Semantic array programming in data-poor environments: assessing the interactions of shallow landslides and soil erosion

This research was conducted with the main objective to better integrate and quantify the role of water-induced shallow landslides within soil erosion processes, with a particular focus on data-poor conditions. To fulfil the objectives, catchment-scale studies on soil erosion by water and shallow landslides were conducted. A semi-quantitative method that combines heuristic, deterministic and probabilistic approaches is here proposed for a robust catchment-scale assessment of landslide susceptibility when available data are scarce. A set of different susceptibility-zonation maps was aggregated exploiting a modelling ensemble. Each susceptibility zonation has been obtained by applying heterogeneous statistical techniques such as logistic regression (LR), relative distance similarity (RDS), artificial neural network (ANN), and two different landslide-susceptibility techniques based on the infinite slope stability model. The good performance of the ensemble model, when compared with the single techniques, make this method suitable to be applied in data-poor areas where the lack of proper calibration and validation data can affect the application of physically based or conceptual models. A new modelling architecture to support the integrated assessment of soil erosion, by incorporating rainfall induced shallow landslides processes in data-poor conditions, was developed and tested in the study area. This proposed methodology is based on the geospatial semantic array programming paradigm. The integrated data-transformation model relies on a modular architecture, where the information flow among modules is constrained by semantic checks. By analysing modelling results within the study catchment, each year, on average, mass movements are responsible for a mean increase in the total soil erosion rate between 22 and 26% over the pre-failure estimate. The post-failure soil erosion rate in areas where landslides occurred is, on average, around 3.5 times the pre-failure value. These results confirm the importance to integrate landslide contribution into soil erosion modelling. Because the estimation of the changes in soil erosion from landslide activity is largely dependent on the quality of available datasets, this methodology broadens the possibility of a quantitative assessment of these effects in data-poor regions.