Small Arctic rivers mapped from Sentinel-2 satellite imagery and ArcticDEM

Abstract

Small rivers (<30 m) are significant components of terrestrial river networks and remote sensing is a necessary tool to study them at the global scale. However, current global hydrography data products generally include only rivers that are wider than 30 m, neglecting smaller waterways and thus limiting our understanding of surface water processes, especially in remote areas. We present a new automated methodology to map two Arctic actively-flowing river networks (Kotuy and Coleen Rivers, with study areas of ~12,000 km2 and 10,000 km2, respectively), including small rivers as narrow as 10 m from Sentinel-2 multispectral satellite imagery and high-resolution ArcticDEM digital elevation data. First, preliminary river network masks were generated by using Gabor filtering and path opening morphological operations on Normalized Difference Water Index (NDWI) images. Second, ArcticDEM data were employed to simulate surface flow paths, and the resultant drainage networks were used to create river areas of interest (AOI) and subsequently eliminate non-water features outside the river AOI. Third, gaps along remotely sensed small river channels were filled using the ArcticDEM-modeled drainage networks, yielding continuous river networks with high spatial resolution. We compared the 10 m Sentinel-ArcticDEM merged River Network (SARN) with the National Hydrography Dataset (NHD) and four 30 m Landsat-derived hydrography data products (GSW, GRWL, FROM-GLC 2017, and G3WBM). We conclude: (1) SARNs have similar quality to the >1st order NHD (>70% of NHD streams appear in SARNs). (2) At a reach scale, SARNs include more small rivers with the drainage densities at least ~4.5 times larger than Landsat-derived hydrography data products. Similarly, coupling ArcticDEM data with 30 m Landsat satellite images increases resultant drainage densities approximately three-fold. (3) At a catchment scale, merging remotely sensed river networks with ArcticDEM-modeled drainage networks significantly improves river connectivity (defined as the longest connected river length) with at least a ten-fold increase over the other hydrography data products examined here. In sum, merging high-resolution Sentinel-2 imagery with ArcticDEM data enable effective, continuous mapping of complex fluvial drainage patterns of Arctic rivers.