Complete Radiometric Grid of Australia (Radmap) v4 2019 with modelled infill

Brief description

The complete infilled K, eTh and eU grids are based on the Radiometric Map of Australia (radmapv4) 2019 (Poudjom Djomani and Minty, 2019a, b, c) with gaps in coverage infilled using environmental correlation machine learning prediction. The radiometric, or gamma-ray spectrometric method, measures the natural variations in the gamma-rays detected near the Earth's surface as the result of the natural radioactive decay of potassium (K), uranium (U) and thorium (Th). However because Uranium and Thorium abundances are calculated by measuring gamma emission associated with their daughter radionuclides they are typically expressed as equivalent eU and eTh. The 2019 radiometric grid is compiled from airborne geophysical surveys conducted by Commonwealth, State and Northern Territory Governments and the private sector. Over 600 airborne gamma-ray spectrometric surveys were merged and gridded to a cell size of approximately 100m (0.001 degrees) to produce the Radiometric Map of Australia (radmapv4) 2019. Gamma-rays emitted from the surface mainly relate to the mineralogy and geochemistry of the bedrock and weathered materials or regolith. To infill gaps in the national gamma-ray grid (radmapv4 -2019) we have compiled a set of national covariates or predictive datasets that capture landscape processes, regolith and geology that are likely correlated to the distribution of K, eTh and eU at the surface. These datasets include satellite imagery (to map surface mineralogy and vegetation), terrain attributes (e.g. slope, relief), gravity (Lane et al, 2020) and surface geology. A boosted regression tree algorithm called XGBoost (open-source software library for gradient boosting machine learning) was used to train relationships between airborne estimates of K, eTh and eU with the covariate datasets. The training set used the Australia Wide Airborne Geophysical Survey (AWAGS) (Milligan et al., 2009). Local model predictions were generated for gaps in the 2019 version of the national grid by clipping subsets of the AWAGS survey lines and in places extracting additional training survey sites from nearby surveys. The strength of the correlations between the training observation and the covariates were highest in semi-arid areas with decreasing correlations from K through to eTh and eU. Modelled grids of K, eTh and eU were merged with the Radiometric Map of Australia (radmapv4 -2019) using the grid merge module in Intrepid Geophysics software. The first step was to scale the modelled dataset to the reference dataset, then apply a DC shift. The second step was to surface adjust the grid, which computes a two dimensional surface calculated from the differences in its value between the reference grid, it then fits a difference surface with the largest mean signal value and reiterates this process until the difference is within a pre-defined threshold. The third step is to merge the modelled dataset with the Radiometric Map of Australia (radmapv4) 2019, using a feathering process where measured radiometric values are ranked higher over the modelled data. The complete infill radiometric grids have been generated for regolith (including soils) and geological mapping and can be used as a seamless dataset for predictive modelling using machine learning. The product can be seen as an interim dataset until the gaps are filled in through new airborne survey acquisition. It is important to recognise that the infill grids are based on correlations between airborne flight-line estimates of the radioelements and covariate thematic datasets. Responses and patterns observed within these gap areas are therefore not reflecting measurements using the airborne spectrometry. Equally, the covariate approach should not be expected to confidently identify gamma-ray ‘outliers’ or anomalies that have been used in other geophysical survey approaches. Lane, R. J. L., Wynne, P. E., Poudjom Djomani, Y. H., Stratford, W. R., Barretto, J. A., and Caratori Tontini, F., 2020, 2019 Australian National Gravity Grids: Geoscience Australia, eCat Reference Number 133023, https://pid.geoscience.gov.au/dataset/ga/133023 Milligan, P., Minty, B., Richardson, M and Franklin, R. 2009 The Australia-Wide Airborne Geophysical Survey - accurate continental magnetic coverage, ASEG, Extended Abstracts, 2009:1, 1-9 Poudjom Djomani, Y., Minty, B.R.S. 2019a. Radiometric Grid of Australia (Radmap) v4 2019 unfiltered pct potassium. Geoscience Australia, eCat reference number 131978. http://dx.doi.org/10.26186/5dd4a7851e8db Poudjom Djomani Y., Minty, B.R.S. 2019b. Radiometric Grid of Australia (Radmap) v4 2019 unfiltered ppm thorium. Geoscience Australia, ecat reference number 131988. http://dx.doi.org/10.26186/5dd4a821a334d Poudjom Djomani, Y., Minty, B.R.S. 2019c. Radiometric Grid of Australia (Radmap) v4 2019 filtered ppm uranium. Geoscience Australia, eCat reference number 131974. http://dx.doi.org/10.26186/5dd48ee78c980

Lineage

Maintenance and Update Frequency: notPlanned

Statement: This product is based on the published 2019 radiometric or gamma-ray grid of Australia. The radiometric grid of Australia (Poudjom Djomani and Minty, 2019a,b,c) is derived by merging over 600 airborne gamma-ray spectrometric surveys by the Commonwealth, State and Territory Governments and held in the national radioelement database of Australia. The cell sizes of the original survey grids range from 50 m through 800 m, but most have a cell size of about 100 m. The original survey grids were re-levelled and then re-sampled to generate the Radiometric Map of Australia grids with a cell size of about 100m (0.001 degrees). Gaps in the 2019 radiometric or gamma-ray grid of Australia have then been infill at the same spatial resolution (0.001 degrees) using a covariant machine learning approach. A boosted regression tree algorithm called XGBoost was used to train relationships between subsets of the airborne measured K, eTh and eU from the Australia Wide Airborne Geophysical Survey (AWAGS) (Milligan et al., 2009) with a suite of covariate datasets that are likely to correlate to the distribution of their radioelements. These covaraites datasets included - satellite imagery, terrain attributes (e.g. slope, relief), gravity (Lane et al, 2020) and surface geology. The predicted grids over the gaps for K, eTh and eU were merged into the existing 2019 using Intrepid Geophysics software.