What controls the spatial distribution of stream-road crossings?

Stream-road crossings arise when natural fluvial networks (emerging over geological timescales)  intersect built transportation networks (which are designed). Such conflicting intersections are resolved by installing conveyance infrastructure, such as culverts and bridges, which mitigate flood risk to adjacent communities and prevent disruptions to the transportation networks. Studying the spatial distribution of stream-road crossings across large, geographically diverse spatial scales can shed light on how the density of these crossings depends on topographic, geomorphic, hydrologic, and urban controls. In this research, we address these questions by performing a large-scale analysis of the stream-road crossings of New York State (~141,000 sq. kilometers).  We use a grid–based scheme at several spatial resolutions (initially 10×10 km and 5×5 km), which allows us to study the effect of the spatial resolution on the observed distributional patterns. Looking beyond the primary effect of stream and road densities, this study focuses on identifying dependence on second–order drivers that characterize the stream–road crossing distribution. For this analysis, we employ large remotely-sensed geospatial datasets as features. Given the high dimensionality of the feature space and the strong presence of multicollinearity, dimensionality reduction techniques are used to identify latent structures and dominant modes of variability, while clustering methods are applied to separate regions with internally consistent geospatial characteristics. We finally compare outcomes across spatial resolutions to generate insights on how inferred relationships depend on various hydrologic, geomorphic, and land-use features.