Severity of salinity accurately detected and classified on a paddock scale with high resolution multispectral satellite imagery

Abstract

We hypothesised that digital mapping of various forms of salt-affected soils using high resolution satellite imagery, supported by field studies, would be an efficient method to classify and map salinity, sodicity or both at paddock level, particularly in areas where salt-affected patches are small and the effort to map these by field-based soil survey methods alone would be inordinately time consuming. To test this hypothesis, QuickBird satellite data (pan-sharpened four band multispectral imagery) was used to map various forms of surface-expressed salinity in an agricultural area of South Australia. Ground-truthing was performed by collecting 160 soil samples over the study area of 159km2. Unsupervised classification of the imagery covering the study area allowed differentiation of severity levels of salt-affected soils, but these levels did not match those based on measured electrical conductivity (EC) and sodium adsorption ratio (SAR) of the soil samples, primarily because the expression of salinity was strongly influenced by paddock-level variations in crop type, growth and prior land management. Segmentation of the whole image into 450 paddocks and unsupervised classification using a paddock-by-paddock approach resulted in a more accurate discrimination of salinity and sodicity levels that was correlated with EC and SAR. Image-based classes discriminating severity levels of salt-affected soils were significantly related with EC but not with SAR. Of the spectral bands, bands 2 (green, 520-600nm) and 4 (near-infrared, 760-900nm) explained the majority of the variation (99 per cent) in the spectral values. Thus, paddock-by-paddock classification of QuickBird imagery has the potential to accurately delineate salinity at farm level, which will allow more informed decisions about sustainable agricultural management of soils.