Identifying common near-surface and subsurface stratigraphic units using EM34 signal data and fuzzy k-means analysis in the Darling River valley

Abstract

In irrigated areas of central and northern New South Wales secondary soil salinisation is of increasing concern. This is particularly the case for the Bourke Irrigation District (mid-Darling River valley) where incipient traces of point-source salinisation are evident. Much of this is attributed to significant changes in the water balance and in particular increased deep drainage and mobilisation of primary salts. In the first instance, near-surface prior-stream channels and subsurface paleochannels evident within Cenozoic alluvial deposits act as significant hydrological features. Owing to the porosity and connectivity of the Cenozoic sediments, deep draining water interacts with ancient and highly saline Cretaceous marine mudstones. Unfortunately, natural resource data on the spatial distribution of near-surface and subsurface stratigraphy are not readily available to understand these processes or determine best management. In this paper, we demonstrate how fuzzy k-means (FKM) analysis of EM34 signal data collected across the Bourke Irrigation District can be used to identify and map common near-surface and subsurface stratigraphic units. EM34 survey was carried out in the horizontal mode of operation with measurements made at coil spacing of 10, 20 and 40 m (i.e. EM34-10, -20 and -40, respectively). In all, 1236 sites were visited on a ∼0.5–1 km grid. The EM34 signal data were classified numerically using the FKM clustering algorithm. The iteration of fuzzy exponents (f) and various indices, including the fuzziness performance index (FPI) and normalised classification entropy (NCE) enabled the determination of f = 1.6 and k = 4 classes to be selected for further investigation. Using fuzzy canonical analysis we find that the EM34-10 and EM34-20 signal data best discriminate the classes. The resulting k = 4 classes (i.e. Class A, B, C and D) are then mapped using a method that ensures summation of class membership (m) values to unity and using local ordinary kriging. Class A represents saline and clay-rich sediments associated with outcrops of the Cretaceous marine mudstone. Class D represents interbedding of sand within the clay units indicative of the Cenozoic alluvial deposits that overly the Cretaceous mudstone. Similarly, Class B represents Cenozoic sediments with the presence of paleochannel(s). Class C characterises the eolian dune and alluvial floodplain physiographic units. The use of a confusion index (CI) highlights areas of uncertainty in the FKM classification mapping and indicates where the collection of additional information is appropriate. Validation of the FKM approach is confirmed by calculating near-surface (0–6 m) and subsurface (6–12 m) class average soil physical and chemical property (i.e. clay content, cation exchange capacity and salinity) variance (i.e. S2 Z ) and total within-class variance (i.e. S 2 T ) of a FKM map. Overall, the average subsurface ECe data produces the greatest reduction in variance. This suggests the FKM classification of EM34 signal data is influenced by saline groundwater-tables in the Cenozoic sediments which lie above the Cretaceous marine mudstone. The FKM class memberships and CI along a short but detailed transect across the Bourke Irrigation District are interpreted with respect to the spatial distribution of near-surface and subsurface soil property data. We conclude the FKM analysis of EM34 provides a better understanding of the spatial distribution and horizontal layering of the various stratigraphic units which characterise the Bourke Irrigation District. The results help explain the broad hydrological processes driving point source secondary soil salinisation and provides a framework for the deployment of measuring and monitoring sites.