Broadband vegetation index performance evaluated for a low‐cover environment

Abstract

Fourteen vegetation indices (VIs) were quantitatively evaluated using a Landsat TM dataset spanning 17 years over the San Luis Valley, Colorado, USA to find the best VI for use in sparsely vegetated arid regions. The VIs were compared to the known linear promotional effect of antecedent precipitation in an area of stable water tables. The r ² value of the linear regression for these indices was used to determine performance. The following published VIs were evaluated: ARVI, DVI, EVI, IPVI, MSAVI, MSAVI₂, NDVI, NDVI∗, PVI, RVI, SAVI, TSAVI, WDVI, and a new index introduced in this paper, NDVI_offset. NDVI∗ and NDVI_offset performed best of all the tested VIs, both with r ² = 0.77. Standard NDVI was a relatively poor performer (r ² = 0.37). The high performance of NDVI_offset is promising for use in regions or multi‐temporal sequences that lack saturated NDVI pixels.

Acknowledgments

Authors enthusiastically thank the Rio Grande Water Conservation District and especially David Robbins, Allen Davey and Ralph Curtis for their unflagging interest and generous support for this work. We also thank Pete Humphrey and Jeff Silverman for assistance with GIS analyses necessary to complete this work and the Alamosa, Colorado office of the US Bureau of Reclamation for supplying monitoring well data from the Closed Basin Project.