The overarching aim of PROSLIDE is to exploit the potential of innovative input data, available ground truth data and novel modelling designs (i.e. data-driven and physically-based) at different scales to improve the predictability of where and when landslides will occur. The main project innovations are based on three closely interlinked pillars that determine the research objectives and associated research questions:
1) Innovative input data
The first objective is to exploit the potential of diverse multi-temporal proximal and remote sensing data to represent landslide controls for the purpose of landslide prediction.
1a) How and to which extent can meteorological RADAR derived rainfall estimates be integrated into data-driven and physically-based models?
1b) What is the potential of environmental parameters derived from new generation satellite EO (e.g. soil moisture) and varying spatial resolution laser scanning (ALS, TLS/ULS) to approximate landslide controls?
2) New modelling designs
The second objective is to develop – on the basis of innovative input data (1) – suitable modelling strategies in order to integrate already available approaches with new modelling designs:
2a) How to optimally integrate heterogeneous data on static and dynamic landslide controls by applying data-driven models at regional scale and physically-based models at catchment scale?
2b) How to take advantage of data-driven and physically-based landslide prediction models to develop an integrated approach?
2c) Can the application of novel data-driven modelling designs (e.g. mixed-effects model) improve the handling of biased landslide information, area heterogeneity and hierarchical data?
2d) To what extent can a calibrated physically-based model (parameter ensemble) represent the parameter variability?
3) Scale dependency and result evaluation
The third objective is to elaborate optimal investigation scales by exploring if, how and why changes in the spatial, temporal, methodical analysis dimension modify subsequent landslide predictions and interpretation possibilities:
3a) How do spatio-temporal landslide predictions vary in response to changes in the underlying spatial investigation unit (e.g. pixel vs. slope-units vs. morpho-climatological units)? Which spatial level of detail is most appropriate?
3b) How do landslide predictions vary in response to changes in the temporal investigation scale and which (combination of) temporal scales are optimal?
3c) Why and how does a variation in the spatial (3a), temporal (3b) and methodical dimension (i.e. data-driven vs. physically-based; 2a,b) affect ensuing results (e.g. predictive performance, geomorphic plausibility, spatial and temporal transferability) and what are potential implications for early warning?