more often mentioned in Kal1 and Kab1.

The main reasons formulated by farmers (Table 6) when asked why they did not adopt certain SWC measures were small farm size, labour requirements, availability of suitable equipment, insufficient money and labour and perceived inefficiency of some techniques. In addition, some producers did not consider soil erosion as a big problem in their fields, so that there was no need for them to adopt SWC measures.

The main constraint for hedges and agroforestry systems was the small size of farms (Table 7). Adoption of hedges is affected by the work requirement and cost. In addition, farmers justify its non-adoption by its low-perceived effectiveness, and the fact that erosion on their plots is not such a big problem. This is also apparent in the high number of farmers that declare little motivation to implement hedges. Besides farm size, adoption of agroforestry systems suffers from cost and equipment availability. Although labour requirement is less of a problem than for hedges, farmer motivation appears as a serious constraint. Both techniques are restricted on rented plots. Drainage ditches are mostly constrained by perceived lack of erosion, labour requirements, perceived effectiveness and motivation but only little by farm size. Motivation and perceived lack of erosion also constrain the adoption of contour bunds. The use of organic amendments such as household waste and compost is constrained by the amount of work required for composting and, in the case of household waste, transportation to the fields. Availability of material is also an issue for mulches and compost, including the availability of manure to be added to the compost.

Based on logistic regression, adoption of four practices shows a positive relationship with erosion severity: drainage ditches, agroforestry, permanent cover and application of household waste (Table 8). This appears to indicate that these practices are implemented, at least in part, in response to erosion.

4.1 Farmer perception of soil erosion problems and causes

A range of impact indicators were reported by farmers, indicating a broad understanding of the impacts of erosion both on soils (reduced depth, loss of SOM and fertility, stoniness where applicable) and crops (root exposure, burial of crops). Besides these impact indicators, farmers also mentioned gullies and, less often, rills. Local farmers refer to gullies as permanent incisions which cannot be removed by routine manual land preparation activities, while rills were seen as small entwined or parallel incisions, easily erased by manual tillage operations. Given the steep slopes and heavy rainfall, gullies are very common in South-Kivu in areas impacted by human activities, and they are fully recognised by farmers as a sign of erosion (Figure 5a–d). Rills were generally among the least cited indicators, despite their common occurrence in the majority of fields and watersheds (personal observations; Figure 5b,c), and the severity of rilling was mostly ranked as medium to low. In the highlands of South Kivu, it is the ephemeral nature of rills that appears to explain the low rating as an indicator. Indeed, rills form only after significant rainfall events and are easily erased by routine land management activities (tillage or weeding). Thus, farmers do not perceive the high quantities of sediments that can be lost by this form of erosion, and the overall severity of erosion may be largely underestimated.

The rate of reporting of erosion signs varied greatly across watersheds. Similarly variable results regarding erosion signs (e.g., rills) have been reported in other studies (Okoba & de Graaff,2005; Takele, Chimdi, & Abebaw, 2015). Rates of reporting (Table 2) were always lowest in the Walungu territory. Based on personal observations, erosion was not less severe in the Walungu area than in the other surveyed watersheds (Heri-Kazi, 2020). However, Wal1 is closest to the City of Bukavu, and brick making is commonplace in the vicinity of this watershed. Wal2 is located in an area where mining activities are widespread. Although villages were selected where agriculture was the main activity, access to non-farming activities that are more lucrative than farming nevertheless seems to have impacted the rates of reporting. It may be that, because of these external sources of income, agriculture carries less importance, and farmers are less sensitive to existing signs of erosion. Whereas pairs of watersheds from the same territory were generally quite similar in terms of frequencies of responses, there is a marked contrast between Kal1 and Kal2, with much higher frequency of responses in Kal2. The explanation in this case seems to lie in the generally steeper slopes in Kal2 compared to Kal1 (Heri-Kazi & Bielders, 2020) but also in higher adoption rates of earthen bunds (+20%) and permanent vegetation cover (+31%) in Kal1 compared to Kal2 (not shown). In addition, plot sizes tend to be smaller in Kal2, such that farmers may be somewhat more sensitive to the degradation of their land than in Kal1. Finally, the Idjwi territory is on average characterised by higher reporting frequencies than the other territories. Besides steeper slopes, part of the explanation may lie in the replacement of banana fields by large-scale cassava cropping as a result of the Banana Xanthomonas Wilt (BXW) disease. Cassava has been shown to cause more serious soil erosion than many other crops (Howeler, 2008).

Similarly to other studies in the east African highlands (e.g., Okoba & de Graaff, 2005 in Kenya; Mashi et al., 2015 in Nigeria; Rutebuka, Kagabo, & Verdoodt, 2019 in Rwanda), farmers believe that high rainfall, steep slopes and runoff are the main causes of erosion in the study area (Table 3). Soils play a lesser role, while less than 20% of farmers on average link erosion to their crops and cropping practices (e.g., Bitijula & Lal, 1983). With the exception of banana plants and coffee trees which have a dense rooting system (soil stabilization), offer a permanent vegetation cover (reduced surface crusting and splash) and are therefore perceived as beneficial for soil conservation, most crops are perceived as neutral with respect to soil erosion. Farmers explained this neutrality by the fact that other external factors such as rainfall and slope steepness are so overwhelming that most crops have little impact on erosion. Besides, farmers do not perceive that growing crops cause erosion because cropping is a necessity, engrained in their livelihoods and there is no alternative to this practice. Similarly, most cropping practices are perceived as neutral, even though some farmers recognised that practices that favour the mechanical destruction of soil structure promote runoff and erosion. Regarding the neutral role of most farming practices, producers again stated that the negative impact of heavy rainfall and steep slopes is so overwhelming that the effect of farming practices becomes negligible, for example, the positive effects of good farming practices on erosion are not perceived. Even so, raising awareness about the role of crops and cropping practices on soil erosion would be an important step in order to convince farmers to adapt their practices and adopt conservation measures.

There were significant differences across watersheds regarding causes of soil erosion. As for the erosion signs, frequency of responses tended to be lowest in the Walungu territory and highest in the Idjwi territory. Frequencies were also again noticeably lower in Kal1 than Kal2. It seems therefore that the same local contexts that explained differences in reporting of erosion signs also affected reporting of causes of erosion.

4.2 Adopted control measures and their effect on soil erosion

Whereas beneficial cropping practices such as composting or spreading of waste are widespread but deemed ineffective for erosion control, specific soil conservation measures were often judged fairly effective by farmers, yet their adoption is rather limited, with the exception of drainage ditches and agroforestry (Table 5). Overall, adoption levels are substantially lower than what has been reported in other regions (Wildemeersch et al., 2013; Assefa & Hans-Rudolf, 2016). This was also observed in the Central Rift Valley of Ethiopia where, despite farmers' awareness of both water erosion and soil fertility decline, no significant investments in land management practices were made (Adimassu, Kessler, Yirga, & Stroosnijder, 2013). This is to some extent surprising given the widespread awareness of farmers of the occurrence of erosion in their fields in the study area. Indeed, de Graaff et al. (2008) showed that the first step in adoption of SWC technology is the knowledge or perception of the erosion problem. This is also supported by Tenge et al. (2004) and Haghjou et al. (2014) who stated that a good perception of erosion is the most decisive factor in the adoption of SWC technologies. It thus appears that other criteria, no less important (access to land and extension services, institutional support,… which are widespread problems in this region and neighbouring countries such as Rwanda and Burundi; that is, Rutebuka et al., 2019), are to be considered in the adoption process or are needed to ensure continued adoption of SWC technologies (Wildemeersch et al., 2013; Assefa & Hans-Rudolf, 2016, Rutebuka et al., 2019). Ragasa et al. (2015) showed that agricultural extension services suffered from the negative effects of decades of insecurity experienced in the study region, which could also be responsible for the low adoption rates of dedicated SWC measures.

Significant differences in adoption rates of SWC measures across regions were observed (not shown), but these regional differences varied from one SWC measure to another. Low adoption rates of ditches and agroforestry in Wal2 may have resulted from a lesser interest to invest in agricultural activities given the importance of mining as a source of income. However, in Wal1, adoption rates were similar to the other watersheds despite the importance of brick making as an external source of revenue, implying that availability of external sources of revenue does not necessarily lead to lower adoption rates. High adoption rates of earthen bunds in Kal1 and Idj1 coincide with a high percentage of farmers growing sweet potatoes, which are typically grown on bunds. Similarly, the high percentage of farmers referring to permanent vegetation cover in Kal1 and Idj1 coincides with a high percentage of farmers growing coffee (Heri-Kazi & Bielders, 2020). Hence the adoption rate of some of the SWC measures is intimately linked to regional differences in cropping systems.

The fact that farmers often did not mention crops and cropping practices as possible causes of erosion may reflect the perception that the problem is largely outside their control, given the overriding importance of rainfall and topography in the process. Nevertheless, our results indicate that much could be gained by simply raising the level of awareness regarding existing technologies since, for many of them, the level of awareness is not vastly different from the level of adoption (Table 5). In addition, farmers identified a range of constraints for each of the technologies that will have to be taken into account when promoting them. Several of these constraints relate to farmer motivation, perception of efficiency or perception that erosion is not severe enough to justify investment in these technologies. Training farmers to better assess signs of erosion, and especially rilling, and, at the same time, enhancing their knowledge regarding proper ways to implement SWC technologies to enhance their effectiveness may thus help with the adoption process.

Positive relationships were observed between the adoption of some SWC techniques and perceived erosion severity (Table 8). This may seem counter-intuitive because one could expect adoption of SWC measures to be correlated to lower erosion severity (e.g., Adimassu et al., 2013). The above-mentioned result, however, supports the perceived seriousness of the erosion phenomena by concerned farmers and may indicate that they may be more likely to invest in water erosion control practices when perceived erosion is high. Additionally, the positive correlation between adoption of practices and severity of erosion may also be an indication that these measures have little effect. This seems to indicate the need for training on the implementation of adequate techniques to combat soil erosion. Indeed, the effectiveness of an SWC measure will, in practice, depend very much on the suitability of the technique for a given form of water erosion (sheet, rill, gully) and how the SWC techniques are implemented (e.g., too great spacing between hedges …), or even by the approach adopted (issue of design versus environment, scale issue: field, hillslope or watershed).

Additional major constraints for specific SWC measures are cash and labour requirements as well as farm size and land tenure. These constraints have frequently been mentioned in other studies which addressed adoption issues of SWC technologies (e.g., Tenge et al., 2004; Birhanu & Meseret, 2013; Wildemeersch et al., 2013; Haghjou et al., 2014). Because of the small size of most farms (about 0.3 to 0.4 ha on average; Van Asten et al., 2013), farm size is seen as a constraint for technologies that compete for space with food crops (e.g., hedges and agroforestry systems). However, this does not necessarily have to be the case, as agroforestry systems and hedges may very well contribute to household food security and income if properly designed (Magcale-Macandog, Rañola, Rañola, Ani, & Vidal, 2010; Sharma et al., 2016). The fact that farm size is mentioned in relation to these technologies seems to indicate that current practices are not optimally designed. Nevertheless, the issue of farm size is likely to be exacerbated in the future because families are large (seven to eight children) and access to land happens essentially through inheritance, leading to further fragmentation (e.g., van Asten et al., 2013). Farmland ownership is a significant constraint for all technologies that involve long-term investments such as agroforestry, hedges and, to a lesser extent, drainage ditches, as has been reported elsewhere (Tenge et al., 2004). Besides the fact that such investments require approval by the landowner, there is the additional risk that, once the land has been improved, the owner will claim the parcel for his own use (Lunze, 2000).

In the study area, adoption rates mostly reflect farmers' own initiative compared to what has been observed in many other regions (with higher reported adoption rates) where adoption occurs in the context of multi-institutional programmes (e.g., Assefa & Hans-Rudolf, 2016; Rutebuka et al., 2019). Indeed, as a result of the persistent insecurity and weak government support, there have been very few soil conservation projects or programmes during the last decades. This also explains the near absence in our study region of dedicated soil erosion control measures (i.e., measures whose main purpose is erosion control such as ditches and terraces, as opposed to measures which may contribute secondarily to erosion control such as organic amendments or intercropping) that farmers would find too costly to implement without external support. In addition, adoption of dedicated SWC techniques by smallholder farmers is frequently limited by the lack of immediate returns on investment (Adimassu, Langan, Johnston, Mekuria, & Amede, 2017). However, in South-Kivu, because of the severity of the erosion process (Heri-Kazi & Bielders, 2020) and low inherent fertility of soils (Bitijula & Lal, 1983), this constraint may be less than in other environments. Indeed, it is expected that stopping erosion will more quickly lead to increased crop productivity, especially if fertility enhancing techniques are promoted alongside the SWC measures.

While dedicated SWC measures (e.g., drainage ditches, contour bunds, hedges) may be most effective at reducing soil erosion, the range of constraints associated with these techniques implies that their adoption will always be limited to certain categories of farmers (more motivated, owning larger farms, with external sources of income or having access to more labour, etc…). Hence efforts should not neglect existing practices whose main purpose is often not to control erosion but which, given proper advice, may significantly contribute to erosion control (e.g., organic amendments, surface mulching, intercropping). Such technologies are already widely adopted (98% for intercropping, 97% for spreading of household waste, 75% for compost; Table 4(b)). In fact, Roose (1993) among others alleged that farmer techniques could be just as effective as new technologies, provided that they are conducted efficiently. Nevertheless, reinforcing these practices also face constraints, in particular, in terms of availability of raw material for compost or mulching or in terms of transportation for household waste. Hence, depending on the farmer's resources and motivation, different solutions combining conventional techniques with dedicated erosion control measures may have to be proposed to the farmer, the so-called 'basket of best-bet technologies' as also proposed for soil fertility management (Giller et al., 2011).

The last finding of this study is that bench or progressive terraces were not reported or observed, even though they have been promoted by soil conservation projects in the region in the past two decades but also during the colonial period (Moeyersons, 1989). If well maintained, this technique can be very effective at controlling erosion and improving productivity (Bizoza & De Graaff, 2012). Besides the very heavy financial and labour constraints associated to this technique, the principal reason for their absence may be the lack of institutional support. Such institutional support is largely responsible for their widespread occurrence in neighbouring Rwanda, for instance (Bizoza, 2014).