Stream network conflation with topographic DEMs

https://doi.org/10.1016/j.envsoft.2018.01.009Get rights and content

Highlights

  • An algorithm for aligning vector stream network with topographic DEM is presented.

  • Avoids ambiguity of channel initiation threshold, and the pitfalls of stream-burning.

  • Does not require any user-defined parameters for stream segmentation.

  • Developed in Python using ArcPy and Numpy libraries.

  • Fits seamlessly into existing catchment modeling framework.

Abstract

This paper presents DEM-Stream-Conflation (DSC) algorithm – a scale-independent robust technique of aligning vector streams with flowpaths dictated by raster DEMs. Designed as an alternative to both stream-burning and threshold-dependent stream segmentation techniques, DSC utilizes the existing vector flowlines to identify the channel heads and a sink filled hydrologically conditioned DEM to resolve the flowpaths. The algorithm conceptually initiates the movement of water on a DEM at the starting node of channel heads, from which it traces the path of water to its ultimate watershed outlet. Each trace represents a stream, which is in perfect alignment with the direction dictated by the raster DEM. The algorithm is tested with different DEMs, and its efficacy is demonstrated through the replication of the original vector drainage pattern, derivation of geomorphic attributes that are independent of tested DEM scale, and the visualization of monotonically decreasing longitudinal stream profiles.

Introduction

A vast array of landforms in their many manifestations limit our ability to model and predict channel locations; and yet an accurate and unambiguous spatial delineation of channel forms is necessary for many models of catchment hydrology (Pilotti et al., 1996). Integrated assessment of hydro-geomorphic system often requires an accurate representation of topography primarily in the form of digital elevation models (DEMs). DEMs are used to quantify basin parameters such as catchment boundaries, area, length, slope, aspect, curvature, drainage density, stream order, and Horton's ratios (Bhadra et al., 2008). A stream network is a conduit that links the upstream catchment processes and its impact on the downstream streams and floodplains. Unambiguous and exact alignment of channel geometry and topographic DEMs, therefore, is necessary to model the routing of water and sediments, the fate and transfer of contaminants, and for many other integrated computer models to assess the catchment processes. Geomorphological research on drainage basins is replete with literature on modeling, mapping, and prediction of channel networks from DEMs and other digital data sources (Clubb et al., 2014, Montgomery and Foufoulageorgiou, 1993, Passalacqua et al., 2010, Pelletier, 2013). Despite our rich knowledge of physical processes governing the initiation, growth, and development of channel networks, no predictive model successfully replicates observed channel networks from a DEM (Clubb et al., 2015, Lindsay, 2016, Montgomery and Dietrich, 1988, Passalacqua and Foufoula-Georgiou, 2015).

The constant area threshold method or some variant thereof (e.g., the slope-dependent area threshold approach) is typically used to delineate stream networks from DEMs (Jenson, 1991, Montgomery and Foufoulageorgiou, 1993, Omran et al., 2016). This method implicitly assumes a certain minimum upstream contributing area to initiate channelization. However, this approach produces different realizations of stream network for different thresholds, which is problematic when the upstream contributing area is not unique but varies spatially and is essentially unknown. By assuming a single basin-wide minimum threshold for the upstream contributing area, we are effectively setting the basin-wide drainage density a priori (Pelletier, 2013). Any additional geomorphic parameters generated from this derived stream network loses much of their physical interpretation; as, they are indirectly dependent upon the ill-defined threshold at the outset (Helmlinger et al., 1993).

More recent methods of predicting channel networks from DEMs seek to identify channel head locations using curvature-based techniques such as GeoNet (Passalacqua et al., 2010, Sangireddy et al., 2016), Pelletier (2013), and DrEICH (Clubb et al., 2014). GeoNet uses a non-linear Perona-Malik filter (Perona and Malik, 1990); Pelletier method deploys an adaptive Wiener filter; and DrEICH operates on the tangential curvature of transformed longitudinal stream profiles. These methods have shown promising results in the identification of channel locations from high-resolution DEMs; however, they also require two to three user-defined parameters that may not be readily available. Moreover, these methods are difficult to implement within a general scheme of catchment hydrologic modeling. Passalacqua and Foufoula-Georgiou (2015) and Clubb et al. (2015) go so far as to claim that the intricacies of these algorithms are confusing even for the experts.

Indeed, the development of automated algorithms for channel detection from topographic DEMs is a challenging task; still, we need automated methods of deriving stream networks that are not dependent on arbitrary parameters. In this paper, we approach this challenge by supplementing raster DEMs with readily available vector data on stream networks. Examples of such vector data include stream features obtained through digitizing paper maps or aerial photography, field surveys, and other legacy systems. These vector networks are not always collocated with DEM defined channel centerlines, but can still be a valuable source of information for inferring the channel locations. An exact alignment between the vector network and the channels derived from DEM is required in the models of catchment hydrology to model the transfer and conveyance of water, sediment, and chemicals uniquely from the watersheds drained by corresponding stream segment. Even though stream-burning is a common practice to align the vector stream with DEM, in the process, a hydrologically conditioned DEM is further modified due to the etching of stream network on the DEM.

The overarching objective of this paper is to develop conflation algorithms that fuses the known drainage pattern into a raster DEM, but without further modifying the hydrologically conditioned DEM, consequently rendering the topographic attributes derived from DEM more representative of catchment characteristics. The strength of our approach is that it preserves the existing topological relationships of the drainage network while simultaneously avoiding the perils of stream-burning, and the ambiguity related to choosing an arbitrary threshold value for channel initiation.

Section snippets

Background on stream conflation

Conflation refers to the alignment and integration of seemingly different but related sources of data to improve the quality of one or both sources. The problem addressed herein is an example of vector to raster data conflation where two data sets referencing the same area are reconciled to obtain a consistent stream network that confirms the local topography defined by DEMs. Stream-burning is a technique commonly used to address the issue of misalignment between vector stream networks and DEMs

Study area and data

The DSC algorithm is developed and tested on a catchment corresponding to a U.S. Geological Survey (USGS) stream-gaging station 02167582 named Bush River Near Prosperity in South Carolina, USA (Fig. 1). The catchment, located in the Piedmont region of South Carolina with an area of 298 square kilometers, is a headwater watershed that is tributary to Saluda River. The relatively flat topography of the watershed is predominantly covered by forest and agriculture landuse classes. A medium scale

Results

The premise of this work is based on the existence vector stream network to infer the general flow pattern, and that of a hydrologically conditioned DEM to resolve the flow direction. The DSC algorithm reproduces the general flow pattern of vector flowlines that is coincident with the path of least resistance depicted by the DEM. DSC avoids the artifacts and further degradation of hydrologically conditioned DEM due to stream-burning and drainage-enforcement algorithms. This is achieved within a

Discussion

Misalignment between vector stream network and topographic DEM limits their use in the quantitative analysis and modeling of drainage basins. The typical stream-burning approach arbitrarily modifies the DEM – severely compromising the derivation of catchment attributes that depend on DEM. At the other extreme, the threshold dependent stream network is associated with a high degree of uncertainty because of the difficulties in objectively defining the minimum support area threshold required for

Acknowledgements

This research was funded by the Optech Inc. Geosensing Systems Engineering Endowment and the Engineering School of Sustainable Infrastructure & Environment (ESSIE), University of Florida.

References (38)

  • F. Clubb et al.

    Reply to comment by P. Passalacqua and E. Foufoula-Georgiou on “Objective extraction of channel heads from high-resolution topographic data'

    Water Resour. Res.

    (2015)
  • A.R. da Paz et al.

    Errors in river lengths derived from raster digital elevation models

    Comput. Geosci.

    (2008)
  • A.C.V. Getirana et al.

    Improving hydrological information acquisition from DEM processing in floodplains

    Hydrol. Process.

    (2009)
  • T. Goulden et al.

    Sensitivity of watershed attributes to spatial resolution and interpolation method of LiDAR DEMs in three distinct landscapes

    Water Resour. Res.

    (2014)
  • F. Hellweger et al.

    AGREE-DEM Surface Reconditioning System

    (1997)
  • K.R. Helmlinger et al.

    On the use of digital elevation model data for hortonian and fractal analyses of channel networks

    Water Resour. Res.

    (1993)
  • J.D. Hunter

    Matplotlib: a 2D graphics environment

    Comput. Sci. Eng.

    (2007)
  • S.K. Jenson

    Applications of Hydrologic information automatically extracted from digital elevation models

    Hydrol. Process.

    (1991)
  • S.K. Jenson et al.

    Extracting topographic structure from digital elevation data for geographic information-system analysis

    Photogramm. Eng. Rem. Sens.

    (1988)
  • Cited by (0)

    View full text