Mapping surface systems of dryland salinity with hyperspectral imagery

Abstract

Hyperspectral images from three sensors were compared for their ability to discriminate and map selected symptoms of salinity in a dryland agricultural area in southern Australia. The symptoms mapped are widespread in this environment: the perennial halophytic shrub samphire (Halosarcia pergranulata), a salt tolerant grass, sea barley grass (Hordeum marinum) and salt encrusted pans. Airborne HyMap and satellite Hyperion imagery were acquired in the late summer dry season to maximise soil and perennial vegetation discrimination, while airborne CASI imagery was acquired in late spring to capture senescence of the annual sea barley grass. Image spectra were used to map surface salinity symptoms using partial spectral unmixing techniques. Saltpans were discriiiminated using the gypsum 1750 nm absorption feature, while full-wavelength image spectra were necessary to map the halophytic plants. Salinity symptom maps were validated at over 100 random field sites and compared with independent mapping of soils and salinity from aerial photography and Landsat imagery. HyMap imagery produced the most accurate maps of samphire and saltpans, compared with CASI and Hyperion, discriminating samphire from other perennial native vegetation and saltpans from other reflective non-saline sands and soils. Discrimination of the sea barley grass with CASI was largely based on the residual greenness the plant retained in late spring, but led to some confusion with other non-saline pastures. By comparison, previous multispectral salinity mapping in the same area required multitemporal imagery and ancillary topographic and soils data to achieve comparable discrimination of surface salt. The hyperspectral mapping identified some saline areas that had not been delineated in previous aerial photo interpretation and provided enhanced spatial information on the distribution of particular salinity indicators. The 3 metre ground resolution of the HyMap imagery also provided greater definition of symptoms within saline areas identified in prior 1:50,000 soil-landscape mapping. The study demonstrates that hyperspectral imagery can improve discrimination of vegetation and mineral indicators of surface salinity, that seasonality of the imagery is important in capturing diagnostic spectral differences, and that consistent spectral mapping can be achieved over sizeable areas using high-resolution airborne imagery. The increased spatial resolution and additional information about landscape composition derived from the hyperspectral imagery are significant improvements over traditional soil and salinity mapping based on aerial photography interpretation.