Measuring channel planform change from image time series: A generalizable, spatially distributed, probabilistic method for quantifying uncertainty

Abstract

Channels change in response to natural or anthropogenic fluctuations in streamflow and/or sediment supply and measurements of channel change are critical to many river management applications. Whereas repeated field surveys are costly and time‐consuming, remote sensing can be used to detect channel change at multiple temporal and spatial scales. Repeat images have been widely used to measure long‐term channel change, but these measurements are only significant if the magnitude of change exceeds the uncertainty. Existing methods for characterizing uncertainty have two important limitations. First, while the use of a spatially variable image co‐registration error avoids the assumption that errors are spatially uniform, this type of error, as originally formulated, can only be applied to linear channel adjustments, which provide less information on channel change than polygons of erosion and deposition. Second, previous methods use a level‐of‐detection (LoD) threshold to remove non‐significant measurements, which is problematic because real changes that occurred but were smaller than the LoD threshold would be removed. In this study, we present a new method of quantifying uncertainty associated with channel change based on probabilistic, spatially varying estimates of co‐registration error and digitization uncertainty that obviates a LoD threshold. The spatially distributed probabilistic (SDP) method can be applied to both linear channel adjustments and polygons of erosion and deposition, making this the first uncertainty method generalizable to all metrics of channel change. Using a case study from the Yampa River, Colorado, we show that the SDP method reduced the magnitude of uncertainty and enabled us to detect smaller channel changes as significant. Additionally, the distributional information provided by the SDP method allowed us to report the magnitude of channel change with an appropriate level of confidence in cases where a simple LoD approach yielded an indeterminate result. © 2020 John Wiley & Sons, Ltd.

Identifying Factors That Influence Accuracy of Riparian Vegetation Classification and River Channel Delineation Mapped Using 1 m Data

“…However, both focused on specific issues, rather than developing broadly applicable techniques. Donovan et al [17] assessed producer's error in the manual delineation process, whereas Lea and Legleiter [54] addressed image registration, which is particularly relevant in change studies that utilize multiple images of a location [55]. While the accuracy of manual digitization is important, manual approaches have practical limitations and cannot be applied over large areas.…”

Biogeomorphic recovery of a river reach affected by mining

“…In summary, although the spatial resolution of the datasets, the river positions marked on the map sources and the extent of inundated area of the image sources reflect the methods of data capture, mapping conventions and river stage at the time of acquisition (Fuller et al, 2013), together they provide a good visual impression of the characteristics of planform change over a 175‐year period. However, when quantitative comparisons of channel positions are attempted, these are subject to the above error sources as well as image co‐registration errors introduced during georeferencing (Leonard et al, 2020). Complete error analysis is challenging, but Leonard et al (2020) suggest that where the signal of channel change is large, and the aim is not to quantify the magnitude of volumetric change, a computationally intensive method of uncertainty analysis is probably not needed.…”

Sediment‐ecological connectivity in a large river network

“… b Orthorectification error (horizontal and vertical root‐mean square error) (e.g., Leonard et al, 2020). …”

“…Photographs were digitized and then orthorecitified using digital elevation models (DEMs) created with structure‐from‐motion (SfM) in AgiSoft Photoscan Professional (Version 1.4) (Bakker & Lane, 2017; Fonstad et al, 2013; Leonard et al, 2020). Average horizontal and vertical orthorectification error in SfM was 4.5 and 4 m, respectively.…”

“…The time period of active erosion in each tributary was thus identified and constrained, with the assumption that a stability or decrease in channel and arroyo dimensions from a T A B L E 1 US Geological Survey stream gages most proximal to the study forests and the years of the six largest recorded floods for each gage. The Yampa River at Deerlodge Park gage record (09260050) can be extended to 1922 using the combined record of the Lily and Maybell gages b Orthorectification error (horizontal and vertical root-mean square error) (e.g., Leonard et al, 2020).…”