Landslide Occurrence and Mitigation Strategies: Exploring Community Perception in Kivu Catchment of Rwanda

Abstract

Landslides are among the most significant disasters that threaten communities worldwide. This study sampled 384 respondents, using standardized interviews and field observations, to analyze how they perceived the factors influencing the incidence of landslides in the Kivu catchment of Rwanda, especially in landslide-prone areas. This study employs a mixed-methods approach that combines household surveys and interviews with key informants to assess how residents perceive landslide causes, warning signs, and impacts, which were analyzed statistically using SPSS. For further analysis, a binary logistic regression model and chi-square tests were used. The chi-square test findings highlighted that heavy rainfall, inappropriate agricultural practices, steep slopes, deforestation, road construction, earthquakes, and climate change were strongly correlated with landslide occurrence, with a p < 0.05 level of significance, while mining activities were not correlated with landslides. On the other hand, a binary logistic regression model revealed that, among the selected factors influencing landslide occurrence in the Kivu catchment, road construction (B = −0.644; p = 0.014), inappropriate agriculturalpractices (−1.177; p = 0.000), steep slopes (B = −0.648; p = 0.018), deforestation (B = −0.854; p = 0.007), and earthquakes (B = −1.59; p = 0.008) were negatively correlated, while heavy rainfall (B = 1.686; p = 0.000) and climate change (B = 1.784; p = 0.001) were positively correlated, and this was statistically significant for landslide occurrence at a p-value < 0.05. In contrast, mining activities (B = −0.065; p = 0.917) showed a negative coefficient that was statistically insignificant with respect to landslide occurrence in the study area. Future studies should integrate surveys with landslide hazard modeling tools for better spatial prediction of vulnerability and economic losses. Therefore, the findings from this study will contribute to sustainable natural disaster management planning in the western region of Rwanda.

1. Introduction

Landslides are the main natural disasters that affect mountainous areas worldwide, particularly in areas with high rainfall, steep slopes, and changes in land use induced by human activity [1,2]. They are ranked as the third most critical disaster globally, causing a high number of fatalities and significant property damage [3,4].

In recent decades, theories and assumptions have been proposed pertaining to the main factors driving landslide incidence [5,6]. The key drivers of landslides were found to be a complex combination of physical, geological, morphological, anthropogenic, climatological, and hydrological factors [7,8]. The other underlying drivers also include slope angle, internal relief, land use, and slope length [9]. Human activities were also found to have an accelerating role through deforestation, mining, and the clearing of slopes for buildings and roads [10].

However, different approaches are adopted to measure local communities’ perceptions of hazard risks. In support of this, the literature shows a variety of techniques, such as gathering information and insights from the community using physical disorder indicators [11]. Studies on landslides have been conducted in the Kivu rift, but have mostly concentrated on landslide risk dynamics modeling [12] and landslide susceptibility and associated factors [13]. Moreover, the literature suggests that it is very important to consider communities’ perceptions of their surroundings, as they greatly influence the socioeconomic activities that take place within them [14]. Studies on perception have considered numerous factors that influence communities’ perceptions of disaster hazards, and they revealed that knowledge, prior experiences, feelings on hazards, communication, readiness, information, and trust influence the way that communities perceive risk, such that they turn their beliefs into actions [15,16]. Furthermore, the main aspect of risk perception that can influence local residents’ behavior is their awareness of landslides [17].

Thus, indigenous local knowledge contributes to successful and practical landslide risk reduction, including identifying, evaluating, and tracking disaster risks, as well as improving early warnings for local disaster hazard reduction initiatives. Local expertise influences the large body of knowledge and skills that develop outside of formal education systems [18]. Local expertise has been traditionally entrenched in society. It supports and forms the basis for community planning and decision-making in areas including education, food security, natural resource management, and human and animal health.

For local communities, landslide factors are perceived depending on expertise and understanding of landslides [19]. On the other hand, the way that communities interpret landslide occurrence is accelerated by a number of complex elements, encompassing social, economic, personal, cultural, psychological, and political factors [20].

It has been argued that people who have recently experienced a landslide are more likely to recognize the related dangers and causes, and they exhibit greater sensitivity [21,22,23]. In this sense, preparedness supports community resilience and responses to landslide [24,25], especially when the risk is adequately explained and when confidence in the authorities is firmly established [26,27]. However, trust between scientists and local residents is essential for the effective application of landslide risk reduction methods, including early warning and emergency control [28,29]. In analyzing attitudes toward landslide risks, sociocultural influences have been found to be crucial [30,31], but most studies struggled to identify relevant and useful indicators to evaluate people’s perception and preparedness strategies regarding landslides.

In the same general context, most studies on landslides in Rwanda focused on assessing landslide susceptibility and mapping [32,33] and modeling landslide occurrence and related factors [34,35].

In addition, a limited number of studies examined how the Rwandan community perceived geo-hazards in general; however, data on landslides in particular, and especially for the Kivu catchment in Rwanda, is still lacking [36]. Various reports indicated that rainfall-induced disasters, including landslides, commonly occur in western Rwanda and cause infrastructure damage, economic losses, and fatalities. However, the risk associated with landslides is still not well understood [37,38,39].

Therefore, this study was conducted in various areas of the Kivu catchment that were directly impacted by landslide events in order to examine how the local community perceived landslide incidences and related adaptation strategies. Specifically, this study assessed community perception on landslide types and associated factors; its perception on landslide severity and likelihood; and community perception on landslide risk reduction strategies.

This findings of this study will contribute to the formulation of policies, which would be considered as a baseline for developing landslide mitigation strategies. Additionally, understanding how people perceive landslides will help the community to plan for development initiatives and appropriate landslide mitigation strategies. Furthermore, this study will function as a reference for future researchers who may be interested in conducting related studies in this distinct field, examining community perceptions of landslides and local adaptation efforts.

Several key landslide-prone areas have been identified in the Kivu catchment of Rwanda, requiring targeted risk mitigation efforts. The most critical selection criteria in these areas include a documented history of landslide occurrence, significant variations in elevation, diverse land use practices, and distinct settlement patterns.

2. Materials and Methods

2.1. Study Area

Situated in western Rwanda, the Kivu catchment is composed of many small lateral catchments that drain the west side of the divide between the Nile and Congo basins. This dividing line runs north–south in the west of Rwanda (Figure 1).

The Kivu catchment of Rwanda is situated at an altitude between 1451 m and 4483 m above sea level (Figure 2a), with a steep slope terrain ranging between 0 and above 120% (Figure 2b). It experiences an average annual rainfall of slightly above 1200 mm per year (Figure 2f), extends to 3469 km², and is covered by six districts, including Rubavu, Rutsiro, Rusizi, Karongi, Nyamasheke, and Nyabihu (https://waterportal.rwb.rw (accessed on 27 March 2023 and 10 July 2025)).

Additionally, the catchment’s lithology is composed of volcanic material, with quartz-rich schists found in fault zones in the south and center, and granite and basalt rock in the north and south (Figure 2e). The granite and pegmatite are found in the intermediate areas [40]. The predominant soil classes found throughout the catchment area are lixisol, alisol, apricot, and nitosol, which comprise pockets of cambisol and ferralsol. However, the andosol class is found in the north (Figure 2d). These soils typically have high rates of infiltration, which contrasts with the mineral and clay soils on flat terrain found in the north and central areas of the catchment (https://waterportal.rwb.rw).

Additionally, the catchment’s land cover is primarily composed of forested and farmed areas on hillside (Figure 2c). Natural open space for grazing and dairy production now makes up the majority of the former Gishwati Forest. The northern region (Park des Volcans), the Nyungwe Forest, and the remaining area of the Gishwati Forest are all designated natural forest areas. Benches, radical terraces, trenches, and contour plantations are implemented throughout the catchment for erosion protection. Built-up areas are mostly limited to the urban centers of Rusizi, Karongi, and Rubavu, and are fairly small in size (https://waterportal.rwb.rw).

2.2. Sampling Technique and Sample Size

The ability to gather a comprehensive set of information is necessary to enable the researcher to fulfill the study’s objectives. Thus, there is a need to select a sample drawn from the whole population [41]. The primary goal of a research survey is to provide an insight into how the findings drawn from a selected population can be extrapolated to the entire population [41].

In sampling analyses, the sample size can be controlled, but it is required to be optimal [42]. Although high-precision research requires a large sample size, smaller samples sizes can be chosen if the intention of a survey is only to gather information about research trends. The margin of error and the degree of significance of a study determines the necessary sample size [41]. A multistage random sampling design was adopted in the current study to guarantee a representative sample. The sampling proceeded through three steps, including systematic random sampling, stratified sampling, and purposive sampling. In the initial step, the Kivu catchment was chosen as the study location, both since it is known as one of the areas in Rwanda with landslide susceptibility, and due to the notable recent settlement growth in landslide-prone areas.

The second phase entailed selecting particular sampling locations inside the escarpment zone. These locations were chosen using stratified random sampling according to their level of landslide susceptibility, which was assessed by a number of criteria, including land use patterns, slope steepness, and proximity to known landslide-prone areas. In the third step, a systematic random sampling approach was used to choose and make a list of households to be interviewed. A substantial sample size of 384 houses was surveyed in the study, guaranteeing the validity and reliability of the findings. In addition to the structured interview, ten key informants were approached, including three district officers in charge of disaster management, a disaster management director at the Ministry in Charge of Emergency Management, two researchers at the University in Rwanda, and a senior consultant in catchment management and restoration.

In this study, the sample of respondents was chosen by employing systematic random sampling with a 95% level of confidence, with a 0.5-degree range of variation and a 5% margin of error as the accuracy level, while utilizing the formula established by Cochran [41], as described in the equation below (1).

$$n={\displaystyle \frac{z^2\stackrel{⃛}{p}\stackrel{⃛}{q}}{e^2}}={\displaystyle \frac{{\left(1.96\right)}^2\left(0.50\right)\left(0.50\right)}{{\left(0.05\right)}^2}} = 384 \mathrm{residents}$$

(1)

The required sample size is represented by n in this equation, along with p ⃛ (the estimated percentage of a feature in the population). In this study, we used p ⃛ = 0.5, since there was no previous research or trustworthy data that showed the percentage of local community members planning to implement landslide mitigation strategies in the Kivu catchment. This is a traditional and cautious assumption that is advised when the actual fraction is uncertain, since it produces the largest sample size and reduces sampling error. [1]. The q ⃛ = 1 − p ⃛ = 0.5, e is the acceptable error margin, Z is the statistical value indicating the level of confidence, and α is the value that researcher chooses to assess the statistical significance of the random sampling. This indicates a permissible likelihood of an error type [43].

2.3. Data Collection

First, various articles, journals and government reports were used to gather secondary information about landslide quantification, causes, impacts, and existing strategies for fighting disaster in the Kivu catchment. The catchment-level data in Rwanda were obtained from the RWB (Rwanda Water Resource Board) website (https://waterportal.rwb.rw). However, a survey questionnaire is employed to evaluate the perception of local communities on landslide occurrence adaptation measures in the Kivu catchment, where 384 respondents were randomly chosen to be interviewed. The structure of the questionnaire encompassed key sections: demographic information (e.g., age, gender, occupation), awareness and perception of landslides (e.g., frequency, causes, risk areas), impacts experienced (e.g., damage to property, livelihoods), and adaptation strategies employed (e.g., early warning systems and community initiatives).

The respondents were randomly selected from different locations of the catchment, based on their exposure to landslide risk, to gain practical and empirical understanding of how their community views the reasons and impacts of landslide, as well as their opinions and strategies on adaptation, and the effectiveness of landslide management in the Kivu catchment of Rwanda. In this regard, approximately 55%, 30%, and 15% of respondents were distributed in moderate, very high, and high-risk areas, respectively (Figure 3). Furthermore, interviews were also conducted with selected key informants including local leaders in charge of disaster management, environmental management professionals, and university researchers, particularly those who are involved in studies related to landslides.

Technically, during the data collection process, the Open Data Kit was employed for questionnaire digitalization, and the data were collected using tablets with KoboCollect v2922.2.3. The data were then cleaned and uploaded to SPSS software for quantitative analysis. This study’s survey design is detailed in Figure 4.

2.4. Data Analysis

In order to describe the socioeconomic characteristics of the community and their perceptions of the factors influencing landslides and various explanatory variables, the study employed descriptive statistics using SPSS (Version 20) in conjunction with a t-test, a chi-square test, and a model of binary logistic regression. In this study, various factors were analyzed: heavy rainfall, inappropriate agricultural practices, steep slopes, road construction, deforestation, mining activities, climate change, and earthquakes. All of these factors were selected based on their relevance to landslide occurrence, as proven in previous published works [44,45,46] (Figure 5).

The chi-square test is a statistical metric that is utilized in sampling analysis to evaluate the correlation between two attributes (variables) [47]. In SPSS, cross-tabulation and chi-square tests are widely jointly employed in statistical analysis. The chi-square test establishes the statistical significance of the observed associations, while cross-tabulation helps in data visualization.

Technically, the chi-square test is given in the following Equation (2):

$${\chi }^2={\sum }_{i=1}^n{\displaystyle \frac{{(O_I-E_I)}^2}{E_i}}$$

(2)

To test the hypothesis while using the chi-square test, in this study, two approaches are employed: the critical value approach and the p-value approach, compared via the level of significance/alpha ($\propto =5\%)$.

Ho: there is no correlation between the chosen independent parameter (factors influencing landslides) and dependent variable (landslide occurrence). This led to the following decisions: if χ² (calculated) is greater than χ² (critical), reject Ho; if the p-value $\le \propto$, then reject the null hypothesis; and if the p-value $> \propto$, fail the null hypothesis.

In this study, the binary logistic regression model was also utilized to test if there was statistical significance between landslide occurrence and the associated factors as perceived by the local community. It is called “binary” because it considers two values: 0 and 1, occurrence and non-occurrence. The predictable value is basically the probability p. In practice, the logistic transformation of p is used to indirectly model the dependent variable, as detailed in Equation (3) [48,49].

$$logit\left(P\right)=\ln \left({\displaystyle \frac{p}{1-p}}\right)=B_O+\left(B_1\times X_1\right)+\left(B_2\times X_2\right)+\left(B_3\times X_3\right)\dots (B_n\times X_n)$$

(3)

where B stands for odds p/(1 − p) and logistic regression model coefficients. The language of odds is employed more frequently than the language of probability in this binary logistic regression model.

In the binary logistic regression model, the dependent variable represents the occurrence of a landslide event, as reported by the respondent, within the past five years or near their household location. Respondents who reported having experienced or observed a landslide within this period were assigned a value of 1, while those who reported no landslide occurrence were assigned a value of 0. This dichotomous variable reflects the actual household-level experience of landslides, rather than the perceived likelihood or site classification. The five-year reference period was adopted to ensure a balance between recall accuracy and sufficient temporal coverage of landslide events. For the second logistic regression model examining community mitigation behavior, the dependent variable was defined as the adoption of any landslide mitigation measure (coded as 1 = yes, 0 = no) by the respondent. This allowed the assessment of socio-economic and environmental factors influencing households’ engagement in mitigation practices.

The independent variables used in the binary logistic regression model were derived from respondents’ perceptions of the key drivers of landslide occurrence. During the survey, respondents were asked whether they considered various factors such as heavy rainfall, steep slope cultivation, deforestation, road construction, and poor drainage systems to be major causes of landslides in their area. Each perception-based factor was recorded as a binary variable at the household level, and was coded as 1 if the respondent identified or agreed with the factor and 0 if not. This approach captures the individual’s awareness and perceived causal understanding of landslides. All variables were entered into the logistic regression model to assess how different perceptions of landslide causes relate to the reported occurrence of landslides and the adoption of mitigation measures.

This study also analyzed statistics on multicollinearity, which is a phenomenon when two or more predictors are correlated [50]. The fundamental assumption of factor analysis is that the observed variables are influenced by a smaller number of underlying factors. Multicollinearity can result in imprecise estimations of the relationships between variables, making it challenging to interpret the findings. As a result, incorrect inferences about the underlying structure of the data may be made. Most of the time, explanatory variables are intercorrelated and produce significant effects on one another. This relationship between explanatory variables compromises the results of multivariable regression analyses. The intercorrelation between explanatory variables is termed as “multicollinearity.” Tests for multicollinearity involve various methods, including tolerance, variance inflation factor (VIF), and condition index. Tolerance is a fundamental diagnostic statistic that is used to assess multicollinearity in multiple regression, serving as a direct indicator of how much unique information an independent variable contributes to the model. Tolerance is calculated for each predictor as follows:

$$Tolerance=1-R^2$$

(4)

where $R^2$ is the coefficient of determination obtained from regressing that particular predictor on all other independent variables in the model; it measures the proportion of variance in a predictor that is not explained by the other predictors. A tolerance value close to 1.0 indicates that the variable is almost perfectly independent, while a low tolerance value, typically below 0.10 (or equivalently, a variance inflation factor above 10), signals that the variable is highly redundant and that its relationship with the other predictors causes instability in the estimation of its regression coefficient. This means that if the tolerance is less than 0.1, there is a serious issue of multicollinearity in a dataset, namely if the fraction of variance in a predictor variable that is not shared by other predictor variables in a regression model is represented by tolerance, which is the reciprocal of VIF [51]. Regression analysis multicollinearity is measured statistically using the variance inflation factor (VIF). It measures the extent to which multicollinearity among predictor variables increases the variance of the predicted regression coefficients. According to Fox [52], the VIF of a predictor variable is determined as the difference between the estimated coefficient’s variance and the coefficient’s variance in the absence of multicollinearity. In a regression model, it is typically calculated for each predictor variable.

The formula for calculating VIF for the predictor variable is

$$VIF={\displaystyle \frac{1}{Tolerance}}={\displaystyle \frac{1}{(1-R^2)}}$$

(5)

where R² is the coefficient of determination for the predictor variable being analyzed [51].

The variance inflation factor (VIF) is a quantitative measure that is used to diagnose the severity of multicollinearity in a multiple regression model, specifically by quantifying how much the variance of an estimated regression coefficient is inflated due to linear dependencies with other predictor variables.

The condition index (CI) is a diagnostic measure that is used to quantify the severity of multicollinearity in a situation where the predictor variables in a multiple regression model are highly correlated by assessing the stability of the coefficient estimates [53].

A common rule of thumb is that a condition index exceeding 15 indicates a potential multicollinearity problem, values between 15 and 30 suggest moderate-to-strong dependencies, and values above 30 signify a severe multicollinearity problem. This means that, if the condition index is greater than 15, multicollinearity is suspected in datasets; on the other hand, if the condition index is greater than 30, there is a serious or strong issue of multicollinearity present in the datasets.

3. Results

3.1. Existing Landslides in the Kivu Catchment

The Kivu catchment, located in the African Great Lakes region, has experienced significant landslides, particularly in western areas with steep terrain, heavy rainfall, and unstable soils. The region’s high vulnerability stems from deforestation, unsustainable agricultural practices, and seismic activity linked to the East African Rift system. The Rwandan Ministry of Environment divides landslides into two categories (high and very high). Figure 6 shows the landslide-prone area, using the attribute data captured by Rwanda Environmental Management Authority in 2019.

3.2. Social Demographic Characteristics of Respondents

The findings of the SPSS examination for several social demographic attributes are displayed in

Table 1. Results for various social demographic characteristics in Kivu catchment (n = 385).

Variable	Frequency	Variable	Frequency
1. Gender		4. Age
Female	185 (48%)	25–40	150 (39%)
Male	199 (52%)	40–55	140 (37%)
2. Marital status		Above 55	35 (9%)
Divorced	4 (1%)	Below 25 years	59 (15%)
Married	271 (71%)	5. Education level
Separated	22 (6%)	None	13 (3%)
Single	87 (23%)	Primary	84 (22%)
3. Living period		Secondary	213 (56%)
10 to 20 years	131 (34%)	University	74 (19%)
Below 5 years	18 (5%)	6. Education field
Between 5 and 10 years	131 (34%)	Agriculture and environment	38 (10%)
More than 20 years	103 (27%)	Not applicable	143 (37%)
		Other	84 (22%)
		Social and health sciences	119 (31%)

. The proportion of respondents who provided answers to each question was used to perform statistical comparisons.

3.3. Perceived Landslide Types in the Kivu Catchment

Landslides, or mass wasting, encompass various types of downslope movement, categorized by the sort of substance, the movement mechanism, and the speed of the movement [54]. They include the following: falls, topples, slides (translational and rotational), spreads, and flows. Figure 7 presents the perceived types of landslides in the current study.

Figure 7 indicates the perceptions of the local community in the Kivu catchment, revealing that slides are the most prevalent, accounting for 282 (74% of the total 384 landslides); followed by flows with 69 instances (18%); falls with 28 cases (7%); and spreads, which are the least common, with only 5 occurrences (1%). This distribution highlights the dominance of slides in landslide events, suggesting that slope instability and shear failure are the primary mechanisms, while flows, falls, and spreads occur less frequently, possibly due to specific triggering conditions like heavy rainfall, steep terrain, or soil liquefaction.

For confirmation, field visits were conducted, and photographic evidence is provided in Figure 8.

3.4. Signs of Landslides in the Kivu Catchment

Table 2. Signs of landslides in the Kivu catchment.

Sign of Landslides in the Kivu Catchment	N	Percentage (%)
Slope cracks	341	62.80
Unexpected seepage	120	22.10
Unexpected springs	45	8.29
Cracks on flat ground	37	6.81
Total	543	100.00

presents the data on visible signs of landslides in the Kivu catchment, highlighting the most common indicators and their prevalence.

The most frequently observed sign is slope cracks, which are reported in 341 cases (62.80%), suggesting that slope instability is a dominant issue in the region. This high percentage indicates that erosion, soil saturation, or tectonic activity frequently causes ground fracturing, serving as a primary warning for potential landslides. The second most common sign is unexpected seepage (22.10%), observed in 120 cases, which may result from groundwater displacement due to soil movement. This seepage often precedes slope failures by indicating changes in subsurface water flow.

Less frequent but still notable signs include unexpected springs (8.29%) and cracks in flat grounds (6.81%), which are reported in 45 and 37 cases, respectively. Unexpected springs suggest significant shifts in hydrological patterns, possibly due to underground soil displacement. Meanwhile, cracks on flat ground, though less common, may indicate deeper-seated slope instability or subsidence.

Table 2 presents the observed signs of landslides in the Kivu catchment, revealing critical indicators of slope instability, with slope cracks (62.80%) the most prevalent, suggesting widespread ground deformation, likely due to erosion, soil saturation, or tectonic stresses. Unexpected seepage (22.10%) ranks second, indicating altered groundwater flow, often a precursor to slope failure. Less frequent but significant signs include unexpected springs (8.29%), pointing to subsurface water displacement, and cracks in flat grounds (6.81%), which may signal deeper soil movement or subsidence. These findings highlight the importance of monitoring terrain changes, as cracks and water-related anomalies serve as early warnings for landslides. The dominance of slope cracks underscores their reliability as a primary indicator, while variations in other signs suggest localized geological or hydrological influences, necessitating tailored risk mitigation strategies in vulnerable areas of the Kivu catchment.

3.5. Farmers’ Understanding of Factors and Consequences of Landslides

Landslide incidences in hilly areas are subject to numerous causal factors and triggered by various natural factors like precipitation, earthquakes, storm waves, water-level fluctuations, and rapid stream erosion [55].

Table 3. Community perception on landslide factors and consequences.

Landslide Factor	N	(%)	Rank
Heavy rainfall	384	57.66	1
Topography	237	35.59	2
Mining	23	3.45	3
Earthquake	22	3.30	4
Total	666	100

presents a detailed analysis of the parameters that contribute to landslides and their key consequences, based on a sample size of 666 incidents for factors and 781 outcomes for implications. The primary factor identified is heavy rainfall, accounting for 57.66% (384 cases) of landslides, followed by topography at 35.59% (237 cases). Mining and earthquakes are relatively minor contributors, at 3.45% (23 cases) and 3.30% (22 cases). This highlights the significant role of weather conditions and natural land formations in triggering landslides, with human activities, like mining, and natural phenomena, like earthquakes, playing a more minor, but still notable, role.

The consequences of landslides (

Table 4. Key consequences of landslides perceived by the community.

Key Landslide Consequence	N	(%)	Rank
Loss and damage of properties	368	47.12	1
Infrastructure damage	289	37.00	2
Injuries	77	9.86	3
Human deaths	47	6.02	4
Total	781	100

) are dominated by loss and damage of properties, which occur in 47.12% (368 cases) of incidents, ranking as the most frequent outcome. Infrastructure damage follows closely at 37.00% (289 cases), underscoring the substantial economic and societal disruptions caused by landslides. Injuries and deaths, while less frequent, still represent significant human costs, occurring in 9.86% (77 cases) and 6.02% (47 cases) of incidents, respectively. These findings emphasize the dual impact of landslides, affecting both material assets and human lives.

The ranking of consequences aligns with the severity and frequency of impacts, with property damage being the most common, followed by infrastructure disruption, injuries, and fatalities. This hierarchy suggests that, while landslides often cause extensive material damage, the immediate threat to human life, though less frequently observed, remains a critical concern.

3.6. Community Perceptions on the Existing Landslide Mitigation Strategies

Landslide control measures aim to prevent landslides or mitigate their impact.

Table 5. The existing landslide control measures in the Kivu catchment.

Existing Landslide Control Measure	N	Percentage
Agroforestry adoption	301	31.65
Relocation from high-risk zones	225	23.66
Terracing on hills	199	20.93
Storm water drainage systems	181	19.03
Others (specify)	23	2.42
Planting trees along riverbanks	22	2.31
Total	951	100.00

outlines various landslide control measures implemented in the Kivu catchment, as perceived by local communities.

Agroforestry emerges as the most widely adopted measure, accounting for 301 cases (31.65%), indicating its prominence in mitigating landslides through soil stabilization and erosion reduction. It is followed by relocation from high-risk zones (23.66%), with 225 cases, reflecting efforts to reduce human vulnerability by moving communities away from hazardous areas. Terracing on hills ranks third, with 199 cases (20.93%), demonstrating its importance in preventing soil erosion on steep slopes. These top three measures collectively represent over 76% of the interventions, highlighting their central role in landslide risk management in the region.

Secondary measures include storm water drainage systems (19.03%), with 181 cases, which help manage excess runoff and reduce water-induced slope instability. Though less frequent, the remaining measures, categorized as “Others” (2.42%) and planting trees along riverbanks (2.31%), still contribute to landslide mitigation. The low adoption rates for these measures suggest either their limited implementation or niche applicability in specific areas. The inclusion of tree planting along riverbanks underscores the recognition of the role of vegetation in stabilizing banks and reducing erosion, though its limited use may indicate challenges in scaling such interventions.

The total number of documented measures (951) provides a comprehensive overview of landslide control efforts in the Kivu catchment. The dominance of agroforestry, relocation, and terracing suggests a focus on preventive and nature-based solutions, while drainage systems address hydrological causes. The minimal representation of other measures may point to gaps in diversification or the need for context-specific strategies. Future efforts could explore scaling up underutilized measures, like riverbank afforestation, while strengthening integrated approaches that combine structural and ecological solutions for more resilient landslide mitigation.

4. Discussion

4.1. Perception of the Factors Influencing Landslides in the Kivu Catchment

Landslides in the Kivu catchment were perceived to be influenced by a combination of anthropogenic and natural factors, including steep slopes, heavy precipitation, road construction, deforestation, inappropriate agricultural practices, climate change, mining, and earthquakes (

Table 6. Factors influencing landslide occurrence in the Kivu catchment.

Factor	N	Percentage	Rank
Heavy rain	369	36.43	1
Steep slope	288	28.43	2
Road construction	107	10.56	3
Inappropriate agricultural practices	102	10.07	4
Deforestation	89	8.79	5
Climate change	26	2.57	6
Earthquake	19	1.88	7
Mining	13	1.28	8
Total	1013	100.00	263.80

). Together, these factors create conditions conducive to landslides, particularly during extreme weather events.

To analyze the factors influencing landslides in the Kivu catchment, statistical approaches including binary logistic regression and a chi-square test were used. The chi-square test, symbolized as χ², helped to ascertain whether there was a significant correlation between two categorical parameters [47], like landslide occurrence and influencing factors. In the current study, the [χ² (determined)] was calculated by utilizing the degree of freedom [df = (r − 1) (c − 1)] and significance degree (α = 0.05) in comparison with the chi-square value [χ² (critical)] from the table.

Table 7. Significance of landslide incidences and their influencing factors.

Variable	χ2 Calculated	χ2 Critical	df	p-Value	χ2 Test
Heavy rainfall	40.335	3.841	1	0.000	SS
Inappropriate agricultural practices	25.982	3.841	1	0.000	SS
Steep slope	10.605	3.841	1	0.001	SS
Road construction	7.593	3.841	1	0.006	SS
Deforestation	22.837	3.841	1	0.000	SS
Mining activities	0.003	3.841	1	0.958	NS
Climate change	5.597	3.841	1	0.018	SS
Earthquake	5.352	3.841	1	0.021	SS

SS = the landslide occurrence is statistically significant for the selected influencing factor, NS = the landslide occurrence is not statistically significant for selected influencing factor,

presents the test results of the chi-square test, indicating the association between landslide occurrence and the chosen influencing factors in the Kivu catchment.

Meanwhile, binary logistic regression allows researchers to model the likelihood of a landslide occurring based on predictive factors, quantifying their influence while controlling for confounding factors [56,57] (

Table 8. The binary logistic regression model results.

Variable	B	S.E.	Wald	df	Sig.	Exp. (B)	95% C.I for EXP(B)
							Lower	Upper
Heavy rainfall	1.686	0.289	34.11	1	0.000	5.395	3.065	9.499
Inappropriate agricultural practices	−1.177	0.313	14.174	1	0.000	0.308	0.167	0.569
Steep slope	−0.648	0.274	5.584	1	0.018	0.523	0.305	0.895
Road construction	−0.644	0.262	6.04	1	0.014	0.525	0.314	0.878
Deforestation	−0.854	0.319	7.18	1	0.007	0.426	0.228	0.795
Mining activities	−0.065	0.623	0.011	1	0.917	0.937	0.276	3.18
Climate change	1.784	0.547	10.651	1	0.001	5.951	2.039	17.371
Earthquake	−1.59	0.599	7.052	1	0.008	0.204	0.063	0.659

). This method can rank variables by odds ratios, indicating which factors most significantly increase landslide risk.

While determining the influence of the eight explanatory factors (landslide influencing variables) on landslide incidence, the following terms and definitions were applied [56,57]: B: regression coefficient in the binary logistic regression model. Exp. (B): odds ratio. S.E.: standard error. Sig.: p-values. Wald: a Wald chi-square test was employed to find out if the coefficients in the model were significant statistically. df: degree of freedom (for the Wald chi-square test), and 95% C.I. for EXP(B) = confidence interval for odds ratio.

Various studies have mentioned natural and anthropogenic factors influencing the occurrence of landslides [55,58]. The results and interpretation of the analysis of the impacts of the eight independent factors on the dependent factors (landslide occurrence) using the binary logistic regression model and the chi-square test are outlined in the following summary.

Table 6 indicates that heavy rainfall is most significant factor that triggers landslides in the Kivu catchment, amounting to 36.43%. Also, the chi-square test reveals an existing significant relationship between heavy rainfall and landslide occurrence in the Kivu catchment (χ² = 40.335, df = 1; p = 0.000). In the current study, binary logistic analysis showed that heavy rainfall correlated positively with landslide occurrence and was significant at the 0.05 level (B = 1.968, p-value = 0.000), whereas the Wald test results (34.11) also demonstrated the same significance.

Various studies confirmed that precipitation can raise pore water pressure and lower soil shear strength, which leads to the existence of landslides [59]. One study found that landslide events are easily triggered during rainstorms, heavy rainfall, or continuous rain [58]. Moreover, ref. [60] indicated that the landslide incidences are directly caused by intensive precipitation. In the current study, we analyzed the temporal variability in rainfall and connected landslide occurrence over a period of 17 years in the Kivu catchment. Figure 9 indicates a marked increase in the frequency of landslide incidence in the Kivu catchment, which is especially notable from 2022 to 2024 compared to previous years. Most landslide incidence is recorded during the months of March, April, and May, which aligns with the long rainy season in Rwanda (March–May). A smaller number of landslide incidents occur in September and October, corresponding to the short rainy season. The frequency of landslides increased dramatically in 2023 and 2024, particularly in March, April, and May. Landslides were inconsistent and not concentrated in any one month during earlier years (2007–2021). The short rainy season was also found to be associated with an increase in landslides in September and October; however, these are less frequent than during the main rainy season in the Kivu catchment (Figure 9). Landslide incidence is rare during the dry season (June, July, August, December), indicating that heavy rainfall is a significant trigger for landslides in the region. In brief, the frequency of landslides in the Kivu catchment is directly related to the country’s bimodal rainfall pattern, where the risk is higher during the long rainy season (March–May) and lower during the short rainy season (September–October). The recent surge in landslides may be the result of shifting rainfall patterns, changes in land use, or other environmental variables that make the area more vulnerable to landslides during the rainy season.

Based on the findings presented in [61], landslide risk is likely to increase in areas characterized by an inappropriate use of agricultural practices, which reveals that the land is not well-covered, and thus subjected to easy runoff facilitated by the geographical and climatic features of the landscape, which in turn generate mudslides and landslides.

The results of the current study underscore the influence of inappropriate agricultural practices on the occurrence of landslides, which was cited by 28.43% of respondents. In addition, the results are confirmed by the statistical methods, and the chi-square test results showed a statistically significant association between inappropriate agricultural practices and landslide occurrence in the Kivu catchment (χ² = 25.982, df = 1; p = 0.000). Furthermore, the binary logistic regression analysis indicated that inappropriate agricultural practices have a significant and negative influence on landslide occurrence in the study area (B = −1.177, p-value = 0.000); the Wald test results (14.174) revealed the significance.

Slope influences the occurrence of landslides by affecting the sliding force and stress distribution of the rock–soil. For slopes with homogeneous and isotropic material composition, the possibility of landslide incidence increases with the increase in slope [62,63].

The current study indicates a correlation between landslides and steep slopes, as perceived by local communities. Advanced statistical analysis through a chi-square test revealed that the occurrence of landslides in the Kivu catchment is significantly influenced by steep slope at (χ² = 10.605, df = 1, p = 0.001). This was confirmed by the binary logistic regression results, which were negatively correlated with the landslide occurrence and statistical significance at B equal to −0.648, p-value equal to 0.018, in accordance with the Wald statistics (5.584). In the study [64], the authors stated that the runoff velocity rises in the area with very steep slope, and the volume of surface water also rises, resulting in a high risk of landslide incidence. Moreover, a study conducted by [65] depicted that slope was a major contributing element to landslides, with a correlation coefficient of 0.895.

The distance to the road can impact the influence of anthropogenic activity on landslide occurrence. Since the construction of roads in the study area necessitates the excavation of the mountain, reducing the strength of the rock–soil and exposing most of the slopes, landslide disasters are easily induced during periods of heavy rainfall [58].

This study discovered a significant relationship between road construction and landslide incidence; indeed, the chi-square test results show that the occurrence of landslides in the Kivu catchment is strongly correlated with road construction at χ² = 7.593, df = 1, p = 0.006. The binary logistic regression results are also significantly and negatively correlated with landslide occurrence (B = −0.644, p-value = 0.014), according to the Wald test results (6.04). This is contrary to the common assumption that steep slopes, and slope cutting due to road construction, increase landslide susceptibility. However, one key informant’s statement, “In hilly regions of Rwanda, especially in areas with known steep slopes and along major roads, the government, its partners and local community have implemented measures such as retaining walls, drainage channels, slope terracing, and vegetation reinforcement to stabilize slopes and minimize erosion and landslide risk”, reveals that even though these steep areas are geomorphologically vulnerable, the presence of mitigation structures has reduced the actual occurrence or perceived frequency of landslides. This aligns with the findings who suggested that slope management and protective infrastructure can substantially lower the probability of landslide occurrence in high-risk terrains [28]

Deforestation and forest clearance have been linked to an increase in landslide activity due to long-lasting root decomposition processes [66]. Furthermore, the chi-square test analysis revealed that the occurrence of landslides in the Kivu catchment is significantly influenced by deforestation, at χ² = 22.837, df = 1, p = 0.000. The binary logistic regression results are also significantly negatively correlated with the landslide occurrence (B equal to −0.854, p-value equal to 0.007), in accordance with the Wald test results (7.18). Deforestation increases the likelihood of landslides by disrupting the stability of slopes [67]. Since tree roots regulate slope stability, it is well known that deforestation increases the incidence of landslides [68].

Climate change significantly influences landslide occurrence through a complex interplay of factors, primarily by intensifying precipitation patterns, altering soil moisture levels, and increasing the frequency of extreme weather events [69]. Moreover, through a chi-square test, this study applied statistical analysis, and the results show that the occurrence of landslides in the Kivu catchment is significantly influenced by climate change, at χ² equal to 5.597, df equal to 1, and p equal to 0.018. The binary logistic regression results are also significantly positively correlated with landslide occurrence (B equal to 1.784, p-value equal to 0.001), in accordance with the Wald test results (10.651). The results of various studies have indicated that there is an indirect relationship between climate change and landslide development [70,71].

Mining activities, particularly open-pit and underground mining, can significantly increase the danger of landslides [72,73]. These activities alter the natural balance of ground stress, weaken slopes, and create new surfaces that are susceptible to movement. The study conducted by [64] indicated that potential landslide zones were found in a mining area. In contrast, the results of this show that there is no direct correlation between mining activity and the incidence of landslides; the chi-square test analysis revealed that the occurrence of landslides in the Kivu catchment is insignificantly influenced by mining activities, at χ² = 0.0030, df = 1, p = 0.9580. The binary logistic regression results are also negatively correlated with landslide occurrence and with statistical insignificance at 5% (B = −0.065, p-value = 0.917), as confirmed by the Wald test results (0.011).

The results of this paper are in line with [74], in which the authors stated that the results of correlation between mining activity and the existence of landslides showed that mining activity has no direct association with landslide occurrence, as they have a correlation coefficient that is nearly zero. According to the literature, mining activities in the Kivu basin increase landslides [75]. The results on the weak association between mining and landslides may be explained by the following reasons: the mining activities in the Kivu catchment may be performed at a very small scale and a limited extent, where their effect is minimal compared to other landslide-induced factors such as heavy rain and steep slope. Additionally, mining may be indirectly implicated through other landslide- inducing factors, such as deforestation or small-scale settlements, but not perceived by the local community. In this study, mining was found not to be directly associated with landslides, and this is attributed to the environmental protection practices that are currently being adopted in mining areas in Rwanda, including those located in the Kivu catchment. This is evident in the key informant interview with a district officer, who stated that “actually, the mining site owners have adopted mitigation practices such proper cut slope angles, drainage systems, benches, vegetation buffer, slope stabilization, and many more; hence contributing to the reduction of landslides risk. The district collaborates with the central entities in law reinforcement, and set sanctions to the companies who do not adopt these practices, and the results were remarkable in these past five years”.

Earthquakes, caused by the devastation of rock–soil formations in and around the fault zone, are the primary indicator of how local tectonic conditions affect landslides. This reduces the slope’s integrity and serves as a crucial channel for groundwater, leading to the slope’s deformation and destruction [58]. Furthermore, the chi-square test analysis revealed that earthquakes significantly influenced the occurrence of landslides in the Kivu catchment (χ² = 5.352, df = 1, p = 0.021). The binary logistic regression results are also significantly negatively correlated with landslide occurrence (B = −1.59, p-value equal to 0.008), in accordance with the Wald test results (7.052).

Landslides can occur in earthquake-prone areas due to a variety of variables, including geography, geology, and rainfall [72]. While not all earthquakes lead to landslides, studies indicate that a significant portion of large-scale landslides are directly or indirectly induced by seismic activity [73].

Lastly, the basis of this research was the following question: “are there any significant factors influencing the occurrence of landslides in the Kivu catchment”? To answer this question, the authors of this paper utilized a binary logistic regression model and a chi-square test. For this case, the eight independent variables—heavy rainfall, inappropriate agricultural practices, steep slopes, road construction, deforestation, climate change, and earthquakes—were selected as the major factors strongly influencing landslide occurrence, whereas mining activities were not. In this study, heavy rainfall and climate change were positively correlated, while inappropriate agricultural practices, steep slopes, road construction, deforestation, and earthquakes were negatively correlated with landslide occurrence. However, mining activities did not show a significant association with the existence of landslides in the Kivu catchment.

The logistic regression model indicated negative coefficients for variables such as deforestation, inappropriate agricultural practices, road construction, steep slopes, and earthquakes. While this might initially appear to contradict their established roles as physical triggers of landslides, the negative coefficients must be interpreted within the perception-based context of this study. Each variable represents whether a respondent perceived the factor as a major cause of landslides (coded 1 = yes, 0 = no). Hence, a negative coefficient does not imply that these factors physically reduce landslide likelihood, but rather that respondents who had not directly experienced landslides were less likely to perceive these causes as significant. Conversely, respondents reporting recent landslides tended to emphasize other causes, particularly heavy rainfall, as the dominant triggers. This finding reflects variations in community perception rather than contradictions in physical processes, underscoring the need for risk education that integrates scientific understanding with local experiences.

4.2. Multicollinearity Analysis

In this study, the multicollinearity analysis employed three methods, including tolerance, variance inflation factor (VIF), and condition index, to test the relationship between dependent (occurrence of landslides) and independent variables (e.g., heavy rainfall, inappropriate agricultural practices, steep slopes, road construction, deforestation, mining activities, climate change, and earthquakes). Based on scientific assumptions, the following conditions are set: (1) the tolerance value must be higher than 0.2; (2) the variance inflation factor must be lower than 5; and (3) the condition index value must be lower than 15. This means that if tolerance is less than 0.1, there is a serious issue of multicollinearity in the dataset if the fraction of a predictor variable’s variance that is not shared by other predictor variables in a regression model is represented by tolerance, which is the reciprocal form of the VIF [51]. On the one hand, this means that, if the condition index is greater than 15, multicollinearity is suspected in datasets, and on the other hand, if the condition index is greater than 30, there is a serious issue of multicollinearity being present in datasets. Prior to regression analysis, multicollinearity among independent variables was assessed using tolerance, variance inflation factor (VIF), and condition index (

Table 9. Summary of multicollinearity statistics.

Model	Collinearity Statistics
	Tolerance	VIF
Heavy rainfall	0.937	1.068
Inappropriate agricultural practices	0.822	1.216
Steep slope	0.818	1.223
Road construction	0.844	1.185
Deforestation	0.832	1.202
Mining activities	0.969	1.032
Climate change	0.967	1.034
Earthquake	0.927	1.079

,

Table 10. Coefficients with detailed multicollinearity statistics.

Model	UC		ST	Exp(B)	Sig.	Collinearity Statistics
	B	S.E	Wald			Tolerance	VIF
Heavy rainfall	0.33	0.051	0.296	6.423	0.000	0.937	1.068
Inappropriate agricultural practices	−0.217	0.056	−0.191	−3.87	0.0.00	0.822	1.216
Steep slope	−0.132	0.052	−0.126	−2.545	0.011	0.818	1.223
Road construction	−0.124	0.049	−0.124	−2.545	0.011	0.844	1.185
Deforestation	−0.162	0.058	−0.137	−2.804	0.005	0.832	1.202
Mining activities	−0.009	0.125	−0.003	−0.074	0.941	0.969	1.032
Climate change	0.298	0.09	0.15	3.306	0.001	0.967	1.034
Earthquake	−0.332	0.107	−0.144	−3.114	0.002	0.927	1.079

UC: unstandardized coefficients, ST: standardized coefficients.

and

Table 11. Collinearity diagnostics.

Dimension	Eigenvalue	Condition Index	Variance Proportions
			(Constant)	Heavy Rain	Inappropriate Agricultural Practices	Steep Slope	Road Construction	Deforestation	Mining	Climate Change	Earthquake
1	3.674	1	0.00	0.02	0.02	0.02	0.02	0.02	0	0.01	0.00
2	1.066	1.86	0.00	0.00	0.03	0.00	0.00	0.03	0.13	0.05	0.53
3	1.011	1.91	0.00	0.04	0.01	0.00	0.00	0.02	0.64	0.04	0.13
4	0.932	1.99	0.00	0.18	0.12	0.00	0.01	0.19	0.07	0.12	0.08
5	0.851	2.08	0.00	0.13	0.00	0.00	0.01	0.00	0.04	0.76	0.05
6	0.548	2.59	0.00	0.45	0.05	0.03	0.33	0.09	0.03	0.01	0.03
7	0.464	2.81	0.00	0.00	0.74	0	0.00	0.59	0.01	0.00	0.0
8	0.282	3.61	0.1	0.13	0.03	0.34	0.63	0.05	0.03	0.00	0.1
9	0.172	4.63	0.9	0.05	0.00	0.61	0.00	0.01	0.04	0.00	0.07

). All tolerance values exceeded 0.2, all VIFs were below 5, and all condition index values were below 15, indicating no significant multicollinearity in the current study. Thus, all variables were retained in the final logistic regression model.

Variance proportions are a crucial component in multicollinearity analysis, particularly when used in conjunction with the condition index, as they move beyond merely detecting the presence of collinearity to diagnosing specific damaging effects on the model’s coefficients [53]. These proportions, derived from variance decomposition in the coefficient covariance matrix, reveal how the variance in each regression coefficient is distributed across the various dimensions of instability identified by the condition indices. In essence, for a high condition index indicating a near-dependency among variables, the variance proportions show which specific regression coefficients are being adversely impacted by that shared dimension.

A key diagnostic rule is that a high condition index accompanied by two or more independent variables (each having large variance proportions above 0.50) in the same dimension pinpoints the specific set of variables involved in a harmful collinear relationship, thereby identifying which coefficient estimates are rendered unstable and statistically unreliable [51]. According to Table 11, a result less than 0.50 indicates a strong relationship between the variables.

It is important to note that both the dependent variable (landslide occurrence) and most independent variables (perceived causal factors such as rainfall, deforestation, and slope cultivation) were derived from respondents’ perceptions. Consequently, the statistical associations identified in the logistic regression model should be interpreted as relationships between community perceptions, rather than as determinants of actual landslide occurrence. This perception-based approach was intentionally adopted to explore how local residents conceptualize landslide causes and link them to their lived experiences, an essential component of community-based disaster risk reduction. Similar perception-driven frameworks have been applied in previous studies in the East African highlands (e.g., [75,76]) that recognize that community understanding of risk factors strongly shapes mitigation behavior, irrespective of the physical accuracy of those perceptions. Therefore, the findings of this study should be viewed as reflecting associative patterns in community belief systems, which provide valuable insights for designing contextually relevant awareness and policy interventions.

The statistical findings of this study provide several actionable insights for disaster risk management and land use planning in the Kivu catchment. The high odds ratios for heavy rainfall and climate change indicate the importance of integrating rainfall monitoring and early warning systems into community disaster preparedness programs. The negative and significant associations for deforestation, inappropriate agricultural practices, and road construction reflect community recognition of human-induced vulnerability, suggesting that planners should enforce land use zoning, promote agroforestry and terracing, and ensure that road development incorporates slope-stabilization measures. Furthermore, since perceptions were found to align closely with mapped landslide-prone areas, these insights can support participatory hazard mapping and risk-awareness campaigns that leverage local experience in planning. Thus, the statistical results have clear practical relevance for designing context-specific, community-driven mitigation and adaptation strategies in the highlands of Rwanda.

4.3. Study Limitations

Even though this study is valuable, due to the fact that it is based on survey and community perception, it may be affected by bias, as well as less emphasis on measured factors (hydrological, geotechnical, and geological) to verify perceptions; therefore, this limits the ability to quantitatively link causes or predict future landslides. Due to the fact that the data were gathered at specific locations and times while landslide risk and adaptation strategies are dynamic, perceptions may change, and this can limit the long-term validity of the results. Furthermore, the study did not integrate the survey with landslide hazard modeling tools, which could allow for better spatial prediction, and this has made the recommendations more qualitative than quantitative.

5. Conclusions

In this study, we assessed how the local community perceived landslide occurrence and associated risk reduction measures. Among the monitored factors influencing landslide incidence in the study area, the respondents identified rainfall as the main factor, followed by steep slopes. Local communities shared their perception of the impacts of landslides.

Furthermore, different factors influencing landslide occurrence were assessed using SPSS, along with the chi-square test and the binary logistic regression model. The chi-square test showed that heavy rainfall, inappropriate agricultural practices, steep slopes, road construction, deforestation, climate change, and earthquakes were associated (p < 0.05) with landslide occurrence in the Kivu catchment, while mining activities were unassociated with this. In addition, the binary logistic regression model indicated that road construction, inappropriate agricultural practices, steep slopes, earthquakes, and deforestation were negatively correlated with landslide occurrence. In contrast, heavy rainfall and climate change were positively correlated with landslide occurrence, while mining activities are negatively correlated, and did not significantly influence landslide occurrence.

Thus, the local community reported that there are existing measures for mitigating landslides in the study area, but they are not effective. Finally, this study suggests that further research is needed to evaluate the different elements influencing the involvement of the local community in the planning process, and how to implement the existing landslide control measures and structure in the Kivu catchment. The government should provide incentives and an updated early-warning system for local communities and organize training and workshops to educate them on how to respond to landslides at an early stage.