Skip to main content
Environmental Science Division

Mapping Surface Hydrologic Features in Desert Landscapes

EVS developed an ephemeral stream mapping algorithm that could help to develop cost-effective monitoring strategies for managing water resources in arid environments.

Water is the limiting factor in many ecologic systems. Changes in water regimes affect many natural resources and ecosystem services. One critical aspect of monitoring water resources in arid environments is the study of surface hydrologic processes involving ephemeral streams — their flow conveyance, sediment transport, and groundwater recharge. Knowledge of ephemeral stream distributions, attributes, and their changes over time is vital for understanding the hydrologic cycle, local ecosystems, and water availability for human use and wildlife in the region. However, quantifying surface hydrologic processes is extremely challenging because the presence of surface flow or runoff events in arid landscapes is episodic, and established methods for accurately mapping ephemeral stream networks and characterizing their functionality have been lacking.

Very-high-resolution imagery of the Palo Verde Mesa, California (148 km2). The 15-cm-resolution images consist of visible and near-infrared bands. Close-up views show various ephemeral channels in the landscape. [Source: Argonne National Laboratory]

Remote-sensing technologies permit the collection of spatially comprehensive data over large areas, as well as automated processing and streamlined data analysis. By applying knowledge about desert landscapes and multispectral imagery at very high resolutions, EVS scientists developed a new algorithm for mapping ephemeral channel networks and their properties. The Argonne algorithm combines a series of spectral transformations and spatial statistical operations with expert knowledge to generate spatially explicit, spatially contiguous data that represent desert vegetation and the ground surface. Based on the patterns of vegetation occurrence and density, as well as surface brightness and its spatial heterogeneity, the algorithm detects stream channels, extracts channel centerlines, and calculates channel length and width.

Image transformations from the input image to ephemeral stream detection. [Source: Argonne National Laboratory]

The Argonne knowledge-based algorithm extracts well-defined single channels, complex braided streams, and small tributaries across a large, heterogeneous desert landscape and generates a highly detailed map of dry stream networks and their geometry. These data greatly complement other publicly available hydrologic data such as the U.S. Geological Survey (USGS) National Hydrologic Dataset by providing much-needed details of surface hydrologic features on drylands. The algorithm could significantly advance hydrologic modeling and facilitate the development of cost-effective monitoring strategies for managing water resources in desert regions.

The Argonne algorithm extracts significant details of ephemeral stream channel networks that could complement the National Hydrography Dataset (NHD) [Source: Argonne National Laboratory]


Accuracy of Ephemeral Stream Channel Maps Derived From Remotely Sensed Imagery
  Pooled West East
Classification Accuracy      
   Overall accuracy (%) 77.0 91.1 79.8
   Producer’s accuracy (%) 85.8 91.8 88.2
   User’s accuracy (%) 52.2 48.5 51.4
   Kappa coefficient 0.55 0.61 0.54
Centerline Extraction Accuracy      
   Total channel length (km) 146.0
   Accurate channel area (km2) 2.0
   Accurate channel length (km) 92.3
   Channel density (km/km2) 8.5
   Average channel width (m) 33.7
   Accurate delineation 70% (%) 56.3 53.0 59.5
   Accurate delineation 50% (%) 66.5 65.5 67.5
   Recognized (%) 89.0 86.0 92.0
The Argonne algorithm can reliably measure ephemeral stream channel attributes, particularly total length and width, which indicates great promise for monitoring ephemeral stream channels over time and provides vital information for understanding the hydrologic cycle, local ecosystems, and water availability for human use and wildlife in the landscape. The values in brackets indicate reference data.