Evaluation of environmental predictors for sand boil formation: Rhine–Meuse Delta, Netherlands

Abstract

During high-water events, backward erosion piping poses a serious threat to the stability of artificial levee structures. Sand boils forming on the land-side of levees are presently the best indicator of active backward erosion piping, but surveying these features across extensive levee works during flood events is time and resource intensive. Using geographic information system (GIS) technology and a probabilistic statistical analysis, this study assimilates geologic, environmental, and spatial relationship data to generate models to predict the likelihood of sand boil development. Data such as unfavorable geologic orientation, normalized difference vegetation index (NDVI), and blanket thickness are assigned to over 100 km of levees along the Waal and IJssel rivers of the Netherlands. A binary logistic regression is then applied to identify factors with statistically significant relationships to historical sand boil locations, and generate a predictive model using the significant factors. The logit models generally support previous findings indicating unfavorable orientation of geologic deposits and NDVI have a significant relationship with the spatial occurrence of sand boils. Despite this statistical significance, the best predictive model yields an accuracy only 10.83% higher than a random prediction model. Given variance in factors between models, sand boil formation appears to be dependent on a combination of conditions experienced on a local scale. Furthermore, attempting to predict sand boils using a single logit model across multiple geologic environments reduces the influence of region-specific conditions, yielding a less effective model than evaluating the systems individually.