On the use of explainable AI for susceptibility modeling: Examining the spatial pattern of SHAP values

• DOI: 10.1016/j.gsf.2024.101800
• Publication Date: 2024-02-02T03:27:05Z
• Authors: Nan Wang; Hongyan Zhang; Ashok Dahal; Weiming Cheng; Min Zhao; Luigi Lombardo
• Keywords: Spatial contextual awareness

Hydro-morphological processes (HMP, any natural phenomenon contained within the spectrum defined between debris flows and flash floods) are globally occurring hazards which pose great threats to our society, leading fatalities economical losses. For this reason, understanding dynamics behind HMPs is needed aid in hazard risk assessment. In work, we take advantage of an explainable deep learning model extract global local interpretations HMP occurrences across whole Chinese territory. We use a neural network architecture interpret results through spatial pattern SHAP values. doing so, can understand prediction on hierarchical basis, looking at how predictor set controls overall susceptibility as well same level single mapping unit. Our accurately predicts with AUC values measured ten-fold cross-validation ranging 0.83 0.86. This predictive performance attests for excellent skill. The main difference respect traditional statistical tools that latter usually lead clear interpretation expense high performance, otherwise reached via machine/deep solutions, though interpretation. recent development AI key combine both strengths. explore combination context modeling. Specifically, demonstrate extent one enter new data-driven interpretation, supporting decision-making process disaster mitigation prevention actions.