Our study is a national-level analysis with specific details on refugee-hosting districts. Rwanda is one of the most densely populated countries in Africa, given its small territorial size and growing population. The land-locked nation with a landmass of 26,338 square kilometers (2,633,800 ha) bordered Uganda (north), Burundi (south), DRC (west), and Tanzania (east). Rwanda is among the fastest-growing economies in Africa, with an annual GDP growth rate of 8.2 percent in 2023 (Figure A3). The national poverty rate fell from 58.9 percent to 38.2 percent between 2000 and 2017 (Bagstad et al., 2020).

Figure 1 provides a map of our study context and illustrates Rwanda’s watersheds. Precipitation patterns vary across provinces and districts in Rwanda. Rwanda has two distinct wet seasons: heavy rains regularly occur during the long wet season (March to May), while rainfall varies geographically during the short rainy season (October to December). The western parts of the country receive more precipitation events than the eastern locations. The highest monthly rainfall amount is often recorded in April. In 2020, for instance, April precipitation ranged from 155 mm in the eastern area to 269 mm in the western region (Fick and Hijmans, 2017). In Rwanda, global warming has reduced rainy days with a marked increase in the frequency and intensity of torrential rainfall events. Unlike percipitation, evapotranspiration in Rwanda is relatively higher in the eastern than in the western region (Brooks et al., 2013).

Our assessment also incorporates land cover, elevation, and slope. The land cover is one of the most critical factors for assessing and understanding the hydrological dynamics for any given watershed area. For instance, highly vegetated or forested areas gain more infiltration capacity during precipitation events. Infiltration capacity represents the permeability of water from surface areas into the ground, potentially contributing to baseflow and recharging groundwater aquifers, which serve as a primary source of clean drinking water. Croplands dominate a vast area for Rwanda’s vegetation cover and significantly enhance national food security and food sovereignty conditions. Similarly, with a mountainous topographic, Rwanda’s elevation levels range from 1,525 to over 4,500 meters above global mean sea level. The highest altitude areas are found in the northwestern and southwestern regions (Nsengiyumva et al., 2018).

Rwanda is currently home to over 130,000 refugees. Its principal refugee-sending countries are the Democratic Republic of the Congo (DRC) (62 percent) and Burundi (37 percent) (UNHCR, 2024b). A meaningful share of the Congolese refugee population has been displaced in Rwanda since the 1990s, a consequence of the First Congo War. Ongoing violence in the Democratic Republic of Congo has fueled continual population displacement into Rwanda. The Burundian refugees mainly arrived in the country following the post-2015 election-induced tensions and conflicts in Burundi (UNHCR, 2021a).

The Government of Rwanda (GoR) is a signatory of the 1951 Refugee Convention and other related protocols and agreements, and the Ministry in Charge of Emergency Management (MINEMA) oversees the protection of refugees and management of refugee response (MINEMA, 2021).

Rwanda upholds progressive refugee-related policies: refugees are not required to live in camps, and those who do reside in encampment areas have free mobility. Encamped refugees receive food assistance from the World Food Program (WFP), and all refugees receive cash assistance for nonfood items (UNHCR, 2021b, 2024c). Those living in camps obtain healthcare and schooling through UNHCR and humanitarian partners. As of 2019, urban-based refugees have access to the national Community Based Health Insurance scheme (Karinganire, 2022). Rwanda guarantees access to free primary and secondary schooling for all refugee children, and those living in or near camps can obtain UNHCR support for school meals and uniforms (UNHCR, n.d.). Importantly, refugees can work legally (MINEMA and World Bank, 2019), though they face social and market barriers to employment, resulting in a high employment rate (Bilgili and Loschmann, 2018).

A small fraction (about 10%) of refugees live in urban areas, mainly in Kigali City and Nyamata (MINEMA and World Bank, 2019). Presently, the majority of refugees reside in one of the following refugee camps: Nyabiheke, Kiziba, Kigeme, Mugombwa, and Mahama. Prior to 2021, Rwanda also operated the Gihembe camp in Gicumbi district. Gihembe was notably closed due to natural hazard threats, but since it was operating during the development of this study, and given its value as a comparison to still-active sites, we keep the Gicumbi camp in our analysis.

The locations of all six camps are shown in Figure 1, and we provide satellite imagery of the camps in Figure 2. Mahama is the largest, accommodating 47% of the total refugee population as of 2024. Kiziba and Kigeme Camp residents each account for 11% of the refugee population, while Nyabiheke and Mugombwa each represent about 9% of the population (UNHCR, 2024a). In this study, the 15 km area is identified as the exposure extent of refugees and host communities to various hydro-climatological risks. This buffer area usually goes beyond the official boundary of a single administrative district or a hydrologically delineated watershed extent. Focusing on the 15 km radius around camps aligns with UNHCR policy frameworks in identifying the refugees’ host community area to support resource allocation prioritization for refugee and host-community support.

The geospatial data included in this analysis were primarily GIS and remotely sensed layers from various sources. Incorporated map layers capture land cover (LC), elevation, precipitation, evapotranspiration, hydrological soil groups, the normalized difference vegetation index, and compound topographic index (Table 1). Compound Topographic Index (CTI) also called the Topographic Wetness Index (TWI), measures how water accumulates in a landscape based on slope and upstream drainage area. Higher values indicate areas prone to saturation, making it useful for identifying flood-prone and landslide-susceptible zones. Normalized Difference Vegetation Index (NDVI) derived from land cover satellite imagery that measures vegetation health and density. To generate the NDVI layer, we used Band 4 and 5 of Landsat 8 to calculate the NDVI (Band 5 – Band 4)/(Band 5 + Band 4) and eventually incorporated into spatial multi-criteria flood risk analysis. NDVI helps assess land cover changes, flood vulnerability, and drought conditions, as greener areas absorb more water and reduce runoff. Drought Index (scPDSI—Self-Calibrating Palmer Drought Severity Index) is a measure of long-term drought severity, adjusted for different climate zones. This index combines precipitation, evapotranspiration, and soil moisture data to track dry conditions, particularly relevant for monitoring water scarcity risks in refugee-hosting districts. Hydrologic Soil Group (HSG) is a classification of soil types based on infiltration and runoff potential, critical for flood and erosion modeling. Soils are grouped from A (high infiltration, low runoff) to D (low infiltration, high runoff), helping predict areas prone to waterlogging or surface runoff. Digital Elevation Model (DEM) is a 3D representation of terrain elevation, derived from Shuttle Radar Topography Mission (SRTM). DEMs are used for mapping watersheds, modeling landslides, and assessing soil erosion risk by identifying steep slopes, drainage paths, and elevation changes.

Previous studies have identified and used these thematic layers for assessing flood, drought, and erosion risks in various locations [Ahmadisharaf et al., 2016; Chitsaz and Banihabib, 2015; De Brito et al., 2018; Kanani-Sadat et al., 2019; Khosravi et al., 2019; Kowalski et al., 2009; Rahmati et al., 2016; Shadmehri Toosi et al., 2019; Tang et al., 2018; World Wildlife Fund (WWF), 2016]. As described below, most of the data used are publicly available and required limited pre-processing, as included in our modeling, except for the LULC map, which required pre-processing and image classification.

To examine Rwanda’s land cover, we used Landsat 8 spectral imagery with less than 10 percent cloud cover (30 m resolution). To avoid temporal data mismatches while controlling for seasonal effects, we restricted our spectral imagery sample to data captured in January and February 2020/2021, falling between Rwanda’s long and short-wet seasons. Due to atmospheric distortions from cloud cover, we could not use finer resolution land cover data from the European Space Agency’s (ESA) Copernicus satellite, Sentinel-2 A/B (10 m resolution).

The land cover (LC) class identification and training process was informed by previous LC maps created by the Regional Center for Mapping of Resources for Development [RCMRD, 2015; Regional Center for Mapping of Resources for Development (RCMRD), 2013], following the RCMRD (2015) methodology, classes with less producer’s accuracy were thoroughly considered during the training stage of image processing.³ The image processing, including the training sample generation and map creations, was carried out using the ArcGIS Pro software (Version 2.4.0, 2019). We used a pixel-based supervised classification method for image processing based on the random forest algorithm. The land cover map (Figure 3) is a key input for the InVEST Sediment Delivery Ratio (SDR) model, as it determines how different land cover types influence soil erosion and sediment transport (see methodology section).

This study uses flood hazard mapping based on a spatial multi-criteria decision-making (MCDM) approach using applications of the analytical hierarchy process (AHP), geographic information system (GIS) techniques, and remote sensing algorithms. These methods have been widely applied and evaluated as quite useful in informing decision-making processes, specifically in limited or no-data regions in many developing countries (Rahmati et al., 2016). The spatial MCDM method assesses, evaluates, and integrates multiple layers to inform a flood risk modeling process (Feizizadeh and Blaschke, 2013). The spatial MCDM method is one of the most widely used data-driven statistical models for prediction science, including hazard mapping and risk assessment (Khosravi et al., 2019).

The AHP approach, one of the MCDM techniques, has the most widespread application for flood vulnerability assessment and risk mapping, according to a thorough review of 128 papers using different MCDM methods (Khosravi et al., 2019). The MCDM tool has been applied to numerous recent flood susceptibility studies around the world (Khosravi et al., 2019; Tang et al., 2018; De Brito et al., 2018; Ahmadisharaf et al., 2016; Rahmati et al., 2016). Similarly, Shadmehri Toosi et al. (2019) incorporated a hydrological model (SWAT) with the MCDM technique to generate comprehensive potential flood hazard maps for northeast Iran.

The AHP technique assigns weights to each input data layer, as suggested by Saaty (1980) (see also Al-Hanbali et al., 2011; Collins et al., 2001; Feizizadeh and Blaschke, 2013). The causal factor determines the relative importance of each map layer when compared to others (Bunruamkaew and Murayama, 2011; Dampha, 2020). In this study, each data type generates an input map layer included in the model as a contributing causal risk factor for assessing the overall biophysical and socio-economic vulnerability of Rwandans and refugees to fluvial and pluvial flood risks. The spatial MCDM model evaluates and harmonizes tradeoffs among various input factors in relation to their flood risk susceptibility scores and weights (Table 2). For example, high-altitude locations might be considered less flood-prone, all else constant. Studies confirm that “at higher elevations, an increase in rainfall can actually reduce the occurrence of flooding” (Nkwunonwo et al., 2020). However, the LULC composition (e.g., highly impervious surfaces) of the same elevation site with a flat terrain could expose the location to a higher flood risk level.

We map these contribution risk factors in Figure 4. After processing all input layers and assigning their corresponding weights, we used ArcGIS’s reclassify spatial analyst tool to categorize the raster values into five different classes, ranging from lowest (1) to highest (5) risk and vulnerability levels (Table 3). We used the natural breaks method in dividing the feature classes into their “natural groupings” using the ArcGIS Pro software. Evidence shows that the severity of each reclassified flood risk contributing factor should be based on the characteristics of the individual input feature (Tran et al., 2009). The more significant the layer’s contribution, the higher the assigned flood risk score and weight (Table 2). For instance, locations near rivers and stream networks were considered highly exposed as they are the most likely to be affected by fluvial flood events. Also, vegetated areas are likely less prone to a particular flood event than areas with a lot more impervious surfaces, especially when the infiltration rate is highest and stormflow (i.e., direct runoff) volume is at its lowest velocity. Similarly, locations with good road networks and less travel time distance are considered less vulnerable in an extreme flood event than places with difficulties in relocating residents.

The flood risk map generated from the spatial MCDM modeling approach is an aggregated overlay of weighted cells representing all reclassified input map layers. The weighted overlay tool in ArcGIS Pro was used for running the final analysis. The flood risk mapping is mathematically defined as:

In the above equation, R is the risk composite score, W i is the weight assigned to the factor i, X i is the score of factor i, Π is a multiplicative sum of the constraints, and C j represents constraint j score (0 or 1). For a pixel area, if the constraint factor is zero (meaning not at risk), the composite risk score of that area becomes zero, resulting in the exclusion of the site from the analysis.

Landslides are one of the deadliest types of natural hazard event that occurs in Rwanda. Nsengiyumva et al. (2018) conducted a nationwide landslide susceptibility assessment in Rwanda (Figure 5). Similar to our flood modeling approach, they applied the Spatial Multi-Criteria Evaluation Model (SMCE) using eight input map layers identified as causal factors of landslide occurrences. These include slope, distance to roads, lithology, precipitation, soil texture, soil depth, altitude, and land cover (Nsengiyumva et al., 2018). Landslide incidences are primarily attributed to extreme and intense precipitation events (Ministry in charge of Emergency Management (MINEMA), 2019; Nsengiyumva et al., 2018). Landslide incidences in Rwanda are most frequent during the two wet rainy seasons (March to May) and (October to December). Nsengiyumva et al. (2018) landslide modeling results were validated by 980 past landslide-affected locations based on Rwanda’s Ministry in Charge of Emergency Management (MINEMA) data. Similar studies have supported using the causal factors included in Nsengiyumva et al. (2018) landslide susceptibility assessment (Duc, 2013; Tegeje, 2017; Kato and Mutonyi, 2011; NASA Earth Observatory, 2010; Uwihirwe et al., 2020). For example, similar studies attributed severe landslide events in Uganda (2010) and Vietnam (2005) to extreme precipitation hazards (Duc, 2013; Kato and Mutonyi, 2011; NASA Earth Observatory, 2010).

To avoid duplication of research efforts, we used the landslide susceptibility map from Nsengiyumva et al. (2018) to inform our target refugee-hosting districts’ exposure and vulnerability to landslide hazards. In addition we compiled the most recent MINEMA records of past landslide-reported disasters from 2013 to 2021 for all refugee-hosting districts in Rwanda. Using these data, we evaluate the effects of landslide incidences on the loss of lives, destruction of livelihood sources, and damage to private properties and public infrastructure in refugee-hosting districts.

Using disaster panel data from Rwanda’s Ministry in Charge of Emergency Management (MINEMA), we conducted a district-level risk mapping of recorded disaster impacts from 2016 to 2020, mainly validating our modeling results. The mapping is based on the assessed losses of human lives, injuries, destruction to private properties and public infrastructure, and damage to people’s livelihood options. The disaster impact mapping focuses more on refugee-hosting districts using protracted time-series data from 2013 to 2021 (Table A1). The 2016–2020 dataset was later imported into the ArcGIS pro software for additional analysis (i.e., from vectorization, rasterization, reclassification, weighted aggregation to map visualization). The combined disaster risk map is weighted because the impacts of disaster types cannot be considered equal. For example, a disaster impact resulting in loss of human lives was weighted higher than impacts leading to damage to public infrastructure such as schools, markets, etc.

We use a widely applied spatially explicit model, the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) Sediment Delivery Ratio (SDR) or Sediment Rentention, to map out soil erosion dynamics in Rwanda (Figure 6). Developed by the Natural Capital Project, the model assesses soil erosion impacts in our study area (Sharp et al., 2020). As applied in this study, the InVEST SDR model offers relevant information about soil erosion, sediment export, sediment deposit, and potential sediment retention. For the purpose of this analysis, our results focuses on soil erosion effects at watersheds scale (30 meter spatial resolution) with a particular focus on camps areas – i.e., within a 15 km radius from the centroid of the refugee camps. A 30-meter resolution provides a balance between detail and computational efficiency. This resolution captures fine-scale changes in terrain, vegetation, and land cover, relevant for detecting erosion hotspots, crucial for targeted erosion control measures. We ran the InVEST SDR based on the same parameters used in Bagstad et al. (2020) study of accessing Rwanda’s ecosystem accounts (Table A2). The methodological details of these models can be accessed here (Sharp et al., 2020).

The InVEST SDR model’s soil erosion as illustrated in Figure 6 is centered around estimating soil loss due to rainfall and runoff using the Revised Universal Soil Loss Equation (RUSLE). The model calculates annual soil loss for each pixel within the study area by considering factors such as rainfall intensity (R Factor), soil susceptibility to erosion (K Factor), topography (LS Factor), land cover (C Factor), and soil conservation practices (P Factor). These factors impact soil erosion dynamics, highlighting the interaction between land-use management choices and natural processes.

For instance, landscape conditions such as topography or terrain represent a crucial role in influencing erosion risk, represented by the LS Factor, which accounts for slope length and steepness. For example, steeper slopes or longer slopes tend to accelerate more soil loss, increasing the erosion potential. On the other hand, human management choices represented by the C and P factors account for the effects of vegetation cover and erosion control measures, respectively. A low C value, for example, indicates more vegetative protection, leading to lesser soil loss, while effective soil management practices reflected in a low p value help reduce erosion.

The model outputs a raster map showing the potential annual soil loss across the landscape, highlighting areas more vulnerable to severe erosion, with a zoom into refugee campus areas. These erosion hotspots as shown below provide valuable insights for conservation planning and land management practices through vegetation cover, contour farming, or other erosion control measures. This detailed pixel-based approach at 30-meters resolutions offers deeper analysis of localized erosion risks which could better inform targeted interventions in refugee camp surroundings. However, the model performs well in capturing overland erosion processes such as rill and inter-rill erosion and does not perform well for detecting gully erosion or streambank erosion.

We also included the drought index data, based on the self-calibrating Palmer Drought Severity (scPDSI), which is globally available at a spatial resolution of 0.5 degrees. The dataset offers salient information on drought-prone locations, particularly among refugee-hosting districts. The scPDSI data used spans from 1901 to 2020, driven by the preliminary version of the CRU TS4.05 monthly climate dataset. The scPDSI is determined based on time series data on precipitation and evapotranspiration. It also incorporates parameters reflecting the characteristics of a location’s soil and surface (Barichivich et al., 2020; Van Der Schrier et al., 2013).

In Rwanda, our study demonstrates high risks to various climate change impacts across encampment areas, though the type of risk varies (Table 4). The country’s current physical flood risk is relatively high. Approximately 18% of the country’s total land area falls under the ‘highest’ and ‘high’ flood-prone categories (Figure 7). Additionally, 64 percent of the national territory currently falls under areas classified as “moderate” flood zones (Figure 8). In urban areas like Kigali, the capital city, flood risk exposure is highly significant, due to the increased presence of impervious surfaces and relatively large human settlement areas within the city. The rising exposure of agricultural lands to flood risks in rural areas poses additional threats to household food security, especially for those engaged in subsistence agriculture.

Physical flood risk levels are equally high in refugee-hosting districts and within a 15 km radius from the refugee camps’ centroid (Figure 9). Among these locations, Mugombwa and Mahama refugee camp areas exhibit the highest physical flood risk potential. In Mugombwa and Mahama, an estimated 26,234 ha (39%) and 16,711 ha (24%) of the landscape within the 15 km radius of camps’ centroids, respectively, are classified as ‘high’ flood-prone areas. Flood exposure levels for both refugees and host communities are also relatively higher in the Mugombwa refugee-hosting district than in the Mahama refugee camp locations. The next highest flood hotspot among all refugee-hosting sites is the Nyamagabe district, which houses the Kigeme refugee camp. Of the total land area (70,630 ha) within the 15 km radius from the Kigeme refugee camp’s centroid, 17% (approximately 11,971 hectares) is classified as high flood risk zones, affecting both settlements and cropland areas.

Using the landslide risk mapping from Nsengiyumva et al. (2018) (Figure 10), we find that Rwanda’s western and northern regions are highly prone to landslide hazards compared to the eastern parts of the country. This past assessment suggests that landslide risk is highest in the Kigeme and Kiziba camp hosting areas. But the landslide risk predictions in Nsengiyumva et al. (2018) do not classify the Gihembe encampment area as facing high risk. Disaster damage data shows that landslides are additionally of high concern in this region. These contrasting findings highlight the potential shortcomings of risk prediction methodologies, demonstrating the importance of triangulating evidence whenever possible.

Figure 11 provides information on landslide related damages. Between 2013 and 2021, landslide disasters resulted in 28 deaths in the Kiziba refugee-hosting district (Karongi) and 11 deaths in the Gihembe refugee-hosting district (Gicumbi). Comparatively, refugee-hosting communities in the eastern and southern parts of the country, like Kirehe (Mahama refugee camp) and Gisagara (Mugombwa refugee camp), have recorded no deaths or casualties since 2013. Regarding damages to residential houses, landslide disaster impacts have either partially or completely destroyed 207 private properties and 23 public infrastructure facilties in the Kiziba refugee camp district between 2013 and 2021. Similarly, 85 private houses and 21 public infrastructure facilities were affected in the Gihembe refugee camp district. Moreover, landslides resulted in damages to croplands, undermining rural livelihood sustenance for both refugees and their host community members. For instance, in the Gihembe refugee-hosting district, 324 hectares of croplands and 104 units of livestock were damaged by or lost to landslides.

The recorded effects of climate-related disasters in Rwanda sharply increased in 2018 and 2019. The 2018 year exhibits very high disaster impacts on croplands (food security source), properties (houses) and human casualties (deaths and injuries) (Figure A1). Figure 12 maps the weighted recorded impacts of weather and climate-induced disasters by administrative districts in Rwanda from 2016 to 2020.

The disaster impact results reveal that among all refugee-hosting districts that fall within a 15 km radius areas from the camp’s centroid, the Mahama camp area has experienced the most recorded damages. For instance, between 2013 and 2021, over 7,767 ha of croplands were damaged in this location. Additionally, 3,897 houses were damaged and destroyed within the 15 km buffer distance area around the Mahama refugee camp (Figure 13).

To aggregate the data by hazards of interest to this study, we added the disaster risk impact findings from the following precipitation-related events (i.e., floods, rainstorms, and lightning) and landslides (as mentioned above) for only refugee-hosting disasters. First, on precipitation-related hazards such as pluvial flooding, rainstorms, and lightning, the vulnerability of the Mahama refugee camp district remains significantly high relative to the recorded disaster impacts in other refugee-hosting communities. The second most affected refugee-hosting district from precipitation-related damages is Gisagara, where the Mugombwa refugee camp is located. The results support our flood risk modeling conclusions that Mugombwa and Mahama refugee camps are the most vulnerable hotspots to physical flood risk. Indeed, they remain the most impacted by precipitation-related disasters, as shown in Figure 14.

As mentioned above, our results focuses on soil erosion effects at watersheds scale (30 meter spatial resolution) with a particular focus on camps areas – i.e., within a 15 km radius from the centroid of the refugee camps. The result shows that the Kiziba refugee campsite watershed boundary has the greatest soil erosion issues, with an estimated an annual soil loss of upto 19 million tons within the watershed area. Comparatively, yearly total soil losses are lowest in Mahama and Mugombwa refugee camps areas (< 41,000 ton/watershed). Our model also suggests that soil erosion risks are moderately in the Kigeme and Gihembe encampment regions (Figure 15). This corresponds to the recent development of deep ravines in Kigeme, which have motivated population relocations and reforestation initiatives (Karinganire, 2023). The findings could enable decision-makers to prioritize sustainable land and watershed management plans around camps.

Using the drought index map from the University of East Anglia, UK [self-calibrating Palmer Drought Severity Index (scPDSI)] (Barichivich et al., 2020), it appears that Rwanda’s drought risk is highest in its southeastern region (Map 17). Hence, we can state with some confidence that the Mahama refugee camp is located in an area that is highly prone to climate-related drought risk. Given the elevated flood risks we identified in the Mahama encampment area, our work suggests that this area is vulnerable to both extreme lows and highs in terms of precipitation. Mugombwa camp sits close to the western extent of the southeastern drought-prone region. While Mugombwa exhibits rather low risk of landslide, soil erosion, or flood, our drought risk assessment suggests that droughts may be the one risk type this encampment area may be susceptible to. Other encampment areas by contrast, have rather low drought risk profiles (see Figure 16).

Our study demonstrates that climate-induced hazards vary spatially across refugee-hosting districts in Rwanda. The hilly and densely populated country is heavily deforested (REMA, 2011), which has augmented the risk of damaging landslide events (Vodacek, 2021). Without sufficient mitigation and adaptation measures, such climate hazards could reduce the safety and livelihood security of all residents of Rwanda. Such hazards are particularly dangerous for vulnerable households, that may struggle to cope with climate risk impacts, including refugees and poor households in nearby host communities.

Rwanda also faces high flood risk in many areas (Mind'je et al., 2019), as well as drought risk, particularly in the Eastern Province (Niyonsenga et al., 2024). Our findings reveals that flood risk is highest in southeastern locations, particularly in districts hosting Mugombwa and Mahama refugee camps. Disaster impact data show significant precipitation-related damages in Mahama’s home district in recent years. Similarly, landslide risk is particularly high in the former Gihembe camp area, as well as in the western districts hosting Kiziba and Kigeme camps. Between 2013 and 2021, landslides in the district hosting Gihembe camp damaged 324 hectares of cropland, while in 2016 alone, landslides in Kiziba camp’s district claimed 110 lives and injured 52 people (MIDIMAR, 2016).

These risks have influenced recent Government of Rwanda (GoR) and UNHCR policies, such as the closure of Gihembe camp in 2021 due to landslide risks (UNHCR Rwanda, 2021). Additionally, Rwanda’s southeastern region faces increasing drought risks, which will likely persist due to climate change-driven precipitation variability. Between 1974 and 2018, an estimated 4.2 million people were affected by droughts in Rwanda (United Nations Office for Disaster Risk Reduction (UNDRR), 2020). Our analysis indicates that Mahama and Mugombwa refugee camps are particularly vulnerable to droughts and water scarcity.

Soil erosion is a critical challenge in Karongi district, where Kiziba refugee camp is located. Using the InVEST Sediment Delivery Ratio (SDR) model, we find that sediment delivery significantly increased in all encampment areas, with Kiziba exhibiting the highest levels. Soil erosion has reduced agricultural productivity in many parts of the country (Kagabo et al., 2013). Refugee camps often contribute to increased deforestation for firewood and shelter materials, reducing vegetation cover and accelerating soil erosion and sediment transport (Food and Agriculture Organization of the United Nations (FAO), 2022). The loss of tree cover weakens slope stability, exacerbating landslide risks and increasing sediment loads in nearby rivers and streams, which threatens water quality (Emadi-Tafti et al., 2021). Also, a large share of Kigeme camp’s population has been relocated to Mahama due to erosion effects and the resulting deep ravines (Karinganire, 2023). In response, reforestation efforts are also underway. For example, in Kigeme camp, tree planting undertaken by refugees and host community members intends to reduce landslide vulnerability (Karinganire, 2023). Our study highlights the ongoing hazard threat in many encampment areas that continue to support large refugee populations: with one exception (Nyabahike), each has at least one climate risk of concern, if not more.

In terms of refugee integration, Rwanda has a progressive asylum policy that aligns with international refugee protection frameworks, including the 1951 Refugee Convention and the OAU 1969 Refugee Convention (Ministry in charge of Emergency Management (MINEMA), 2021; Organization of African Unity, 1969). The government has also taken steps to mainstream refugee needs and concerns into national development plans, particularly through its National Strategy for Economic Inclusion of Refugees and Host Communities (2021–2024) (Ministry in Charge of Emergency Management (MINEMA) and United Nations High Commissioner for Refugees (UNHCR), 2021). This approach integrates refugees into public services, health care, and education, reducing dependence on humanitarian aid. The country provides refugees with the right to work reflecting its commitment to inclusive policies. As a signatory to the Global Compact on Refugees (GCR), Rwanda actively promotes self-reliance and socio-economic inclusion of refugees through initiatives like the Comprehensive Refugee Response Framework (CRRF) (Global Compact on Refugees, 2025). In addition to its social protection systems, the Rwandan government has taken steps to address the threats of climate-related disasters. As a signatory of the United Nations Framework Convention on Climate Change (UNFCCC), Rwanda is advancing strategies to reduce greenhouse gas emissions and build resilience (Republic of Rwanda, 2020). Rwanda has adopted the Sendai Framework of Action for disaster risk reduction (2015–30) and has invested in educating Rwandans on natural and climate hazard risks (Nahayo et al., 2018). Rwanda has mainstreamed refugee affairs and disaster risk reduction (DRR) strategies into some of its national sectoral policies and district development plans such as National Disaster Risk Reduction and Management Policy (MINEMA, 2023), National Contingency Plan for Storms (MINEMA, 2018b), Revised Green Growth and Climate Resilience Strategy (GGCRS) (Republic of Rwanda, 2022) Additionally, the country has developed regulations and contingency plans to mitigate, respond, and recover from natural and climate-related disasters. Rwanda also maintains a national disaster loss database to monitor impact. The database supports analytical scientific studies like ours to inform public policy (United Nations Office for Disaster Risk Reduction (UNDRR), 2018).

Despite these national and local efforts, many people in Rwanda remain highly vulnerable to climate change impacts. For a large share of Rwandan households, livelihoods are heavily dependent on economic activities that are the most vulnerable to negative shocks related to natural hazards. This is especially the case for agriculture, but also applies to those engaged in forestry, tourism, and fisheries. Over the past two decades, the frequency and intensity of natural and climate-induced disasters, particularly floods and droughts, increased significantly, leading to an increase in the toll of deaths and injuries as well as economic and environmental losses (United Nations Office for Disaster Risk Reduction (UNDRR), 2020). Natural hazard damages have also resulted in increasingly negative consequences for agricultural production, as well as augmenting infrastructural damages (Ministry of Disaster Management and Refugee Affairs (MIDIMAR), 2016, 2017; Ministry in charge of Emergency Management (MINEMA), 2018a, 2019, 2020).

Our analysis acknowledges limitations due to scale mismatches and data availability constraints. We integrate high-resolution hazard models (30 m–100 m) with district-level disaster records, ensuring spatial coherence, but camp-specific disaster data remain limited. The multi-criteria GIS modeling approach involves some subjectivity in factor weighting, which, while informed by previous research, would benefit from further ground-truth validation. Additionally, while broad-scale drought indices (e.g., scPDSI) provide regional and district-level insights, they may not fully capture localized water stress within refugee settlements.

Future research should prioritize high-resolution monitoring using drone imagery, LiDAR, and Sentinel-2 data to better track land cover changes in encampment areas. Improving spatial disaggregation techniques for disaster impact data would refine linkages between district-wide hazards and refugee camp risks. Finally, integrating scenario-based climate projections (CMIP6) and hydrological models could strengthen long-term risk assessments. Finally, policy-driven research on nature-based solutions and infrastructure adaptation could support resilience-building efforts in refugee-hosting areas.