| Intro |
Desert shrubs are the dominant vegetation type in arid deserts and serve as crucial elements in
sand retention, biodiversity maintenance, and carbon sequestration. However, due to their patchy
and scattered distributions and spectral resemblance to herbaceous plants, desert shrub mapping
relies on high-resolution imagery, which is less accessible for large-scale mapping. Here, a set of
desert shrub mapping indices for Sentinel-2 (DSMIS) and for universal medium-resolution imagery
(DSMIL) is developed to distinguish desert shrubs with dense vimen canopies. The index exploits
the canopy structure and spectral characteristics, which have a sparse and multilayered canopy
and a high proportion of desiccated branches, resulting in a consistently low reflectance in the rededge
to near-infrared range. The effectiveness of DSMI indices was examined in Ordos, China. In
the experiment, an optimal threshold of 10.3 was obtained via DSMIs with Sentinel-2, which
achieved an overall accuracy of 91.6% and identified a minimum desert shrub coverage of 0.23.
In comparison, a threshold of 9.7 was obtained by DSMIL with Landsat-8, achieving an accuracy of
90.1% and identifying a minimum coverage of 0.17. The performance of DSMI was superior to that
of the commonly used random forest, and this index could further improve classification as a
complement to machine learning methods. The late stage of the nongrowing season was identified
as the optimal period for desert shrub mapping with the index. DSMI also performed well at
two test sites with diverse desert shrub dominant species and growing conditions. This study
provides a novel index for desert shrub mapping and a practical tool for monitoring desertification
in arid desert regions. It also offers a new perspective for historical dynamic studies of desert shrubs
and other land cover types where only optical data are available. |
English
中文