Machine Learning Models WCSThis web coverage service (WCS) provides access to datasets generated from the application of machine learning for geoscience predictive modelling and knowledge discovery. Datasets include: complete infilled K, Th and U grids, 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; grids from surface conductivity models generated using a machine learning covariate prediction approach based on AEM data; and grids from major oxide models that expand on point measurements of oxide concentration in surface rock and regolith obtained by laboratory assay. Maximum amount of data generated by the WCS is limited to ~1GB.radiometricmachine learninggridpotassiumthoriumuraniumAustraliaweb serviceAEMconductivitysoilsmodellingWCSmajor oxidesurn:ogc:service:wcs2.0.11.1.11.1.0http://www.opengis.net/spec/WCS/2.0/conf/corehttp://www.opengis.net/spec/WCS_protocol-binding_get-kvp/1.0.1http://www.opengis.net/spec/WCS_protocol-binding_post-xml/1.0http://www.opengis.net/spec/WCS_service-extension_crs/1.0/conf/crs-gridded-coveragehttp://www.opengis.net/spec/GMLCOV/1.0/conf/gml-coveragehttp://www.opengis.net/spec/GMLCOV/1.0/conf/special-formathttp://www.opengis.net/spec/GMLCOV/1.0/conf/multiparthttp://www.opengis.net/spec/WCS_service-extension_scaling/1.0/conf/scalinghttp://www.opengis.net/spec/WCS_service-extension_crs/1.0/conf/crshttp://www.opengis.net/spec/WCS_service-extension_interpolation/1.0/conf/interpolationhttp://www.opengis.net/spec/WCS_service-extension_interpolation/1.0/conf/interpolation-per-axishttp://www.opengis.net/spec/WCS_service-extension_interpolation/1.0/conf/nearest-neighborhttp://www.opengis.net/spec/WCS_service-extension_interpolation/1.0/conf/linearhttp://www.opengis.net/spec/WCS_service-extension_interpolation/1.0/conf/cubichttp://www.opengis.net/spec/WCS_service-extension_range-subsetting/1.0/conf/record-subsettingNONE© Commonwealth of Australia (Geoscience Australia) 2021. This product is released under the Creative Commons Attribution 4.0 International Licence. http://creativecommons.org/licenses/by/4.0/legalcodeGeoscience AustraliaClient ServicesGPO Box 378CanberraACT2601Australiaclientservices@ga.gov.auXMLapplication/gml+xmlimage/jpegimage/pngimage/tiffimage/tiff;application=geotifftext/plainhttp://www.opengis.net/def/crs/EPSG/0/4283http://www.opengis.net/def/crs/EPSG/0/4326http://www.opengis.net/def/crs/EPSG/0/3857http://www.opengis.net/def/crs/EPSG/0/20248http://www.opengis.net/def/crs/EPSG/0/20249http://www.opengis.net/def/crs/EPSG/0/20250http://www.opengis.net/def/crs/EPSG/0/20251http://www.opengis.net/def/crs/EPSG/0/20252http://www.opengis.net/def/crs/EPSG/0/20253http://www.opengis.net/def/crs/EPSG/0/20254http://www.opengis.net/def/crs/EPSG/0/20255http://www.opengis.net/def/crs/EPSG/0/20256http://www.opengis.net/def/crs/EPSG/0/20257http://www.opengis.net/def/crs/EPSG/0/20258http://www.opengis.net/def/crs/EPSG/0/28348http://www.opengis.net/def/crs/EPSG/0/28349http://www.opengis.net/def/crs/EPSG/0/28350http://www.opengis.net/def/crs/EPSG/0/28351http://www.opengis.net/def/crs/EPSG/0/28352http://www.opengis.net/def/crs/EPSG/0/28353http://www.opengis.net/def/crs/EPSG/0/28354http://www.opengis.net/def/crs/EPSG/0/28355http://www.opengis.net/def/crs/EPSG/0/28356http://www.opengis.net/def/crs/EPSG/0/28357http://www.opengis.net/def/crs/EPSG/0/28358http://www.opengis.net/def/crs/EPSG/0/32748http://www.opengis.net/def/crs/EPSG/0/32749http://www.opengis.net/def/crs/EPSG/0/32750http://www.opengis.net/def/crs/EPSG/0/32751http://www.opengis.net/def/crs/EPSG/0/32752http://www.opengis.net/def/crs/EPSG/0/32753http://www.opengis.net/def/crs/EPSG/0/32754http://www.opengis.net/def/crs/EPSG/0/32755http://www.opengis.net/def/crs/EPSG/0/32756http://www.opengis.net/def/crs/EPSG/0/32757http://www.opengis.net/def/crs/EPSG/0/32758http://www.opengis.net/def/crs/EPSG/0/3110http://www.opengis.net/def/crs/EPSG/0/3111http://www.opengis.net/def/crs/EPSG/0/3112http://www.opengis.net/def/crs/EPSG/0/3031http://www.opengis.net/def/crs/EPSG/0/3032http://www.opengis.net/def/crs/EPSG/0/3308http://www.opengis.net/def/crs/EPSG/0/3107http://www.opengis.net/def/interpolation/OGC/1/nearest-neighborhttp://www.opengis.net/def/interpolation/OGC/1/linearhttp://www.opengis.net/def/interpolation/OGC/1/cubicAluminium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of aluminium oxide (Al2O3) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__aluminium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Aluminium Oxide Prediction MedianMachine learningml_oxides__aluminium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Aluminium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of aluminium oxide (Al2O3) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__aluminium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Calcium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of calcium oxide (CaO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__calcium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Calcium Oxide Prediction MedianMachine learningml_oxides__calcium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Calcium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of calcium oxide (CaO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__calcium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Iron Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of oxides of iron (Fe2O3tot) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__iron_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Iron Oxide Prediction MedianMachine learningml_oxides__iron_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Iron Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of oxides of iron (Fe2O3tot) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__iron_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Machine learning radiometric map filtered K, Th, UThis grid is a merged three-band (ternary) grid for each of the K, Th and U radiometric infill grids. These 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.radiometricmachine learninggridpotassiumthoriumuraniumAustraliaml__radmap_v4_2019_filtered_ML_KThU_merged_3bandRectifiedGridCoverage112.899914375 -43.7605153.671914375 -8.9995112.899914375 -43.76050004081041153.671914375 -8.999500000938234Machine learning radiometric map filtered percent KThis grid is the potassium (K) radiometric infill and has been generated for regolith (including soils) and geological mapping and can be used as a seamless dataset for predictive modelling using machine learning.radiometricmachine learninggridpotassiumAustraliaml__radmap_v4_2019_filtered_ML_pctkRectifiedGridCoverage112.89991437500001 -43.76050000000001153.671914375 -8.999500000000005112.89991437500001 -43.76050004081041153.671914375 -8.999500000938237Machine learning radiometric map filtered ppm ThThis grid is the thorium (Th) radiometric infill and has been generated for regolith (including soils) and geological mapping and can be used as a seamless dataset for predictive modelling using machine learning.radiometricmachine learninggridthoriumAustraliaml__radmap_v4_2019_filtered_ML_ppmthRectifiedGridCoverage112.89991437500001 -43.76050000000001153.671914375 -8.999500000000005112.89991437500001 -43.76050004081041153.671914375 -8.999500000938237Machine learning radiometric map filtered ppm UThis grid is the uranium (U) radiometric infill and has been generated for regolith (including soils) and geological mapping and can be used as a seamless dataset for predictive modelling using machine learning.radiometricmachine learninggriduraniumAustraliaml__radmap_v4_2019_filtered_ML_ppmuRectifiedGridCoverage112.92091437500001 -43.6365153.63991437500002 -10.0565112.92091437500001 -43.6365000407851153.63991437500002 -10.056500001370653Magnesium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of magnesium oxide (MgO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__magnesium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Magnesium Oxide Prediction MedianMachine learningml_oxides__magnesium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Magnesium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of magnesium oxide (MgO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__magnesium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Manganese Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of manganese oxide (MnO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__manganese_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Manganese Oxide Prediction MedianMachine learningml_oxides__manganese_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Manganese Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of manganese oxide (MnO) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__manganese_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Oxide Covariate ShiftMachine learningml_oxides__national_oxide_covariate_shiftRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 0-4m 5th Percentile GridThis grid represents the lower (5th) percentile of a conductivity prediction model at the 0-4 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_0_4m_conductivity_lower_percentile_5thRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 0-4m 95th Percentile GridThis grid represents the upper (95th) percentile of a conductivity prediction model at the 0-4 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_0_4m_conductivity_upper_percentile_95thRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 0-4m Median GridThis grid represents the median values of a conductivity prediction model at the 0-4 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_0_4m_conductivity_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 30m 5th Percentile GridThis grid represents the lower (5th) percentile of a conductivity prediction model at the 30 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_30m_conductivity_lower_percentile_5thRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 30m 95th Percentile GridThis grid represents the upper (95th) percentile of a conductivity prediction model at the 30 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_30m_conductivity_upper_percentile_95thRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity 30m Median GridThis grid represents the median values of a conductivity prediction model at the 30 meter depth interval, trained using a national compilation of AEM survey line data. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. The grid is part of a sequence of two conductivity depth interval models including; 0-4m and 30m. Over 460K training points/measurements were used in a 5 K-Fold training and validation split. A further 28626 points/measurements were used to assess the out of sample performance (i.e. points not used in the model validation). Modelling was preformed using the gradient boosted (GB) tree algorithm (https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.GradientBoostingRegressor.html). We use the median of the models as the conductivity prediction and the upper and lower percentiles (95th and 5th) to measure the model uncertainty. Grids show conductivity (S/m) in log 10 units. Reported out-of-sample r-squares for the 0-4m and 30m depths are 0.75 and are 0.73, respectively. Where we don’t have AEM survey coverage the model is finding relationships with the covariates and predicting into those areas. Where those relationships are not well understood (i.e. where we see a departure in the feature space characteristics from what the model can ‘see’) the model prediction is likely to be less certain. Differences in the features the model uses in training and the full feature space is captured covariate shift map. Users need to be mindful of the uncertainties shown by the 5th-95th percentiles and high values in the covariate shift map.AEMml_nat_conductivity__national_30m_conductivity_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766National Surface Conductivity Covariate Shift GridThe national conductivity prediction models have been generated using a covariate machine learning approach. The covariate shift grid shows where the covariate model relationships are learned, indicated by shift values approaching 0, and where we see a departure from the covariate features, indicated by shift values approaching 1 (i.e. covariate values not ‘seen’ by the model training).AEMml_nat_conductivity__national_surface_conductivity_covariate_shiftRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Northern Australia Surface Conductivity 0-50cm 5th Percentile GridThis grid represents the lower (5th) percentile of a conductivity prediction model at the 0-50cm depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__0_50cm_lowerRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 0-50cm 95th Percentile GridThis grid represents the upper (95th) percentile of a conductivity prediction model at the 0-50cm depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__0_50cm_upperRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 0-50cm Median GridThis grid represents the median values of a conductivity prediction model at the 0-50cm depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the greater the uncertainty.AEMml_nthaus_conductivity__0_50cm_medianRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 22-27m 5th Percentile GridThis grid represents the lower (5th) percentile of a conductivity prediction model at the 22-27m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__22_27m_lowerRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 22-27m 95th Percentile GridThis grid represents the upper (95th) percentile of a conductivity prediction model at the 22-27m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__22_27m_upperRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 22-27m Median GridThis grid represents the median values of a conductivity prediction model at the 22-27m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the greater the uncertainty.AEMml_nthaus_conductivity__22_27m_medianRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 9-11m 5th Percentile GridThis grid represents the lower (5th) percentile of a conductivity prediction model at the 9-11m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__9_11m_lowerRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 9-11m 95th Percentile GridThis grid represents the upper (95th) percentile of a conductivity prediction model at the 9-11m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the, greater the uncertainty.AEMml_nthaus_conductivity__9_11m_upperRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity 9-11m Median GridThis grid represents the median values of a conductivity prediction model at the 9-11m depth interval for the Northern Australian region, generated from a machine learning covariate prediction approach based on the Northern Australian regional Aus-AEM survey dataset. The grid is part of a sequence of three conductivity depth interval models including; 0-50cm, 9-11m and 22-27m. Machine learning is used to find predictive relationships between the AEM depth conductivities and a suite of national environmental covariates (e.g. radiometrics, satellite imagery, terrain attributes) with higher spatial resolution. The machine learning trains on the conductivity values derived from the AEM depth inversions and predicts conductivities at the resolution of the covariates which are at 85m resolution. Reported out-of-sample r-squares for each interval in order of increasing depth are 0.74, 0.64, and 0.67. The model prediction equates to the median grid and the 5th and 95th percentiles provide an estimate of the prediction uncertainty. The higher the inter-percentile range the greater the uncertainty.AEMml_nthaus_conductivity__9_11m_medianRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Northern Australia Surface Conductivity Covariate Shift GridConductivity grids have been generated using a covariate machine learning approach. The covariate shift grid shows where the covariate model relationships are learned, indicated by shift values approaching 0, and where we see a departure from the covariate features, indicated by shift values approaching 1 (i.e. covariate values not ‘seen’ by the model training).AEMml_nthaus_conductivity__Cov_shiftRectifiedGridCoverage-1555680.0 -2666375.3460527651596571.074089495 -1052892.0116.53165280234649 -24.676787741228072147.87081529772945 -9.221628000620374Phosphorus Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of phosphorus oxide (P2O5) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__phosphorus_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Phosphorus Oxide Prediction MedianMachine learningml_oxides__phosphorus_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Phosphorus Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of phosphorus oxide (P2O5) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__phosphorus_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Potassium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of potassium oxide (K2O) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__potassium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Potassium Oxide Prediction MedianMachine learningml_oxides__potassium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Potassium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of potassium oxide (K2O) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__potassium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Silicon Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of silicon oxide (SiO2) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__silicon_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Silicon Oxide Prediction MedianMachine learningml_oxides__silicon_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Silicon Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of silicon oxide (SiO2) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__silicon_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Sodium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of sodium oxide (Na2O) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__sodium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Sodium Oxide Prediction MedianMachine learningml_oxides__sodium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Sodium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of sodium oxide (Na2O) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__sodium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Titanium Oxide Lower PercentileThis layer provides a seamless national grid of the lower (5th) percentile of titanium oxide (TiO2) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__titanium_oxide_lower_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Titanium Oxide Prediction MedianMachine learningml_oxides__titanium_oxide_prediction_medianRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766Titanium Oxide Upper PercentileThis layer provides a seamless national grid of the upper (95th) percentile of titanium oxide (TiO2) concentrations in surface rock and regolith, generated using a machine learning approach that expands on geochemical point measurements of oxide concentration in rock or regolith samples using correlations with national covariates. The upper and lower percentiles (95th and 5th) provide a measure of model uncertainty.Machine learningml_oxides__titanium_oxide_upper_percentileRectifiedGridCoverage-2000000.0 -4900000.02200000.0 -1050000.0108.06971127443157 -44.86805196721959158.25939863332658 -8.45788596860766