Remote Sensing Large‐Wood Storage Downstream of Reservoirs During and After Dam Removal: Elwha River, Washington, USA

Abstract

Abstract Large wood is an integral part of many rivers, often defining river‐corridor morphology and habitat, but its occurrence, magnitude, and evolution in a river system are much less well understood than the sedimentary and hydraulic components, and due to methodological limitations, have seldom previously been mapped in substantial detail. We present a new method for this, representing a substantial advance in automated deep‐learning‐based image segmentation. From these maps, we measured large wood and sediment deposits from high‐resolution orthoimages to explore the dynamics of large wood in two reaches of the Elwha River, Washington, USA, between 2012 and 2017 as it adjusted to upstream dam removals. The data set consists of a time series of orthoimages (12.5‐cm resolution) constructed using Structure‐from‐Motion photogrammetry on imagery from 14 aerial surveys. Model training was optimized to yield maximum accuracy for estimated wood areas, compared to manually digitized wood, therefore model development and intended application were coupled. These fully reproducible methods and model resulted in a maximum of 15% error between observed and estimated total wood areas and wood deposit size‐distributions over the full spatio‐temporal extent of the data. Areal extent of wood in the channel margin approximately doubled in the years following dam removal, with greatest increases in large wood in wider, lower‐gradient sections. Large‐wood deposition increased between the start of dam removal (2011) and winter 2013, then plateaued. Sediment bars continued to grow up until 2016/17, assisted by a partially static wood framework deposited predominantly during the period up to winter 2013.

Keywords

• large wood
• fluvial geomorphology
• Elwha River
• deep learning
• image segmentation
• dam removal