Karteren en modelleren van ravijnerosie op continentale schaal van Afrika

Gully erosion is a major driver of land degradation and a key cause of irreversible soil loss worldwide. Understanding and quantifying this process is therefore essential for designing effective land-use and conservation strategies. However, our ability to predict gully erosion - especially at larger scales - remains limited. This is due to both the lack of suitable predictive models and the scarcity of datasets that capture gully erosion processes, particularly in generally vulnerable and data-scarce regions like Africa. Moreover, as a result of rapid population growth and climate change, the pressure on land in these regions is expected to increase significantly. The aim of this research is to address these research gaps by compiling a comprehensive database of gully locations and characteristics and developing models that can predict both the occurrence and activity of gully erosion across Africa at high spatial resolution. The thesis consists of three main phases, each contributing a different model component. In the first phase, a model was developed to predict the occurrence of gully heads (i.e. the vertical walls at the upslope end of the gully channel). For this, we assembled a new, extensive dataset of over 31,000 manually mapped gully heads across more than 1,200 sites of 1 km x 1 km on the African continent. All gully heads were identified by trained experts using high-resolution Google Earth imagery. This dataset served as the basis for building a logistic regression model that predicts Gully Head Susceptibility (GHS) at a 30-meter resolution using publicly available GIS data. Results indicate that higher gully densities are primarily found in areas with steeper slopes, medium-textured soils, and low vegetation cover and, to a lesser extent, in areas with high rainfall intensity and tectonic activity. The model performs well (AUC ≈ 0.8) and provides the first high-resolution gully head susceptibility map for Africa. Although this data-driven model is robust and practically applicable, it is based solely on statistical relations and does not explicitly describe the physical processes that drive gully initiation. This limits its generalizability, especially under changing environmental conditions. To address this, the second phase of the research explores a more process-based modelling approach using the well-established Curve Number (CN) method. This method estimates runoff volumes based on land use, rainfall, and soil type. We introduced a new Gully Head Initiation (GHI) index, which represents the ratio between runoff-induced shear stress and the soil's resistance to erosion. Applied at both the continental scale and within two small catchments in the Ethiopian Highlands, the GHI model showed moderate predictive performance (AUC ≈ 0.65 - 0.67). While it underperforms relative to the statistical GHS model, the GHI index has the advantage of being easily interpretable and transferable, and offers a conceptually grounded description of gully initiation. This highlights the potential of more process-based approaches for large-scale gully erosion modelling. After initiation, gullies can further expand through the upslope migration of the gully head. The total amount of soil loss caused by gullies therefore not only depends on where gullies form, but also on which gullies are actively expanding. Since not all gully heads remains active, identifying the active gully heads is crucial for estimating total soil loss. In the third and final phase of this research, the focus therefore shifted to assessing gully head activity. By comparing high-resolution imagery over a 10-year interval, we identified more than 900 active gully heads (i.e., those showing more than 10 meters of upslope retreat). A separate model was developed to predict the susceptibility to active gully head occurrence. The results show that active gullies are more likely to occur in areas with lower slopes, fine-textured soils, high rainfall intensity, and recent cropland expansion - revealing significant differences between spatial patterns of inactive and active gully head occurrence. Although the Active Gully Head Susceptibility (AGHS) model shows moderate performance (AUC ≈ 0.66), its combination with the GHS model from phase one offers a preliminary, but promising tool for identifying active gully erosion hotspots across Africa. Together, these models and the resulting maps offer valuable tools for understanding, predicting, and managing gully erosion in Africa. They demonstrate the value of integrating data-driven and process-based approaches and mark a first step toward large-scale, integrated erosion control. The findings can support policymakers and land managers in assessing erosion risks across regional and national borders. Moreover, because gully initiation and activity are strongly influenced by land use (change) and rainfall intensity, this study underscores the sensitivity of gully erosion to environmental change. At the same time, it lays the groundwork for future modelling efforts to incorporate climate and land use change scenarios to inform proactive soil conservation strategies across vulnerable, rapidly changing landscapes.