Optical Remote Sensing of Submerged Aquatic Vegetation: Opportunities for Shallow Clearwater Streams

Remote sensing has rarely been used as a tool to map and monitor submerged aquatic vegetation (SAV)
in rivers, due to a combination of insufficient spatial resolution of available image data and strong attenuation
of light in water through absorption and scattering. The latter process reduces the possibility to
use spectral reflectance information to accurately classify submerged species. However, increasing availability
of very high resolution (VHR) image data may enable the use of shape and texture features to help
discriminate between species by taking an object based image analysis (OBIA) approach, and overcome
some of the present limitations.
This study aimed to investigate the possibility of using optical remote sensing for the detection and
mapping of SAV. It firstly looked at the possibilities to discriminate submerged macrophyte species based
on spectral information only. Reflectance spectra of three macrophyte species were measured in situ
across a range of submergence depths. The results showed that water depth will be a limiting factor
for the classification of species from remote sensing images. Only Spiked Water Milfoil (Myriophyllum
spicatum) was indicated as spectrally distinct through ANOVA analysis, but subsequent Jeffries–Matusita
distance analysis did not confirm this. In particular Water Crowfoot (Ranunculus fluitans) and Pondweed
(Potamogeton pectinatus) could not be discriminated at 95% significance level. Spectral separability of
these two species was also not possible without the effect of an overlying water column.
Secondly, the possibility to improve species discrimination, using spatial and textural information was
investigated for the same SAV species. VHR image data was acquired with a Near Infrared (NIR) sensitive
DSLR camera from four different heights including a telescopic pole and a Helikite UAS. The results show
that shape and texture information can improve the detection of the spectrally similar Pondweed and
Water Crowfoot from VHR image data. The best performing feature ‘length/width ratio of sub-objects’
was obtained through expert knowledge. All of the shape and texture based features performed better
at species differentiation than the spectrally based features.
In conclusion this study has shown that there is considerable potential for the combination of VHR
data and OBIA to map SAV in shallow stream environments, which can benefit species monitoring and
management.