Historical LULCC
Over the last three decades, the LKV catchment has witnessed considerable LULCC, driven by political, socio-economic, and demographic factors. Notably, the initial decade considered in this study (1990–2000) revealed the most substantial changes, characterized by a significant forest loss, and marked rise in urban and agricultural land. These changes are partly attributed to the early 1990s conflicts in the region, when forests became battle zones due to their strategic locations (Arakwiye et al., 2021; Plumptre, 2003). Additionally, the region witnessed major population movements in the aftermath of the 1994 genocide against Tutsis in Rwanda. This includes the return to Rwanda of people who had fled during the war and genocide periods from 1959 and the influx of refugees due to the civil wars in the Democratic Republic of Congo in the late 1990s and early 2000s (Bagalwa et al., 2016; Mugiraneza et al., 2019). These pressures led to an extensive conversion of forested land for settlement and agricultural expansion, to meet the growing demands for food and energy (Munanura et al., 2018; Rukundo et al., 2018). This period was defined by the fragmentation of natural forest parks, with Gishwati-Mukura in the LKV catchment losing more than half of its original land during the period (Kanyamibwa, 1998; Muhire et al., 2021).
In the second (2000–2010) and third (2010–2020) decades considered in this study, the region experienced contrasting trends in LULCC. The findings revealed a regrowth of forest cover and an equilibrium between expansion and reduction of agricultural land. This shift mirrors Rwanda’s strategic initiatives aiming at fostering sustainable environmental and natural resource management, transitioning towards a green economy. This commitment is reflected in the country’s Green Growth Strategy, focusing on addressing environmental challenges and land degradation. The increase in forest cover aligns with several of the country’s policies and strategic goals. These include the increasing forest cover to 30% by 2020 (Akinyemi, 2017) and developing the National Land Use Development Master Plan, which further optimized the land use across the landscape (Banerjee et al., 2020). Additionally, the results indicate the expansion of larger forest and grassland patches, particularly in the third decade, with occasional forest fragmentation amid agricultural land. This can be interpreted as a consequence of the country’s policy on land consolidation (Bizoza, 2021), following initiation of soil and water conservation measures such as agroforestry and terracing, to sustain the agricultural land.
Furthermore, the results of this study showed a continuous increase in built-up areas, with a peak of about 160% in the last decade compared to the second decade. The LULC evolution maps highlight this urban growth particularly in the North and Southern parts of the catchment. This trend coincides with the observed urban sprawl. Factors contributing to this expansion include rapid population growth, infrastructure development, and overall urbanization processes in these areas (Amisi et al., 2022). Additionally, there is a consistent decrease in the area covered by river, which can be attributed to measures taken for riverbank protection such as the planting of trees. It is important to note, however, that the results may not entirely reflect the reality, given that rivers within the LKV catchment are too narrow to be accurately detected using 30 m satellite image resolution.
The results of LULCC patterns in LKV align with the global trends observed in tropical catchments, where widespread deforestation and the expansion of anthropogenic activities were predominant in the decades leading up to the 2000s. However, noticeable reverse trend have been observed in the post-2000, with afforestation and reforestation efforts gaining momentum (Kayitesi et al., 2022). This change in direction is largely due to the implementation of various global and regional initiatives aimed at forest restoration and the promotion of sustainable land use practices including New York Declaration on Forests, Bonn Challenge, and African Forest Landscape Restoration Initiative (Dave et al., 2018).
Driving variables and Future LULC scenarios
The analysis of explanatory variables for LULCC revealed that factors including proximity to urban centres and population density play a significant role in most LULC transitions. This aligns with other studies, mainly in developing countries (Bongasie et al., 2024; Khwarahm et al., 2021), highlighting the role of urbanization and population growth in influencing LULC dynamics. Terrain slope emerges as another key factor, especially in transitions between agriculture, grassland, and forest categories, confirming the significant role of physical landscape in determining LULCC patterns, as supported by other studies (Akintuyi et al., 2021; Mandal et al., 2023).
In terms of agricultural land expansion, proximity to rivers stands as one of the primary drivers, underscoring the dependence on river water for irrigation, and the subsequent growth in agricultural areas along these water sources. This factor was also found to be influential for grassland expansion, associated primarily with pastureland, underlying the significance of water access for grazing and livestock needs, in agreement with findings from various studies (Najmuddin et al., 2018; Wassenaar et al., 2007). As revealed by (Li et al., 2021) soil organic content was specifically distinguished as a significant factor for agricultural land extension, while soil texture categories played a role for other LULC classes, though with less impact. While the presence of parks did not show significant impact on forest expansion, implying that factors other than their presence may shape forest changes, it is important to highlight that these protected areas effectively discourage encroachment by nearby communities (Riggio et al., 2019).
The predicted future LULCC based on three scenarios indicate distinct trajectories in the LKV catchment for both 2030 and 2050. The GGE scenario envisions a continuation of the existing land use trends in line with ongoing forest conservation efforts. It highlights evidence of positive environmental outcomes, notably in the promotion and protection of forests. This suggests that the existing policy framework in Rwanda is already effective in influencing LULC outcomes favourably (Bullock et al., 2021), which may not be as pronounced in other developing countries, as showcased by various case studies (Dietz et al., 2023; Khwarahm et al., 2021; Mungai et al., 2022). The EFP scenario further enhances these outcomes, aiming for increased environmental sustainability by preserving existing forested land and actively fostering forest expansion on other lands, alongside the protection of national parks. However, this scenario may pose a challenge to agriculture, which is crucial for food security (Bullock et al., 2021). Conversely, the DAA scenario presents the opposite situation, emphasizing the growth of urban and agricultural lands, potentially conflicting with ongoing forest conservation. Therefore, the GGE scenario emerges as an intermediate option, advocating for a balance between economic progress and environmental conservation.
Future LULC scenarios are anticipated to intersect significantly with various environmental impacts. Rwanda has experienced a 2.3% annual population growth rate between 2012 and 2022. According to the National Institute of Statistics Rwanda, the population is projected to reach 16.4 million by 2032 and 23.6 million by 2052 (NICR, 2023). Accommodating this increase in population will necessitate more space for settlement and agricultural land, potentially leading to the predicted DAA scenario. Such expansion could intensify environmental challenges, including floods, landslides, and soil erosion (Avashia & Garg, 2020; Remondi et al., 2016). Therefore, strategic measures are needed to mitigate these effects. Concurrently, research has demonstrated the effectiveness of green strategies like GGE and EFP in managing environmental hazards like floods and landslides (Locatelli et al., 2020; Nickel et al., 2014). However, these scenarios emphasizing sustainable development and conservation measures, could serve as counterbalance to the pressures of demographic expansion and urbanization. Moreover, the direction of future LULC will depend on a variety of different factors, including demographic trends, socio-economic developments, and the ongoing climate change.
Methodological considerations
This study employed an integrated approach, combining the capabilities of Google Earth Engine for LULC classification, coupled with Land Change Modeler for change detection and prediction of future LULC scenarios. The methodology involved the use of merged seasonal composites to capture the dynamic seasonal variations of the landscape. This, combined with the use of spectral indices such as NDVI, NDWI, and NDBI indices, along with topographic features like slope and elevation, significantly improved the discrimination and extraction of different LULC classes. This was particularly effective in the heterogenous and seasonally variable landscape of the LKV catchment. Future LULC scenarios were predicted based on the influence of natural and socio-economic driving variables on historical LULC class transitions. The Multi-Layer Perceptron neural network allowed the model to efficiently estimate the potential for various land use transitions, enabling predictions of future land use patterns.
Nevertheless, getting a sufficient number of satellite images, was a challenge, particularly in the 1990s and 2000s where the few available images were mostly clouded. This issue complicated the development of training sites due to unclear images, potentially impacting the accuracy of classified images from these periods. To address this challenge, Landsat images with a five-year span have been aggregated to represent each study period. Labels to train the model in supervised image classification relied on high resolution Airbus satellite images available in Google Earth Pro. Additionally, variations in illumination conditions in mountainous areas, influenced by the sun’s position, slope, and aspect, often resulted in inconsistent reflectance within the same LULC type, particularly on shaded slopes. For example, in the 2000 dataset grass areas in mountains were misclassified as agriculture in shaded regions. Various methods were attempted to address this issue, which ultimately proved ineffective. Consequently, the classification of these areas heavily depended on ground truth data, supplemented by the comparison with historical satellite images, and classification from previous and subsequent periods.
Modelling future LULC scenarios was based on the change rate observed between 2010 and 2020 and on the model calibrated over the previous periods. This rate was then projected to predict the transitions up to 2030 and 2050. However, these results have to be interpreted carefully, as the past trends do not necessarily reflect future changes if the surrounding conditions dramatically change in space and in time. The LKV catchment, faces various uncertainties in terms of change demand, partly due to conflicts in the African Great Lakes Region, which could result in unforeseen outcomes. Furthermore, in the developed scenarios, demographic factors such as natural birth and immigration rates were not directly integrated in the model. Nonetheless, their impact was implicitly reflected, particularly through the significant increase in the Built-up class observed from 2010 to 2020, representing a substantial rise compared to previous decades. Population density was additionally used as a proxy in the explanatory drivers in this study.
A potential way for enhancing the future prediction of LULC scenarios is to incorporate the Rwanda master plan for 2050, a document that was not accessible during the research phase. This master plan is a crucial policy document intended to guide the LULC development, and could significantly shape future landscape transformations. The integration of this plan into future LULC scenario modelling would substantially enhance the accuracy and applicability of future LULC scenarios.