Mapping the Symptoms of dryland salinity with EO-1 Hyperion satellite imagery

Abstract

Salinity mapping with aerial photography and multispectral satellite imagery has significant limitations. This study evaluated hyperspectral satellite imagery to map dryland salinity symptoms near the River Murray mouth in South Australia. Hyperion imagery has advantages of high spectral resolution combined with broader spatial coverage than a single airborne hyperspectral swath. This study aimed to exploit this high spectral resolution in order to discriminate key indicators of salinity: saline soils and perennial halophytic vegetation, such as samphire (Halosarcia pergranulata). Automatically-generated endmembers derived from the Hyperion image contained significant noise, particularly in the far-NIR region and lacked diagnostic absorption features in the SWIR, making them unsuitable as a basis for image feature extraction. Instead, partial unmixing used image reference spectra from regions-of-interest that corresponded to known samphire patches and exposed saltpans. KHAT values of 0.50 and 0.38 were calculated for the comparison of image and field samphire and saltpan maps respectively. The higher accuracy of the samphire delineation may be due to lower image noise in the visible and NIR, regions containing spectral characteristics associated with vegetation succulence. Higher accuracies may be achieved with improved SNR in the SWIR region because this region contains important spectral properties related to other halophytic adaptations. The mapping accuracy for the saltpans may also be adversely affected by data noise that masked mineral absorption features in the SWIR. If future high spectral resolution satellites have improved SNR, then salinity mapping with satellite imagery has enormous potential. Image analysis of low-noise hyperspectral imagery may be able to discriminate some halophytic vegetation and isolate spectral features relating to saline mineralogy. Because satellite platforms have the advantage of broader spatial coverage and regular repetition, satellite-based hyperspectral imagery is well suited to monitoring salinity over medium sized catchments.