Overcoming Data Scarcity Related Issues for Landslide Susceptibility Modeling with Machine Learning

Landslide susceptibility maps can be a useful tool to support holistic urban planning in mountainous environments. Data-driven methods for landslide susceptibility modeling work well even in data scarce areas, and there is an increasing relevance of machine learning methods that help analyze efficiently large and complex datasets. In this contribution we present some of our study examples to show how data quality, quantity, complexity, and preparation can have major effects on the outcomes of landslide susceptibility modeling. The aforementioned aspects are too often neglected in spite of their relevance, both in data scarce, but also data rich areas. We also use these examples to discuss the way we evaluate landslide susceptibility models, as the spatial performance of landslide susceptibility maps often differs from the mathematical performance. We finally discuss the necessity of standards for input data, modeling results and result communication to improve the usability of landslide susceptibility models in urban planning.