Using full and partial unmixing algorithms to estimate the inundation extent of small, isolated stock ponds in an arid landscape

Wetlands
By: , and 

Links

Abstract

Many natural wetlands around the world have disappeared or been replaced, resulting in the dependence of many wildlife species on small, artificial earthen stock ponds. These ponds provide critical wildlife habitat, such that the accurate detection of water and assessment of inundation extent is required. We applied a full (linear spectral mixture analysis; LSMA) and partial (matched filtering; MF) spectral unmixing algorithm to a 2007 Landsat 5 and a 2014 Landsat 8 satellite image to determine the ability of a time-intensive (i.e., more spectral input; LSMA) vs. a more efficient (less spectral input; MF) spectral unmixing approach to detect and estimate surface water area of stock ponds in southern Arizona, USA and northern Sonora, Mexico. Spearman rank correlations (rs) between modeled and actual inundation areas less than a single Landsat pixel (< 900 m2) were low for both techniques (rs range = 0.22 to 0.62), but improved for inundation areas >900 m2 (rs range = 0.34 to 0.70). Our results demonstrate that the MF approach can model ranked inundation extent of known pond locations with results comparable to or better than LSMA, but further refinement is required for estimating absolute inundation areas and mapping wetlands <1 Landsat pixel.

Additional publication details

Publication type Article
Publication Subtype Journal Article
Title Using full and partial unmixing algorithms to estimate the inundation extent of small, isolated stock ponds in an arid landscape
Series title Wetlands
DOI 10.1007/s13157-019-01201-7
Volume 40
Year Published 2020
Language English
Publisher Springer
Contributing office(s) Southwest Biological Science Center
Description 13 p.
First page 563
Last page 575
Google Analytic Metrics Metrics page
Additional metadata about this publication, not found in other parts of the page is in this table