You are here:
Improved wetland classification using eight-band high-resolution satellite imagery and a hybrid approach
Citation:
Lane, C., H. Liu, Brad Autrey, O. Anenkhonov, V. Chepinoga, AND Q. Wu. Improved wetland classification using eight-band high-resolution satellite imagery and a hybrid approach. Remote Sensing. MDPI, Basel, Switzerland, 6(12):12187-12216, (2014).
Impact/Purpose:
Study uses novel technology to identify and classify wetland resources
Description:
Although remote sensing technology has long been used in wetland inventory and monitoring, the accuracy and detail level of derived wetland maps were limited or often unsatisfactory largely due to the relatively coarse spatial resolution of conventional satellite imagery. This research explores and evaluates the utility of the newly emerged high resolution WorldView-2 (WV2) satellite system in identifying and classifying freshwater deltaic wetland vegetation and aquatic habitats. We designed a hybrid approach to the classification of WV2 multispectral imagery, consisting of initial unsupervised classification, training data acquisition and analysis, indicator species analysis (ISA), and final supervised classification. The indicator species analysis was introduced for the first time to interpret and label different groups of training sites for the final supervised classification. The hybrid classification approach was successfully applied to the wetland classification and inventory for the Selenga River Delta of Lake Baikal, Russia. With the high spatial and spectral resolutions of WV2 satellite imagery, we discriminated between different aquatic habitats and wetland vegetation at the genus and community levels with unprecedented accuracy and spatial detail. We achieved overall classification accuracy on the order of 86.5% for 22 classes of aquatic and wetland habitats at the finest scale and > 91% accuracy for broad substrate and vegetation classes at more generalized levels. We found that additional variables, such as the NDVI and image texture, were valuable for improving the overall classification accuracy and particularly separating scrub/shrub wetland from emergent herbaceous wetlands. Our analysis demonstrated that the addition of four new spectral bands of WV2 satellite system contributed to the increase in the overall classification accuracy by 3.5%, though the discrimination of certain communities increased more markedly. The coastal band of WV2 imagery was found to be useful in better separating between different open water and aquatic (vegetated) habitats, while yellow, red-edge and NIR2 bands contributed to better discrimination between different vegetated aquatic and terrestrial habitats.
