A data driven gully head susceptibility map of Africa at 30 m resolution

Gullying and landscape evolution: Lavaka in Lac Alaotra, Madagascar shed light on rates of change and non-anthropogenic controls

Gully evolution remains poorly understood, largely because multidecadal analysis is lacking. Large gullies (called lavaka) that pepper Madagascar's highlands are generally attributed to human impact; but longitudinal data are few, and anthropogenic causation is inferred not verified. We focus on Lac Alaotra, Madagascar's largest lake and wetland, its major rice-growing region, and an ecological hotspot surrounded by fault-controlled steep hills with abundant lavaka. Analysis of historical aerial photographs and recent orthoimagery reveals that the proportion of highly active lavaka has decreased since mid-20th century. At the same time, human population, farming intensity, and livestock density have increased exponentially. This suggests that background factors, including seismicity, are primary drivers of lavaka formation. Although human activities may contribute to erosion overall, land management policies that overemphasize human causation of gullying and neglect background forcing factors (in Madagascar and elsewhere) are unlikely to be effective in erosion mitigation.

How the EU Soil Observatory contributes to a stronger soil erosion community

New policy developments have emerged in relation to soil conservation after 2020. The Common Agricultural Policy (CAP) 2023-2027, the proposal for a Soil Monitoring Law and the mission 'A Soil Deal for Europe' have shaped a new policy framework at EU level, which requires updated assessments on soil erosion and land degradation. The EU Soil Observatory (EUSO) successfully organised a scientific workshop on 'Soil erosion for the EU' in June 2022. The event has seen the participation of more than 330 people from 63 countries, addressing important topics such as (i) management practices, (ii) large scale modelling, (iii) the importance of sediments in nutrient cycle, (vi) the role of landslides and (v) laying the foundations for early career scientists. As a follow up, among the 120 abstracts submitted in the workshop, we received fifteen manuscripts, out of which nine were selected for publication in the present special issue. In this editorial, we summarize the major challenges that the soil erosion research community faces in relation to supporting the increasing role of soils in the EU Green Deal.

Interpretability of simple RNN and GRU deep learning models used to map land susceptibility to gully erosion

Gully erosion possess a serious hazard to critical resources such as soil, water, and vegetation cover within watersheds. Therefore, spatial maps of gully erosion hazards can be instrumental in mitigating its negative consequences. Among the various methods used to explore and map gully erosion, advanced learning techniques, especially deep learning (DL) models, are highly capable of spatial mapping and can provide accurate predictions for generating spatial maps of gully erosion at different scales (e.g., local, regional, continental, and global). In this paper, we applied two DL models, namely a simple recurrent neural network (RNN) and a gated recurrent unit (GRU), to map land susceptibility to gully erosion in the Shamil-Minab plain, Hormozgan province, southern Iran. To address the inherent black box nature of DL models, we applied three novel interpretability methods consisting of SHaply Additive explanation (SHAP), ceteris paribus and partial dependence (CP-PD) profiles and permutation feature importance (PFI). Using the Boruta algorithm, we identified seven important features that control gully erosion: soil bulk density, clay content, elevation, land use type, vegetation cover, sand content, and silt content. These features, along with an inventory map of gully erosion (based on a 70 % training dataset and 30 % test dataset), were used to generate spatial maps of gully erosion using DL models. According to the Kolmogorov-Smirnov (KS) statistic performance assessment measure, the simple RNN model (with KS = 91.6) outperformed the GRU model (with KS = 66.6). Based on the results from the simple RNN model, 7.4 %, 14.5 %, 18.9 %, 31.2 % and 28 % of total area of the plain were classified as very-low, low, moderate, high and very-high hazard classes, respectively. According to SHAP plots, CP-PD profiles, and PFI measures, soil silt content, vegetation cover (NDVI) and land use type had the highest impact on the model's output. Overall, the DL modelling techniques and interpretation methods used in this study proved to be helpful in generating spatial maps of soil erosion hazard, especially gully erosion. Their interpretability can support watershed sustainable management.

Assessment of badland susceptibility and its governing factors using a random forest approach. Application to the Upper Llobregat River Basin and Catalonia (Spain)

Badlands are considered hotspots of sediment production, contributing to large fractions of the sediment budget of catchments and river basins. The erosion rates of these areas can exceed 100 t ha-1 y-1, leading to significant environmental and economic impacts. This research aims to assess badland susceptibility and the relevance of its governing factors at different spatial scales using the well-known machine learning approach random forest (RF). The Upper Llobregat River Basin (ULRB, approx. 500 km2) and Catalonia (approx. 32,000 km2) have been selected as study areas. Previous studies stated that the RF approach is successful at making predictions for the same area where it has been trained, but the results of testing it in a different area remains unexplored. This work aims to evaluate the feasibility of upscaling to the large region of Catalonia a RF model trained in the small ULRB area. Two badland datasets of both small and large regions and a total of eleven governing factors have been used to determine the areas susceptible to badlands. Models performance has been analyzed through three different evaluation metrics: overall accuracy, kappa coefficient and area under receiver operating characteristic curve (AUC). The outcomes of this work confirmed that RF is a powerful tool for badland susceptibility analysis, specially when predictions are made in the same scale and spatial context where the model has been trained. Upscaling a RF model defined in the ULRB to the large area of Catalonia has been possible, but improved results have been obtained when the training of the models has directly been performed in the large region. Our final RF modelling results have facilitated the development of a large scale (32,000 km2) Badland Susceptibility Map for the full extension of Catalonia with a predictive overall accuracy of 97%, which strongly emphasizes lithology and Normalized Difference Vegetation Index (NDVI) as the main conditioning factors of badland distribution.