{
    "componentChunkName": "component---src-gatsby-entities-sub-topic-tsx",
    "path": "/sub-topics/heterogeneity-modelling",
    "result": {"data":{"platform":{"subTopic":{"__typename":"Platform_SubTopic","_schema":{"label":"Sub Topic","pluralLabel":"Sub Topics"},"openGraph":{"title":"Heterogeneity Modelling","description":{"text":"Heterogeneity Block Modelling Introduction\n==============================================\n\nTo enable an analysis of the [heterogeneity](/pages/wikiore-heterogeneity) and any subsequent [bulk ore sorting](/pages/wikiore-bulk-sorting) <b title=\"Materials response to a..."},"image":null},"id":"65d2edd7ebf7bb9388ffb16d","slug":"heterogeneity-modelling","name":"Heterogeneity Modelling","path":"/sub-topics/heterogeneity-modelling","outlineNumber":"2.3.1","color":{"__typename":"Platform_Color","id":"65d2ed7febf7bb9388ffab3a","slug":"core-12","name":"Core 12","value":"#3b6673","darkValue":"#2b4b56","veryDarkValue":"#1c3239"},"intro":{"id":"65d2edd7ebf7bb9388ffb16d_intro","text":"Heterogeneity Block Modelling Introduction\n==============================================\n\nTo enable an analysis of the [heterogeneity](/pages/wikiore-heterogeneity) and any subsequent [bulk ore sorting](/pages/wikiore-bulk-sorting) <b title=\"Materials response to a given preconcentration technology.\">amenability</b> , a model is required to highlight the local variability in addition to an un-biased grade estimate.  "},"description":{"id":"65d2edd7ebf7bb9388ffb16d_description","text":"New [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model)s are required that focus on identifying [heterogeneity](/pages/wikiore-heterogeneity) rather than grade directly.  However, grade or “[assay](https://en.wikipedia.org/wiki/Assay)” data is by far the most extensively available empirical dataset and is logically used to evaluate [heterogeneity](/pages/wikiore-heterogeneity) potential.The key underlying and intrinsic value driver for positive [bulk ore sorting](/pages/wikiore-bulk-sorting) opportunity, is adequate [heterogeneity](/pages/wikiore-heterogeneity) within a mining volume.  Without this [heterogeneity](/pages/wikiore-heterogeneity) a bulk ore “sense and divert” system cannot separate higher value “pods” of ore (a pod can be considered the equivalent of a selective mining unit or SMU) from adjacent delivered “pods” of low value [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) or waste. The “pods” should be considered a scalar feature and have the ability to vary from 10’s of tonnes (eg: 30 second <b title=\"Length of time for continual measurement in sensor applications before a destination is determined.\">integration time</b> on a conveyor), to “bucket” sized 60-90t, or [truck](https://en.wikipedia.org/wiki/Haul_truck) scale of approximately 300t.\n\n\nModel Limitations[](#Model_Limitations)\n---------------------------------------\n\nLinear estimation techniques such as ordinary [kriging](https://en.wikipedia.org/wiki/Kriging) (OK)  common in many [resource estimation](https://en.wikipedia.org/wiki/Mineral_resource_estimation)s or [inverse distance weighting](https://en.wikipedia.org/wiki/Inverse_distance_weighting) (such as ID squared – ID 2 ) are commonly utilised.  However there is a requirement to account with sample or point data that is generally positively skewed in most mining environments.  This is further complicated where the desired <b title=\"Smallest Mining Unit (a.k.a Minimum Mining Unit MMU)\">SMU</b> is often far smaller than the actual input sample spacing (eg: drillholes).  Therefore any attempts to generate small blocks (or SMU’s) that are at a scale suitable for [bulk ore sorting](/pages/wikiore-bulk-sorting) assessments (eg: 10’s tonnes on conveyor’s to 300t for [truck](https://en.wikipedia.org/wiki/Haul_truck)s), would generate very high and unusable estimation variances per block.  This would likely result in unrealistic grade tonnage curves, that differ significantly from the true <b title=\"A spatially constrained unit within the resource model.\">block</b> valuesordinary [kriging](https://en.wikipedia.org/wiki/Kriging) is also impacted by outliers which are important in identifying [heterogeneity](/pages/wikiore-heterogeneity) or variability.  Overall an OK model aims to achieve an unbiased model globally, but will smooth the <b title=\"Spatial or temporal distribution of grade.\">grade profile</b> , reduce the variability ([heterogeneity](/pages/wikiore-heterogeneity)), and will not preserve the input sample distribution.  This is a natural consequence of the methodology and is suitable for resources, reserves and most planning purposes.The key driver or input into any model are the drillholes and the impact they have on change of support. Vann & Guibal (1998) summarise support as the impact on any volume (eg: a fixed [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model) unit) that is impacted by changes in size, shape or orientation.  A simple example is to use different sizes (supports) in the same volume.  Different sized would generate different histograms (distributions), but the mean will stay the same.\n\nIn the image below, this is explained by the use of drillholes and their common resultant [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model) output.  Drill spacing is commonly widely spaced (eg: 50m) as in “A”, which results in [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model)s in the order of 25m x25m as shown in “B”. Note the mean of the blocks in “B” is 1.01%.  However, the desired output to assist in a [heterogeneity](/pages/wikiore-heterogeneity) study would be a [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model) with much smaller SMU’s (or blocks) such as “C”.  This would require <b title=\"Exploration and infill drill hole data (typically diamond drill core).\">resource drilling</b> in the order of 12.5 m spacing as shown in “D” which is not a viable option.  The figure blow shows the impact of change of support as both examples average 1.01% copper, however maximum and minimum in “B” are 1.11% and 0.94% respectively, while “C” has 1.76% and 0.18% respectively.  Or in other words the change of support has smoothed the variability (and sample histogram) and has vastly reduced the [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) of the <b title=\"Specialised block models determined from high frequency data (i.e. blast holes).\">high resolution model</b> from 0.32  to 0.08 in the <b title=\"Subjective term for the larger sized component of feed\">coarse</b> resolution model.  \n\n [![Support Effect](https://res.cloudinary.com/shapeable/image/upload/v1687303452/wikiore/192.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303452/wikiore/192.jpg) \n\nThe above only references the impacts of change in support in a simple 2-dimensional view (X and Y).  In the above example even a <b title=\"A spatially constrained unit within the resource model.\">block</b> of 6.25m x 6.25m x 15m (15m in Z direction) approximates to 1500 tonnes.  A <b title=\"A spatially constrained unit within the resource model.\">block</b> of 6.25m x 6.25m x 6.25m (XYZ) equates to approximately 600 tonnes (or 2 [truck](https://en.wikipedia.org/wiki/Haul_truck)s) which is still orders of magnitude larger than the opportunity presented in [bulk ore sorting](/pages/wikiore-bulk-sorting). As noted earlier <b title=\"Exploration and infill drill hole data (typically diamond drill core).\">resource drilling</b> provides far higher resolution in the Z dimension.  Compositing drill hole data into larger intervals has the same change of support effect as discussed. In the below figure, compositing the drill samples tends to smooth the results, impacts on the outliers and reduces the variability.\n\n\n [![Block Model Change Support Histo Example](https://res.cloudinary.com/shapeable/image/upload/v1687303453/wikiore/193.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303453/wikiore/193.jpg) \n\nOrdinary kriged models are commonly at a <b title=\"Subjective term for the larger sized component of feed\">coarse</b>r scale than is generally useful for a [heterogeneity](/pages/wikiore-heterogeneity) assessment where a key input is the [heterogeneity](/pages/wikiore-heterogeneity) signature.  Ordinary kriged models provide excellent grade prediction for resource estimates but are restricted in their applicability for local preservation of [heterogeneity](/pages/wikiore-heterogeneity).To overcome this, CRC ORE chose to utilise [non-linear estimation](https://www.cubeconsulting.com/wp-content/uploads/2020/11/Technical-Note-What-is-Non-Linear-Estimation_FINAL.pdf) techniques such as Uniform Conditioned, Localised Uniform Conditioned and [conditional simulation](https://www.whittleconsulting.com.au/wp-content/uploads/2018/05/Conditional-Simulations-Tools-for-Modelling-Uncertainty-in-Open-Pit-Optimisation.pdf) models to honour the <b title=\"Smallest Mining Unit (a.k.a Minimum Mining Unit MMU)\">SMU</b> histogram and provide smaller <b title=\"A spatially constrained unit within the resource model.\">block</b> resolution to assist in identifying [heterogeneity](/pages/wikiore-heterogeneity) signatures.  The estimation methods are common, readily accepted estimation techniques utilised in the mining industry.  The process can be readily adopted and replicated by any of the mining estimation software packages.  The [heterogeneity](/pages/wikiore-heterogeneity) is further supported by also analysing a suite of metrics readily calculated from standard [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model)s as identified in the table below.  The key goal is to identify the [heterogeneity](/pages/wikiore-heterogeneity) signature, rather than the absolute metal content, or to utilise any non-industry standard estimation processes.To assist in the validation of the models, <b title=\"Specialised block models determined from high frequency data (i.e. blast holes).\">high resolution model</b>s (HRM) were created from high density <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data (nominally 6m x 7m x 15m in XYZ). Creating a model from the complete <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data set is most likely to closely represent the underlying true grade distribution on a <b title=\"Individual levels within an open pit.\">bench</b> -by-<b title=\"Individual levels within an open pit.\">bench</b> basis. The model is considered to be the high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> and provides an index for analysing variability, [heterogeneity](/pages/wikiore-heterogeneity) and for referencing against other models created from wider-spaced sample data.There are some limitations with this approach. Models and estimates are only as reliable as the input data and the assumptions and parameters used during estimation. While a dense data coverage mitigates against model-induced estimation artefacts, other factors affect the differences between models created from different types of data. These factors include:\n\n*   The differences in sample support between 15m <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s and diamond drill hole samples, even after compositing;\n*   <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b> boundary location uncertainty which itself is a function of both data density and quality;\n*   The statistical data representivity of the data and correct representation of the histogram and [variogram](https://en.wikipedia.org/wiki/Variogram) by the data; and\n*   The sample population statistics and sample bias (or [conditional bias](http://www.ccgalberta.com/ccgresources/report04/2002-115-conditionalbias.pdf)).\n\n\nThe high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model was estimated using ordinary [kriging](https://en.wikipedia.org/wiki/Kriging) into 3.125m x 3.125m x 15m (XYZ) parent blocks based on [variogram](https://en.wikipedia.org/wiki/Variogram) models derived from the data and search neighbourhoods designed to minimise the smoothing and averaging associated with kriged estimates. The [block model](https://en.wikipedia.org/wiki/Mineral_resource_estimation#Resource_block_model) origin and extent were aligned with the <b title=\"Digitised model of the deposit.\">resource model</b> estimate to improve downstream comparisons. The model was restricted to a buffer of within 30m of existing <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> samples to minimise extrapolation.  The same [heterogeneity](/pages/wikiore-heterogeneity) metrics were generated in the <b title=\"Specialised block models determined from high frequency data (i.e. blast holes).\">high resolution model</b> as the <b title=\"Digitised model of the deposit.\">resource model</b> for comparison.\n\nVariability Metrics[](#Variability_Metrics)\n===========================================\n\nVariability and [heterogeneity](Heterogeneity)[](#Variability_and_heterogeneity)\n--------------------------------------------------------------------------------\n\nThe high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model was developed as part of the assessment of the potential value of exploiting the natural variability of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29). Central to this assessment is using the model to identify not only the possible improvement by selecting smaller ore parcels but also using the model to identify preferred locations for the application of the improvement technology. Identifying where there is an economic value requires a measure that is a proxy for value potential. The value proxy is likely to be related to the nature and quantum of [grade variability](/pages/wikiore-heterogeneity) within a local neighbourhood.\n\nAt this stage the best proxy measure is unknown and is subject to further research. Consequently, a number of simple measures of [grade variability](/pages/wikiore-heterogeneity) derived from the distribution of 3.125m x 3.125m blocks within larger 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s. While these variability metrics are not necessarily the best predictor of value, they provide a starting point for further investigation and analysis. Each metric is described in Table 3 and Section 3.2.  \n\nDescription of included metrics[](#Description_of_included_metrics)\n-------------------------------------------------------------------\n\n|Field|Description|\n|-----|-----------|\n|`M%_12_5`|metal percentage in first 12.5% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_25_0`|metal percentage in first 25% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_37_5`|metal percentage in first 37.5% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_50_0`|metal percentage in first 50% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_62_5`|metal percentage in first 62.5% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_75_0`|metal percentage in first 75% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_87_5`|metal percentage in first 87.5% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`M%_100_0`|metal percentage in first 100% of 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`SUB_VAR`|variance of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`SB_STDEV`|[standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`SUB_CV`|coefficient of variation of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`25_METAL`|first quartile metal  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`75_METAL`|third quartile metal  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`25_CuT`|first quartile grade of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`75_CuT`|third quartile grade of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`IQR`|Interquartile range (CuT grade)  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`SUB_SKEW`|Skewness  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`QtCoDisp`|Quartile Coefficient of Dispersion  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`COUNT`|Number of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`mid_RNG`|Mid-range  of 3.125x3.125x15 blocks in 25x25x15 <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>|\n|`VAR`|Variance of <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> grades in the estimation neighbourhood|\n|`VARZSTR`|Estimation variance of <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> grade estimate|\n|`DIFF`|sum of differences between calculated <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> and a 1:1 mass  /metal correlation|\n|`ABSDIFF`|sum of absolute differences between calculated <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> and a 1:1 mass/metal correlation|\n\n\n**`M%_12_5` – `M%_100_0`**: These fields together represent the metal frequency distribution with in the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>, discretised on 12.5% tonnage intervals. <b title=\"A spatially constrained unit within the resource model.\">Block</b> metal content is sorted from low to high and then accumulated sequentially. This distribution is equivalent to the <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> from the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . That is, the metal recovered from selecting the first 12.5% to tonnes in the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> followed by the second 12.5% of tonnes until the entire <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> is selected. These metrics assume perfect selection within the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . There is no account for the spatial distribution of the discretised intervals.\n\n\n**`SUB_VAR`:** The variance of the 3.125m x 3.125m x 15m blocks inside the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Defined as the average squared difference between the 3.125m blocks and the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> average grade.\n\n**`SUB_STDEV`:** The [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) of the 3.125m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Defined as the square root of the variance.\n\n**`SUB_SKEW`:** The skewness of the 3.125m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Defined as the average cubed difference between the 3.125m blocks and the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> average grade.\n\n**`SUB_CV`:** The coefficient of variation of the 3.125m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Defined as the [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) (SUB\\_STDEV) divided by the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> average grade\n\n**`25_METAL` and `75_METAL`**: The first and third quartile of the 3.15m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> on a contained metal basis. The <b title=\"A spatially constrained unit within the resource model.\">block</b> metal is calculated for all blocks in the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> and then the metal quartiles reported. This is a measure of the spread of metal in the blocks.\n\n**`25_CuT` and `75_CuT`:** The first and third quartile of the 3.15m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> on a grade-only basis. This is a measure of the spread of grades in the blocks.\n\n**`IQR`:** The inter-quartile range (on a grade basis) of the 3.15m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . That is, the 75th quartile less the 25th quartile. This is a measure of spread and centrality.\n\n**`QtCoDisp`**: Quartile coefficient of dispersion. The inter-quartile range (IQR) divided by the sum of the 75th and 25th quartiles. This is a normalised version of the IQR similar to the normalised [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) represented by the coefficient of variation (CV). It is a measure of spread and centrality that can be used for comparative purposes. Since to IQR and the differences in the 75th and 25th quartiles are both expressed in units of grade, the quartile coefficient of dispersion is unitless and therefore a relative measure.\n\n**`COUNT`:** A simple count of the number of blocks per <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Used as a reference for the reliability of other metrics. In most cases only <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s on the edge of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) have less than 64 blocks.\n\n**`MID_RNG`:** The mid-range of the 3.15m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> on a grade-only basis. The mid-range is calculated as the sum of the minimum <b title=\"A spatially constrained unit within the resource model.\">block</b> grade and the maximum <b title=\"A spatially constrained unit within the resource model.\">block</b> grade divided by 2. It is a simple measure of centrality and skewness, commonly compared to the average.\n\n**`VAR`:** Variance of <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> grades in the estimation neighbourhood. That is, the variance of the grades of all the samples found within the search neighbourhood for each 3.125m x 3.125m x 15m block. This is a measure of the local distribution of variance commonly used as an indicator of deviation from <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b> stationarity.\n\n**`VARZSTR`:** The estimation variance, or mean squared error variance of the estimate. Defined as the variance of the difference between the true <b title=\"A spatially constrained unit within the resource model.\">block</b> grade and the estimate <b title=\"A spatially constrained unit within the resource model.\">block</b> grade. The estimation variance is calculated in the solution of the ordinary[kriging](https://en.wikipedia.org/wiki/Kriging) equations. The estimation variance is a measure of the quality of the estimate. A low [kriging](https://en.wikipedia.org/wiki/Kriging) variance means the variance of the <b title=\"A spatially constrained unit within the resource model.\">block</b> is approximately equal to the true <b title=\"A spatially constrained unit within the resource model.\">block</b> grade. A high [kriging](https://en.wikipedia.org/wiki/Kriging) variance means the variance of the estimate is relatively high compared to the true <b title=\"A spatially constrained unit within the resource model.\">block</b> variance. Note the estimation variance is not normalised and therefore should not be used as a block-to-<b title=\"A spatially constrained unit within the resource model.\">block</b> comparison metric.\n\n**`DIFF`:** The sum of the differences between a linear <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> (1:1) and the modelled <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> of the 3.125m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Calculated by subtracting `M%_12_5` to `M%_87.5` from a 1:1 curve. Not the differences are not absolute and therefore negative differences will be cancelled by positive differences.\n\n**`ABSDIFF`:** The sum of the absolute differences between a linear <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> (1:1) and the modelled <b title=\"Relative proportion of feed mass in the accept stream after preconcentration.\">mass pull</b> of the 3.125m x 3.125m x 15m blocks in the 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . Calculated by subtracting `M%_12_5` to `M%_87.5` from a 1:1 curve. This represents the deviation of the blocks within a <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> from a homogenous distribution and is a measure of separation ability.  \n\n\nLUC and High Resolution Blast Hole Model Comparisons[](#LUC_and_High_Resolution_Blast_Hole_Model_Comparisons)\n-------------------------------------------------------------------------------------------------------------\n\nA set of localised uniform conditioning (LUC) estimates was developed using resource drill hole samples only – <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s were excluded from the <b title=\"Local Uniform Conditioning\">LUC</b> models. One of the main objectives of the High resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> Modelling project was to assess the ability to model and estimate [grade variability](/pages/wikiore-heterogeneity) (i.e., [heterogeneity](/pages/wikiore-heterogeneity)) from widely-spaced resource drill hole samples compared to the more closely-spaced <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> samples. This ability to predict ‘future ores’ or the variability of material to be mined in the future is a key success factor for generating value. If a good estimate of the likely variability performance of future mining areas cannot be developed from the existing resource-spaced drill hole data, then alternative modelling and/or data collection strategies must be developed. Like-for-like, overlapping regions from the ground-truth and <b title=\"Local Uniform Conditioning\">LUC</b> models was created for comparison. Because the <b title=\"Local Uniform Conditioning\">LUC</b> models were based on larger <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s (50m x 50m x 15m XYZ) the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model variability metrics were re-calculated to this <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> size. That is, all of the 3.125m x 3.125m x 15m blocks within each 50m x 50m x 15m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> were used to determine the variability metrics. Therefore, the metrics in each <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> are supported by 256 <b title=\"A spatially constrained unit within the resource model.\">block</b> values.\n\nThe larger <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> size should result in a more reliable statistical value, albeit at the loss of some spatial resolution. The same metrics were developed for the ground-truth model and the <b title=\"Local Uniform Conditioning\">LUC</b> model. The <b title=\"Local Uniform Conditioning\">LUC</b> model metrics were derived from the blocks within each <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> estimated by the ‘localisation’ phase of the uniform conditioning. The <b title=\"A spatially constrained unit within the resource model.\">block</b> localisation was by discretising the uniform conditioned <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> <b title=\"Graph of cumulative tonnes and grade within a specific confine (e.g. pit, year etc…).\">grade-tonnage curve</b> and assigning the <b title=\"A spatially constrained unit within the resource model.\">block</b> discretised values on the basis of ranking. Individual blocks within each <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> were estimated using ordinary [kriging](https://en.wikipedia.org/wiki/Kriging) of the resource data. These kriged values were ranked from highest to lowest and the uniform conditioning distribution assigned accordingly. This localisation approach, while a standard industry practice, is not the only possible method. It relies on ranking the grade values of blocks within a <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>. The ranking is dependent on the underlying kriged <b title=\"A spatially constrained unit within the resource model.\">block</b> grades. This approach works reasonably well where the sample data is not too widely spaced and there is a reasonable amount of data. When the data is more widely spaced, the localisation approach can (and does) result in artefacts reflecting the similarity of samples selected during the <b title=\"A spatially constrained unit within the resource model.\">block</b> estimation. If all the blocks within a <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> are estimated using the same group of samples (i.e. the search selects the same data) the only difference in <b title=\"A spatially constrained unit within the resource model.\">block</b> grade (and hence ranking) is the relative block-to-sample distance and orientation. This can be problematic. Caution should be taken if using the outcomes of the <b title=\"Local Uniform Conditioning\">LUC</b> for assessing variability where the samples are widely spaced (on the order of more than 3 times the average sample spacing).  \n \n\nComparison Methodology[](#Comparison_Methodology)\n-------------------------------------------------\n\nComparing models is never an easy task. Even comparing straightforward resource estimates usually involves some degree of averaging and/or comparing global outcomes (e.g., the <b title=\"Graph of cumulative tonnes and grade within a specific confine (e.g. pit, year etc…).\">grade-tonnage curve</b> ). A direct block-to-<b title=\"A spatially constrained unit within the resource model.\">block</b> comparison of two estimates is almost always a meaningless exercise because of the 3D [geomet](https://en.wikipedia.org/wiki/Geometallurgy)ries involved and the subtle impact of changes in input data, search strategies and other parameters. Estimated grades may shift over small distances in 3D and yet the overall comparison may show that two models are very similar in their efficacy. In this case, comparing different models of spatial variability, it was necessary to correctly frame the problem. The key question is – do the models predict the same (overlapping) regions of high/low spatial variability? We are not interested in the absolute value of any of the variability metrics themselves, instead we are interested in understanding if the metric map similar regions. With this in mind, the approach to comparing the models involved:\n\n*   Calculating the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> -by-<b title=\"A discrete section within a resource model (i.e. block).\">panel</b> variability metrics as described in Section 1.1 for both the ‘ground-truth’ model and overlapping volumes of <b title=\"Local Uniform Conditioning\">LUC</b> model;\n*   Calculating the statistics for a selection of the variability metrics (CV, QtCofDisp, IQR, Mid\\_RNG, MidRngRT). In particular the 75th  and 90th percentiles;\n*   Copying only <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s above the 75th and 90th percentiles;\n*   Measuring the degree of overlap for <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s above these two thresholds.\n\n  \nThis approach gives four combinations of <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s (Table 2) for each metric considered; Types 1 and 4 where the models both agree on the variability classification and Types 2 and 3 where they disagree.\n\n|Class|LUC Model|Ground Truth Model|Outcome|\n|-----------------------------------------------------|\n|Type 1|Above threshold|Above threshold|Models agree|\n|Type 2|Above threshold|Below threshold|Models disagree|\n|Type 3|Below threshold|Above threshold|Models disagree|\n|Type 4|Below threshold|Below threshold|Models agree|\n\n  \nThe comparison illustrates that each of the examined metric performs differently with the greatest overall agreement being for the Mid-Range. This simple metric (the maximum <b title=\"A spatially constrained unit within the resource model.\">block</b> grade less the minimum <b title=\"A spatially constrained unit within the resource model.\">block</b> grade divided by 2) highlights <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s where the individual <b title=\"A spatially constrained unit within the resource model.\">block</b> grades have the greatest overall spread. Perhaps unsurprisingly, the volumes identified as high variability are different depending on the metric selected. For example, the Mid-Range metric appears to highlight areas in the centre of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) where there are extremely high grades (hence a high mid-range); whereas, the CV metric appears to highlight areas on the edges of the high-grade core of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) where there is a distinct edge-related mixing of high and low grades. This demonstrates the importance of understanding and correctly framing the question before selecting a metric. In this case, the mid-range metric would be excellent for identifying <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s where there are a wide range of <b title=\"A spatially constrained unit within the resource model.\">block</b> grades – regardless of the maximum grade in the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> . The CV metric on the other hand is ‘normalised’ by the average <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> grade and therefore a high CV indicates a high relative variability within the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> (i.e., high contrast)  \n \n\n|Threshold|Class|Mid Range RT|Mid Range|QtCoDisp|IQR|CV|\n|-------------------------------------------------------------------|\n|Average|Type 1|60%|29%|19%|27%|18%|\n|               |Type 2|40%|10%|32%|11%|8%|\n|               |Type 3|0%|4%|18%|8%|21%|\n|               |Type 4|0%|57%|31%|53%|53%|\n|75th Percentile|Type 1|5%|20%|8%|17%|13%|\n|           |Type 2|33%|7%|18%|10%|11%|\n|           |Type 3|8%|4%|17%|8%|12%|\n|           |Type 4|54%|69%|57%|65%|64%|\n|90th Percentile|Type 1|0%|7%|2%|5%|3%|\n|           |Type 2|0%|4%|9%|6%|7%|\n|           |Type 3|0%|3%|8%|5%|7%|\n|           |Type 4|100%|87%|81%|84%|83%|\n\nFigure 1 and Figure 2 illustrate the different type classes for the CV metric. Figure 3 and Figure 4 illustrate the same <b title=\"Individual levels within an open pit.\">bench</b>es for the Mid-Range matric.  \n \n\n [![GEblockCVmetricType1](https://res.cloudinary.com/shapeable/image/upload/v1687303455/wikiore/167.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303455/wikiore/167.jpg) \n\n_Figure 1. CV metric - TYPE 1 class based on 75th percentile (black outlines) (3040mRL)_\n\n [![GEblockCVmetricTypeClass](https://res.cloudinary.com/shapeable/image/upload/v1687303457/wikiore/168.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303457/wikiore/168.jpg) \n\n_Figure 2. 3040m RL coloured by TYPE CLASS based on CV metric at 75th Percentile_\n\n  \nBased on the preliminary analysis, the degree of agreement between the ground-truth metrics and the <b title=\"Local Uniform Conditioning\">LUC</b> derived metrics is dependent on a number of factors. These include the data density in the <b title=\"Local Uniform Conditioning\">LUC</b> estimate, the data-to-<b title=\"A spatially constrained unit within the resource model.\">block</b> configuration and the quality of the <b title=\"Local Uniform Conditioning\">LUC</b> model itself. In addition, metrics that have some degree of ‘normalisation’ to the mean grade of the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> appear more robust. There is also an impact related to the classification threshold. The higher the threshold applied, the better the classification agreement between the ground-truth and <b title=\"Local Uniform Conditioning\">LUC</b> models.  \n \n\n [![GEblockMIDmetricType1](https://res.cloudinary.com/shapeable/image/upload/v1687303459/wikiore/169.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303459/wikiore/169.jpg) \n\n_Figure 3. Mid-Range metric - TYPE 1 class based on 75th percentile (black outlines) (2965mRL)._\n\n [![GEblockMIDmetricTypeClass](https://res.cloudinary.com/shapeable/image/upload/v1687303461/wikiore/170.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303461/wikiore/170.jpg) \n\n_Figure 4. 2965m RL coloured by TYPE CLASS based on Mid-Range metric at 75th Percentile._\n\nPredicting Local Variability[](#Predicting_Local_Variability)\n=============================================================\n\nLimitations of current approach[](#Limitations_of_current_approach)\n-------------------------------------------------------------------\n\nThe high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model is based on closely spaced <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data. This is the most detailed data available prior to blasting and ore digging; however, there are still limitations with the approach adopted including:\n\n*   The <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data has limited resolution in the vertical dimension. Each sample represents the average grade of the entire 15m <b title=\"Individual levels within an open pit.\">bench</b> at the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> location. Any variability within the vertical dimension will not be captured in the high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> estimate. Given the observable variability in the XY plane and the nature of the dataset [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) it is highly likely that a material amount of [grade variability](/pages/wikiore-heterogeneity) occurs vertically. If the is intent on assessing the value of ore/waste separation methods that exploit within <b title=\"Individual levels within an open pit.\">bench</b> variability, the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model will not be a reliable estimate;\n*   The estimate is based on <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s derived from the statistical distribution of grades in the blast  hole data. Ideally estimation <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s should be related to the underlying <b title=\"Imprinting events on geology (e.g. faults, intrusions etc…).\">geological controls</b> . Using grade-based <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s may cause estimation artefacts and incorrectly over or under represent variability at the model scale. Ideally a two-stage <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>ing approach is recommended:\n    *   In the first stage estimation (grade estimation) <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s should be based on zones with similar geological and statistical properties (i.e., stationarity); and\n    *   The second stage is the development of specific ‘variability <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s’ to highlight and isolate regions where the [grade variability](/pages/wikiore-heterogeneity) is high/low compared to the immediate neighbourhood. These ‘variability <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s’ should form the basis of a second phase of modelling where the primary concern is developing an unbiased models of the preferred variability metric; and\n*   The model uses ordinary [kriging](https://en.wikipedia.org/wiki/Kriging), a linear estimator. While the closely spaced <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data ameliorates the smoothing and averaging associated linear estimation, the true [grade variability](/pages/wikiore-heterogeneity) is likely to be under-stated by the [kriging](https://en.wikipedia.org/wiki/Kriging) process.\n\nAlternatives and challenges[](#Alternatives_and_challenges)\n-----------------------------------------------------------\n\nPredicting the variability of small parcels (e.g., 200 tonnes or less) based on small, widely spaced samples is challenging. There are two associated concepts that must be considered:\n\n1.  The true [grade variability](/pages/wikiore-heterogeneity) – an aspect of the geology and [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) system itself. Variability is part of the physical world. It can be measured, analysed and where appropriate explained; and\n2.  The uncertainty of our understanding and models of the true variability. Uncertainty is related to knowledge and data. Incomplete (non-exhaustive) data and modelling both affect our knowledge of variability. In the absence of complete data, we create models in regions that are not sampled. The modelling method (and the parameters applied in the modelling phase) impacts on the apparent variability, resulting in either an over or under estimation of the truth.\n\n [![GEblockVariabilityVsUncertainty](https://res.cloudinary.com/shapeable/image/upload/v1687303462/wikiore/173.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303462/wikiore/173.jpg) \n\n_Variability vs. Uncertainty._\n\n  \nThis concept is easily demonstrated using the ground-truth model developed from the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data. As an experiment, we can treat the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model as ‘reality’, that is, we can say that the true grade and [grade variability](/pages/wikiore-heterogeneity) is represented by the model. For example, consider a slice through the model at 2800m RL (Figure below). Based on this as ‘reality’ we can quantify the ‘real’ [grade variability](/pages/wikiore-heterogeneity) in 25m x 25m <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>s.\n\n[![GEblockHighResBHmodel](https://res.cloudinary.com/shapeable/image/upload/v1687303464/wikiore/172.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303464/wikiore/172.jpg)\n\n_High resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> model slice at 2800m RL._ \n\n  \nThe next step in this experimental demonstration is to ‘sample the reality’ to create a subset of sample data. This is equivalent to drilling the deposit. Because this is an experimental demonstration, we can apply different sample patterns and density. These sampling patterns can then be used to estimate the [grade variability](/pages/wikiore-heterogeneity) and these estimates can be compared to ‘reality’.  \n \n\n [![GEblockImpactOfSparseOnLuc](https://res.cloudinary.com/shapeable/image/upload/v1687303466/wikiore/171.jpg)](https://res.cloudinary.com/shapeable/image/upload/v1687303466/wikiore/171.jpg) \n\n_The impact of sparse data on LUC models._\n\n  \nWhile the differences illustrated in the figure above are subtle, each of these three models are based on a much higher data density than the resource estimate. As data density decreases the <b title=\"Local Uniform Conditioning\">LUC</b> estimates become smoother and more estimation-driven [geomet](https://en.wikipedia.org/wiki/Geometallurgy)ric artefacts are apparent. This affects any analysis of spatial variability based on naïve <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>\\-by-<b title=\"A discrete section within a resource model (i.e. block).\">panel</b> statistical comparison. While there are alternatives to <b title=\"Uniform Conditioning\">UC</b> and <b title=\"Local Uniform Conditioning\">LUC</b> these alternatives have similar issues. It is simply unrealistic to model short-scale variability from widely spaced data without making assumptions about the representivity (and stationarity) of the data.\n\nThe main alternatives to <b title=\"Uniform Conditioning\">UC</b> and <b title=\"Local Uniform Conditioning\">LUC</b> modelling include _**[conditional simulation](https://www.whittleconsulting.com.au/wp-content/uploads/2018/05/Conditional-Simulations-Tools-for-Modelling-Uncertainty-in-Open-Pit-Optimisation.pdf).**_\n\nUnlike UC, [conditional simulation](https://www.whittleconsulting.com.au/wp-content/uploads/2018/05/Conditional-Simulations-Tools-for-Modelling-Uncertainty-in-Open-Pit-Optimisation.pdf) (or CS) is _not_ based on an estimation algorithm. [geostatistical simulation](https://en.wikipedia.org/wiki/Geostatistics#Simulation) approaches do not try to minimise estimation error. Instead, simulation is designed to reproduce the sample histogram and the [variogram](https://en.wikipedia.org/wiki/Variogram) model. In essence, this is seen as preserving the variability of the data. The challenge with simulation is that there are multiple solutions. The sample histogram and [variogram](https://en.wikipedia.org/wiki/Variogram) model can be reproduced with grades in a vast array of [geomet](https://en.wikipedia.org/wiki/Geometallurgy)ric (spatial) patterns. The ‘conditioning’ part of CS refers to one method limiting the possible solution space by forcing the simulation to honour the spatial locations of the input data as well as the histogram and [variogram](https://en.wikipedia.org/wiki/Variogram). This still leaves many possible simulation outcomes but improves the overall character of the simulation models. In most uses CS refers to a suite of models (called realisations) that are considered as a combined set. No individual model is ‘more correct’ than another (they are ‘equi-probable’). The conventional use of a set of simulations is to examine the between-model histograms and interpret this distribution as a probability.\n\n  \nFor [heterogeneity](/pages/wikiore-heterogeneity) modelling, CS may offer an alternative approach. Since one of the primary objectives is to preserve the underlying variability of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29), at first glance CS is ideally suited to [heterogeneity](/pages/wikiore-heterogeneity) studies. The challenge is developing a valid and robust method to interpret the CS realisations and their meaning in terms of [heterogeneity](/pages/wikiore-heterogeneity). This is an area of active research. One of the primary problems with using CS to predict [heterogeneity](/pages/wikiore-heterogeneity) is separating the apparent variability due to data (uncertainty-driven) from the true variability of the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) (geology-driven). The conditioning aspect of CS reduces the degrees-of-freedom for the spatial arrangement of grade. This means that the modelled variability is more likely to represent the ‘truth’ where there is a lot of data and, where the data is widely spaced the modelled variability will combine aspects of both uncertainty and geology.\n\nIt is worthwhile exploring CS as an alternative to UC. There are several possible use cases including:\n\n*   Applying some form of image analysis to characterise regions that display similar patterns across multiple realisations (e.g., to assist with <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>ing). This may help identify areas where the grade always displays high or low variability (or give a probability of being in a high/low variability <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>);\n*   Combining the CS realisations and models sets with some form of data density index to help understand the uncertainty/geology components of the modelled variability. This in turn could be used to target future drill programs dedicated to reducing uncertainty in areas where there is apparent higher [grade variability](/pages/wikiore-heterogeneity);\n*   A naïve approach where the simple inter-realisation distribution at <b title=\"A spatially constrained unit within the resource model.\">block</b> or point scale is interpreted as the probability of grade being above/below some selection threshold (as in a sensed grade). This basic approach could be coupled with downstream modelling of the above/below grade <b title=\"A spatially constrained unit within the resource model.\">block</b> proportions to estimate the grade of each. This would be similar to a form of categorical indicator estimation.\n*   Alternate [non-linear estimation](https://www.cubeconsulting.com/wp-content/uploads/2020/11/Technical-Note-What-is-Non-Linear-Estimation_FINAL.pdf) (e.g. MIK). <b title=\"Uniform Conditioning\">UC</b> is not the only form of [non-linear estimation](https://www.cubeconsulting.com/wp-content/uploads/2020/11/Technical-Note-What-is-Non-Linear-Estimation_FINAL.pdf). Among the alternatives multiple indicator [kriging](https://en.wikipedia.org/wiki/Kriging) (MIK) is the most commonly known approach. Like UC, [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) estimates are used to understand the block-grade histogram and the estimate provides information on the proportion of each block/<b title=\"A discrete section within a resource model (i.e. block).\">panel</b> above a series of Cut-Offs (known as indicator thresholds). Also like UC,[MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) estimates can be ‘localised’ to form a localised multipleindicator [kriging](https://en.wikipedia.org/wiki/Kriging) (LMIK) estimate. Unlike <b title=\"Uniform Conditioning\">UC</b> however, there is no direct way of modelling the <b title=\"Smallest Mining Unit (a.k.a Minimum Mining Unit MMU)\">SMU</b> size. [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) is effectively a point estimate (provides the distribution of point data inside a <b title=\"A discrete section within a resource model (i.e. block).\">panel</b>). While there are some change-of-support mechanisms that can be applied to [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) to allow for <b title=\"Smallest Mining Unit (a.k.a Minimum Mining Unit MMU)\">SMU</b> size, these approaches are dependent on modelling to pre-determined distributions (e.g. Lognormal change of support of Affine correction).\n\n  \nThe underlying difference between <b title=\"Uniform Conditioning\">UC</b> and [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) is best explained by considering the nature of the mineralised system to be modelled. <b title=\"Uniform Conditioning\">UC</b> is based on an assumption of bi-Gaussianity. Effectively this means that if you were to traverse from one location with a known grade to a second location with a known grade you would intersect all the grade values between those know grades during the traverse. Or, more simply put, the grade distribution can be contoured. This is known as a diffusive grade distribution. In contrast, [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) is a point estimator and it does not assume this bi-Gaussian distribution. This means that [MIK](https://en.wikipedia.org/wiki/Kriging#Multiple-indicator) is more robust where the [mineralisation](https://en.wikipedia.org/wiki/Mineralization_%28geology%29) is characterised by grade discontinuities (i.e., you can’t contour the grades), a pattern commonly referred to as a mosaic distribution.\n\n*   *   Disjunctive [kriging](https://en.wikipedia.org/wiki/Kriging);\n    *   Multi-Gaussian [kriging](https://en.wikipedia.org/wiki/Kriging);\n    *   Median indicator [kriging](https://en.wikipedia.org/wiki/Kriging);\n    *   A variety of co-[kriging](https://en.wikipedia.org/wiki/Kriging) and indicator co-[kriging](https://en.wikipedia.org/wiki/Kriging) approaches; and\n    *   Probability [kriging](https://en.wikipedia.org/wiki/Kriging).\n*   Approaches that are not primarily geostatistically driven. [geostatistics](https://en.wikipedia.org/wiki/Geostatistics) is not the only field where modelling and understanding spatial variability is required. In particular the fields of ecology, [computer vision](https://en.wikipedia.org/wiki/Computer_vision) and spatial mapping (i.e., GIS) have alternative approaches where spatial variability is characterised as part of an evaluation process.  Consideration of these fields may identify a novel approach to modelling spatial variability of mineral resources. In particular:\n    *   Moran’s I and Geary’s C; Variability metrics derived from GIS and spatial science;\n    *   Ecology-based metrics that describe ‘clumpiness’ and ‘connectedness’;\n    *   Image analysis based techniques; and\n    *   [computer vision](https://en.wikipedia.org/wiki/Computer_vision) techniques.\n\nRecommendations[](#Recommendations)\n===================================\n\nModelling[](#Modelling)\n-----------------------\n\nConstructing a ‘ground-truth’ model from <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data is a relatively straightforward project. An industry standard approaches for this estimate was adopted. There are however, opportunities to improve the modelling. These include:\n\n*   Data-related improvements. The current model is based on <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data with a low vertical resolution. This means that there is no within-<b title=\"Individual levels within an open pit.\">bench</b> variability included in the model in the vertical dimension. The model predicts the same grade throughout a 15m high column of rock measuring 3.125m x 3.125m. It is highly likely that vertical variability exists, and it may be greater than the XY (plan) variability, therefore the ‘ground-truth’ model may under predict the <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> [heterogeneity](/pages/wikiore-heterogeneity);\n*   Model-related improvement. The main model improvement recommended is to develop robust and reliable geology-based <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s. This estimate is based on subdividing the grade distribution of all <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s into four classes and estimating within each class. While this is a practical solution, it is no substitute for well developed and internally consistent geological <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s.\n\n Analysis[](#Analysis)\n----------------------\n\nAs well as improving the data and modelling practices, a number of actions to improve understanding and modelling of variability, [heterogeneity](/pages/wikiore-heterogeneity) and the potential value of exploiting the natural interaction of in situ rock properties with a variety of [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) technologies can be taken. In an ideal world, the [heterogeneity](/pages/wikiore-heterogeneity) analysis and modelling would be developed from first principles, similar to the methods used for other [metallurgical](https://en.wikipedia.org/wiki/Metallurgy)/[geomet](https://en.wikipedia.org/wiki/Geometallurgy)allurgical performance models. This would involve:\n\n*   A phase of descriptive geology mapping focusing on rock properties that are considered important to the performance of the intended [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) system. For example, for [bulk sorting](/pages/wikiore-bulk-sorting) of 100 tonne parcels post excavation, the size, shape, orientation of packages of rock with similar geology may be an important performance driver;\n*   This geological mapping phase should be supported by dedicated laboratory testing. End-member specimens of a variety of geology types should be tested at <b title=\"Individual levels within an open pit.\">bench</b> or [pilot plant](https://en.wikipedia.org/wiki/Pilot_plant) scale and their measured performance compared to the prediction based on geological understanding. This test program should include comprehensive pre and post-characterisation for the input and output streams with an associated analysis of potential performance drivers. The test program should identify rock properties that are positively (or negatively) correlated to the [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) process;\n*   Using the results of the test program, a second-phase of geological mapping should be completed. This should focus on identifying the location and distribution of zones with character similar to those identified as important in the test program. This may include similar [lithology](https://en.wikipedia.org/wiki/Lithology) or alteration, similar texture, similar fabric or structural preparation amongst other likely drivers;\n*   The [geomet](https://en.wikipedia.org/wiki/Geometallurgy)allurgical mapping (above) should then be combined with grade distribution analyses, identifying suitable proxies (if any) and assisting the development of dedicated [heterogeneity](/pages/wikiore-heterogeneity)  <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s – for that specific [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) approach (note, different Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> levers may require different [heterogeneity](/pages/wikiore-heterogeneity)  <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b>s if their performance is driven by contrasting physical properties);\n\n  \nOnly after the above steps are complete should numerical-spatial modelling of heterogenetiy and the subsequent [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) performance be considered.\n\nIt is recognised that this is a substantial piece of work involving applied research as well as the application of proven techniques. An iterative approach is recommended, as is taking advantage of as much of the existing data and information as possible. One example of adopting existing data would be to analyse any time-based grade data already available. In most operations, there is already some form of shift-by-shift sampling, albeit usually after [crushing](https://en.wikipedia.org/wiki/Crusher) and [grinding](https://en.wikipedia.org/wiki/Mill_%28grinding%29). This shift data is the ideal place to start trying to validate existing or proposed modelling approaches. Similarly, longer-term reconciliation of the mine’s predicted tonnes and grade deliveries compared to the ore treatment plant’s receipts can provide an indication of model efficacy.\n\nAlternatively, the existing plant data can be used for a preliminary analysis of the impact of a [beneficiation](https://en.wikipedia.org/wiki/Beneficiation) technology. This is particularly the case if there is a fine time-scale resolution (hours or less) on the available data. While production presented to the ore treatment plant is usually blended (the objective being directly opposite to that of Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b>), if the <b title=\"Material entering a predetermined system.\">feed</b> [grade variability](/pages/wikiore-heterogeneity) crosses the Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> selection threshold, a minimum upgrade potential can be determined and evaluated.  \n \n\nHigh Resolution Blast Hole Modelling[](#High_Resolution_Blast_Hole_Modelling)\n-----------------------------------------------------------------------------\n\nGeology and grade models are based on data collected prior to the mining process. With current technology, it is not feasible to collect, analyse and model exhaustive data of mineralised systems. Consequently, the developed models are all based on sparse data which is used to predict the geology and grade at unsampled volumes and locations. This approach results in trade-offs between:\n\n*   The amount of data, particularly the distance between sampled points;\n*   The quality of the samples; and\n*   The nature of the predictive model, specifically the trade-off between local and global [conditional bias](http://www.ccgalberta.com/ccgresources/report04/2002-115-conditionalbias.pdf) and the smoothing of the modelled grades.\n\nThe data is typically collected in multiple stages with each stage increasing the amount of sampled points and reducing the distance between these points. The intent of multi-stage data collection is to manage cost and focus sampling on areas most important in terms of the <b title=\"Extraction sequence determined to produce optimal outcomes from long-term mine plan.\">production schedule</b> . Thus, early stages of mining project development are dominated by relatively long and wide spaced drill holes where the sample quality is maximised. As the production timeline approaches, shorter holes are drilled from previously mined locations, reducing drill cost and allowing closer-spaced data to be collected. In some cases, there is a trade-off between this higher density drilling and sample quality.\n\nAt most [open pit](https://en.wikipedia.org/wiki/Open-pit_mining) mines the final stage of sampling is based on <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> drilling. Samples are collected from the <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> cone with one sample representing the full 15m <b title=\"Individual levels within an open pit.\">bench</b> . <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s are usually drilled on 5-7m centres and samples are usually collected from all <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s. Therefore, a dense data coverage is achieved.\n\nCreating a model from the complete <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> data set is most likely to closely represent the underlying true grade distribution on a <b title=\"Individual levels within an open pit.\">bench</b> -by-<b title=\"Individual levels within an open pit.\">bench</b> basis. The model is considered to be the high resolution <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b> and provides an index for analysing variability, [heterogeneity](/pages/wikiore-heterogeneity) and for referencing against other models created from wider-spaced sample data.\n\nThere are some limitations with this approach. Models and estimates are only as reliable as the input data and the assumptions and parameters used during estimation. While a dense data coverage mitigates against model-induced estimation artefacts, other factors affect the differences between models created from different types of data. These factors include:\n\n*   The differences in sample support between 15m <b title=\"Cylindrical hole used to load explosives into unbroken material.\">blast hole</b>s and diamond drill hole samples, even after compositing;\n*   <b title=\"Geological, statistical or spatial groupings used for modelling.\">domain</b> boundary location uncertainty which itself is a function of both data density and quality;\n*   The statistical data representivity of the data and correct representation of the histogram and [variogram](https://en.wikipedia.org/wiki/Variogram) by the data; and\n*   The sample population statistics and sample bias (or [conditional bias](http://www.ccgalberta.com/ccgresources/report04/2002-115-conditionalbias.pdf)).\n\nNone of these factors have been investigated as part of this program of work. It is recommended that to  treat the results of this modelling and [heterogeneity](/pages/wikiore-heterogeneity)/variability prediction with caution. A more comprehensive study is required to understand the drivers of both the local grade distribution and the impact of both data and parameter uncertainty on the predictive performance.\n\n [](#a8df6fbcc43d31d99e5112eb009ed8a2d)\n---------------------------------------\n\n[heterogeneity](/pages/wikiore-heterogeneity) Modelling and Classification[](#heterogeneity_Modelling_and_Classification)\n-------------------------------------------------------------------------------------------------------------------------\n\nThe meaning of ‘[heterogeneity](/pages/wikiore-heterogeneity)’ is not clearly defined and is likely to vary from person to person. Furthermore, the generic use of ‘[heterogeneity](/pages/wikiore-heterogeneity)’ may cause cognitive bias when attempting to model and evaluate Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> lever opportunities. It is extremely likely that different rock properties will give rise to different Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> performance for each different lever or technology applied. There is unlikely to be a single, cross-functional and appropriate definition of [heterogeneity](/pages/wikiore-heterogeneity). It is recommend to define the problem of [heterogeneity](/pages/wikiore-heterogeneity) by understanding the properties of the solution and the nature of the technologies being applied. This ‘correct framing’ of the problem is essential for creating fit-for-purpose Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> response models. Attempting to define classes or groups of similar Bulk separation below current <b title=\"Term for operational limitations imposed within grade control.\">minimum mining unit</b> response is recommended as a first step.  \n \n\nAlternate Opportunities[](#Alternate_Opportunities)\n---------------------------------------------------\n\nIn many respects evaluating ‘[heterogeneity](/pages/wikiore-heterogeneity)’ or variability is similar to other image analysis or pattern recognition problems. Solutions for classifying and analysing images exist outside to the mining, geology and [geostatistics](https://en.wikipedia.org/wiki/Geostatistics) fields. It is recommended to investigating options and techniques from other industries including:\n\n*   [computer vision](https://en.wikipedia.org/wiki/Computer_vision);\n*   Image classification (particularly [machine learning](https://en.wikipedia.org/wiki/Machine_learning) approaches, supervised and unsupervised);\n*   Image enhancement techniques (particularly those aimed at enhancing contrast); and\n*   Geophysical inversion techniques.\n\nIn each of the above cases, it is also important to recognise and evaluate the impact of <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> size, <b title=\"A discrete section within a resource model (i.e. block).\">panel</b> origin and legend/scale classification. The combination of these three parameters can materially impact any upscaling analysis and application of different colour-scales or legends could give rise to an almost infinite range of images for analysis.  \n \n\n### References[](#References)\n\nAbzalov, M. 2006.  Localised Uniform Conditioning (LUC): A New Approach for Direct Modelling of Small Blocks.  Mathematical Geology, vol 38, pp393-411.\n\nRutter J. 2017. [Grade Engineering](/pages/wikiore-introduction-to-grade-engineering) and<b title=\"Grade Engineering\">GE</b> View. CRC ORE White Paper Cooperative Research Centre for Optimising Resource Extraction (CRC ORE). Brisbane Australia\n\nRutter J, and Dunham S. 2020. [heterogeneity](/pages/wikiore-heterogeneity) and [bulk ore sorting](/pages/wikiore-bulk-sorting) – Methods to estimate ore sorting performance in <b title=\"Methodologies used to separate gangue from valuable material.\">preconcentration</b> Digital Conference 2020 (The Australasian Institute of Mining and[metallurgy](https://en.wikipedia.org/wiki/Metallurgy): Melbourne).\n\nScott M, and Rutter J. 2020. Operational Deployment of Sensor Technologies for [bulk ore sorting](/pages/wikiore-bulk-sorting) at [Mogalakwena](https://www.angloamerican.com/about-us/our-stories/mine-profile-mogalakwena)[PGE](https://en.wikipedia.org/wiki/Platinum_group) Mine, in <b title=\"Methodologies used to separate gangue from valuable material.\">preconcentration</b> Digital Conference 2020 (The Australasian Institute of Mining and[metallurgy](https://en.wikipedia.org/wiki/Metallurgy): Melbourne).\n\nVann & Guibal, 1998.  Beyond ordinary[kriging](https://en.wikipedia.org/wiki/Kriging) - An Overview of [non-linear estimation](https://www.cubeconsulting.com/wp-content/uploads/2020/11/Technical-Note-What-is-Non-Linear-Estimation_FINAL.pdf) – Keynote Address.  Symposium on Beyond ordinary[kriging](https://en.wikipedia.org/wiki/Kriging).\n"},"organisations":[],"embeds":{"citations":[],"imageAssets":[]},"video":null,"banner":null,"fileAssets":[],"citations":[],"feedEntries":[],"posts":[],"moderators":[]}}},"pageContext":{"lang":{"id":"en","path":"","iso":"en","name":"English","label":"English"},"availableEntities":["Catalyst","CommodityGroup","Innovation","Organisation","Page","Person","Post","SubTopic","Topic","Trend","OrganisationType","HorizonType","Economy","FeedEntry","Video","Citation","Objective","ProfileType","ExpertiseLevel"],"detailEntities":["Catalyst","CommodityGroup","Innovation","Organisation","Page","Person","Post","SubTopic","Topic","Trend"],"site":{"id":"65d2edd1ebf7bb9388ffb0df","slug":"connect-ore","name":"ConnectOre","url":"https://connectore.org","title":"ConnectOre","twitter":null,"threads":null,"facebook":null,"linkedin":null,"instagram":null,"flickr":null,"tiktok":null,"youtube":null,"ownerName":"ICAA","recaptchaKey":"6LfG2J0qAAAAAHDCFfJnrWOj6DiWJGik5BAa_7p8","googleSiteVerification":null,"platformName":"ConnectOre","platformUrl":"https://platform.connectore.org","supportEmail":null,"contactEmail":null,"mainMenu":{"id":"65d2edaaebf7bb9388ffaf79","slug":"website-navigation"},"linearMenu":null,"entityViews":[{"id":"67528c24c135ebcfc668f1ba_65d2edd1ebf7bb9388ffb0df","name":"Value Chain","hash":"value-chain","label":null,"slug":"value-chain","count":0,"showCount":null,"disabled":null,"icon":{"id":"6752a61b8b51dfa7b77b1e37","name":"Connections Navigator Icon Glyph","slug":"connections-navigator-icon-glyph","component":"ConnectionsNavigatorIconGlyph"},"childEntityTypes":[{"id":"6667fa4ff794dd6e045135c1","name":"Value Chain Theme","internalName":"Trend"},{"id":"6667fa58f794dd6e045135c6","name":"Topic","internalName":"Topic"}],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]},{"id":"6757ef1ede8adbd45cf19474_65d2edd1ebf7bb9388ffb0df","name":"Bulletin Board","hash":"bulletin-board","label":null,"slug":"bulletin-board","count":0,"showCount":null,"disabled":null,"icon":{"id":"681c2cebbccc5a72d8382f31","name":"Tag Icon","slug":"tag-icon","component":"TagDotIconGlyph"},"childEntityTypes":[],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]},{"id":"67528d88c135ebcfc668f31b_65d2edd1ebf7bb9388ffb0df","name":"Innovations","hash":"innovations","label":null,"slug":"innovations","count":0,"showCount":null,"disabled":null,"icon":{"id":"67528cddc135ebcfc668f304","name":"Grid Icon Glyph","slug":"grid-icon-glyph","component":"GridIconGlyph"},"childEntityTypes":[{"id":"67528d74c135ebcfc668f315","name":"Innovation","internalName":"innovation"}],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]},{"id":"67625640fbc7c50d8644c592_65d2edd1ebf7bb9388ffb0df","name":"Community","hash":"community","label":null,"slug":"community","count":0,"showCount":null,"disabled":null,"icon":{"id":"67528cddc135ebcfc668f304","name":"Grid Icon Glyph","slug":"grid-icon-glyph","component":"GridIconGlyph"},"childEntityTypes":[{"id":"67528404c135ebcfc668ee1c","name":"Person","internalName":"Person"}],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]},{"id":"67528e1bc135ebcfc668f33a_65d2edd1ebf7bb9388ffb0df","name":"Stakeholder Map","hash":"stakeholder-map","label":null,"slug":"stakeholder-map","count":0,"showCount":null,"disabled":null,"icon":{"id":"67528cddc135ebcfc668f304","name":"Grid Icon Glyph","slug":"grid-icon-glyph","component":"GridIconGlyph"},"childEntityTypes":[{"id":"66680141f794dd6e0451367a","name":"Organisation","internalName":"Organisation"}],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]},{"id":"67529040c135ebcfc668f4ae_65d2edd1ebf7bb9388ffb0df","name":"References","hash":"references","label":null,"slug":"references","count":0,"showCount":null,"disabled":null,"icon":{"id":"67528cddc135ebcfc668f304","name":"Grid Icon Glyph","slug":"grid-icon-glyph","component":"GridIconGlyph"},"childEntityTypes":[{"id":"6752902fc135ebcfc668f4a4","name":"Article","internalName":"FeedEntry"},{"id":"67529039c135ebcfc668f4aa","name":"Citation","internalName":"Citation"},{"id":"67528de9c135ebcfc668f328","name":"Catalyst","internalName":"catalyst"},{"id":"67528df9c135ebcfc668f32d","name":"Page","internalName":"page"}],"type":{"id":"67528b63c135ebcfc668f19c","name":"Explorer","slug":"explorer"},"slices":[]}],"entityAppViews":[],"entityOnboardingViews":[{"id":"680c3a4ebc80e4d0f3cfa2ef_65d2edd1ebf7bb9388ffb0df","name":"Trend Expertise Step","hash":"trend-expertise-step","label":null,"slug":"trend-expertise-step","count":0,"showCount":null,"icon":null,"childEntityTypes":[],"type":{"id":"680c394fbc80e4d0f3cfa2d8","name":"Onboarding","slug":"onboarding"},"slices":[{"id":"680c39f2bc80e4d0f3cfa2e0_65d2edd1ebf7bb9388ffb0df","name":"Trend Expertise Onboarding Grid","slug":"trend-expertise-onboarding-grid","label":null,"updated":"2025-04-26T01:42:10.20","layout":{"id":"680c3993bc80e4d0f3cfa2dd","name":"Slice Onboarding Trend Expertise Grid","slug":"slice-onboarding-trend-expertise-grid","component":"SliceLayoutOnboardingTrendExpertiseGrid"},"connectedEntities":[],"headerFontType":null,"pretitle":null,"pretitleRich":{"id":"680c39f2bc80e4d0f3cfa2e0_65d2edd1ebf7bb9388ffb0df_pretitleRich","text":""},"title":null,"titleRich":{"id":"680c39f2bc80e4d0f3cfa2e0_65d2edd1ebf7bb9388ffb0df_titleRich","text":"Are you **an industry expert** in any of these Value Chain Themes?"}}]},{"id":"680c3a85bc80e4d0f3cfa2fa_65d2edd1ebf7bb9388ffb0df","name":"Profile Types Step","hash":"profile-types-step","label":null,"slug":"profile-types-step","count":0,"showCount":null,"icon":null,"childEntityTypes":[],"type":{"id":"680c394fbc80e4d0f3cfa2d8","name":"Onboarding","slug":"onboarding"},"slices":[{"id":"680c3a82bc80e4d0f3cfa2f7_65d2edd1ebf7bb9388ffb0df","name":"Profile Types Onboarding Grid","slug":"profile-types-onboarding-grid","label":null,"updated":"2025-04-26T01:44:34.63","layout":{"id":"680c3a13bc80e4d0f3cfa2e4","name":"Slice Onboarding Profile Types Grid","slug":"slice-onboarding-profile-types-grid","component":"SliceLayoutOnboardingProfileTypesGrid"},"connectedEntities":[],"headerFontType":null,"pretitle":null,"pretitleRich":{"id":"680c3a82bc80e4d0f3cfa2f7_65d2edd1ebf7bb9388ffb0df_pretitleRich","text":""},"title":"Which best describes you?","titleRich":{"id":"680c3a82bc80e4d0f3cfa2f7_65d2edd1ebf7bb9388ffb0df_titleRich","text":""}}]},{"id":"680c3aaabc80e4d0f3cfa2ff_65d2edd1ebf7bb9388ffb0df","name":"Objectives Step","hash":"objectives-step","label":null,"slug":"objectives-step","count":0,"showCount":null,"icon":null,"childEntityTypes":[],"type":{"id":"680c394fbc80e4d0f3cfa2d8","name":"Onboarding","slug":"onboarding"},"slices":[{"id":"680c3aa8bc80e4d0f3cfa2fc_65d2edd1ebf7bb9388ffb0df","name":"Objectives Onboarding Grid","slug":"objectives-onboarding-grid","label":null,"updated":"2025-04-26T01:45:12.76","layout":{"id":"680c3a3fbc80e4d0f3cfa2ea","name":"Slice Onboarding Objectives Grid","slug":"slice-onboarding-objectives-grid","component":"SliceLayoutOnboardingObjectivesGrid"},"connectedEntities":[],"headerFontType":null,"pretitle":null,"pretitleRich":{"id":"680c3aa8bc80e4d0f3cfa2fc_65d2edd1ebf7bb9388ffb0df_pretitleRich","text":""},"title":"What are your priority objectives?","titleRich":{"id":"680c3aa8bc80e4d0f3cfa2fc_65d2edd1ebf7bb9388ffb0df_titleRich","text":""}}]}],"gptLanguages":[{"id":"65d2eda5ebf7bb9388ffaf1f","name":"English","slug":"english","iso":"en","locale":"en_US","path":"/en","label":"English","menuLabel":null},{"id":"666f70b6e051ad9ef086d1be","name":"Arabic","slug":"arabic","iso":"ar","locale":"ar_AE","path":"/ar","label":"عربي","menuLabel":null},{"id":"666f7161e051ad9ef086d1c4","name":"Spanish","slug":"spanish","iso":"es","locale":"es_ES","path":"/es","label":"español","menuLabel":null},{"id":"666f71c9e051ad9ef086d1c8","name":"German","slug":"german","iso":"de","locale":"de_DE","path":"/de","label":"Deutsch","menuLabel":null},{"id":"65d2eda5ebf7bb9388ffaf20","name":"French","slug":"french","iso":"fr","locale":"fr_FR","path":"/fr","label":"Français","menuLabel":null}],"gptQuestionTemplate":{"id":"662c7f6e81ef1920e3638518","name":"Standard Open Question","slug":"standard-open-question","path":"/prompt-templates/standard-open-question","__typename":"PromptTemplate","_entityTypeName":null,"label":"CopperAI","type":{"id":"662c7f1981ef1920e363850f","name":"Ask","slug":"ask"},"description":{"text":"This template is tailored for providing accurate responses based on specific context. It integrates contextual details from \"contextual information\" and addresses the \"target question\". The response is always in English, prioritizing clarity. If the answer is unknown, the template openly acknowledges this, ensuring honesty and avoiding speculation. This format is ideal for precise, context-sensitive, and truthful answering."},"summary":{"text":"Ask a general question about Website content"},"gptModel":{"id":"67564c4f2f6749787d0e03dc","name":"OpenAI GPT 4o","slug":"open-ai-gpt-4-o","dataSetId":"openai:gpt-4o"},"variablesPrompt":{"text":""},"submitLabel":"Ask","languages":[],"icon":{"id":"67528b9cc135ebcfc668f1a3","name":"AI Lower Case Icon Glyph","slug":"ai-lower-case-icon-glyph","component":"AiLowerCaseIconGlyph"},"variables":[{"id":"6668076bf794dd6e045138d9","name":"question","slug":"question","useVectorStore":true,"label":"What would you like to know?","labelAnother":null,"help":{"text":""},"options":{"text":""},"defaultValue":null,"entityTypes":[],"controlType":{"id":"6667ff19f794dd6e04513656","name":"Textarea","slug":"textarea"},"labelContextual":null,"labelAnotherContextual":null,"helpContextual":{"text":""},"optionsContextual":{"text":""},"defaultValueContextual":null,"entityTypesContextual":[],"controlTypeContextual":{"id":"6667ff19f794dd6e04513656","name":"Textarea","slug":"textarea"}}]},"advertisements":[],"logoVerticalOffset":null,"logoHorizontalOffset":null,"logoVerticalOffsetMobile":null,"logoHorizontalOffsetMobile":null,"logoVerticalOffsetTablet":null,"logoHorizontalOffsetTablet":null,"logoVerticalOffsetDesktop":null,"logoHorizontalOffsetDesktop":null,"logoHeightMobile":44,"logoHeightTablet":null,"logoHeightDesktop":null,"headerHeightMobile":null,"headerHeightTablet":null,"headerHeightDesktop":null,"logo":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1738807465/copper-connect/site/copper-connect_logo__CopperOre_Master_Brand_Identity-colour_cd4vgg.webp","type":"image/webp","width":945,"height":266},"logoInverted":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1738807452/copper-connect/site/copper-connect_logoInverted__CopperOre_Master_Brand_Identity-white_chnucx.webp","type":"image/webp","width":945,"height":266},"footerMenu":null,"footerSecondaryMenu":null,"footerContent":{"text":"© {year} ConnectOre. All Rights Reserved\n"},"creator":null,"poweredBy":{"id":"65d2edadebf7bb9388ffafc8","name":"Shapeable","slug":"shapeable","url":"https://shapeable.ai","logo":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1674010034/platform/organisation/shapeable_logo__shapeable.png","type":"image/png","width":1174,"height":368},"logoInverted":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1729474494/copper-connect/organisation/shapeable_logoInverted__shapeable-logo-inverted_sdcfpk.png","type":"image/png","width":392,"height":118},"logoSubtle":null},"poweredByLabel":null,"poweredByContent":{"text":""},"explorerPage":{"name":"Explorer","title":null,"slug":"explorer","path":"/explorer"},"termsPage":{"name":"Terms of Use","title":null,"slug":"terms-of-use","path":"/terms-of-use"},"homePage":null,"knowledgeHubPage":{"name":"News","title":null,"slug":"news","path":"/news"},"privacyPolicyPage":{"name":"Privacy Policy","title":null,"slug":"privacy-policy","path":"/privacy-policy"},"summary":{"text":"A platform to accelerate industry-wide innovation through collective intelligence and collaborative action"},"thumbnail":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1713942350/copper-connect/site/copper-connect_thumbnail__site-thumb_bmicl8.png"},"openGraph":{"title":"ConnectOre","date":"2026-07-31T04:44:47.80","description":{"plain":"A platform to accelerate industry-wide innovation through collective intelligence and collaborative action\n"},"image":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1713942350/copper-connect/site/copper-connect_thumbnail__site-thumb_bmicl8.png","type":"image/png","thumbnails":{"full":{"url":"https://res.cloudinary.com/shapeable/image/upload/v1713942350/copper-connect/site/copper-connect_thumbnail__site-thumb_bmicl8.png"}}}},"termsAndConditions":{"text":""},"privacyPolicy":{"text":"At ConnectOre we respect your privacy. We want to ensure that you get the information, content, and experiences that matter most to you. ConnectOre is committed to protecting the privacy of its stakeholders, communities, and other contacts.\n\n## Scope\n\nThis privacy policy applies to all personal data processed by full-time and part-time employees, volunteers when acting on behalf of ConnectOre contractors and partners doing business on behalf of ConnectOre, as well as all legal entities, all operating locations in all countries, and all business processes conducted by ConnectOre.\n\n## Information Collected\n\n#### What information do we collect?\n\nConnectOre collects the following personal data in line with the use purposes explained in a subsequent section:\n\n  * Your name and contact details\n  * Online profile data/usage\n  * Contact information\n  * Social media profile information\n  * Education and professional information\n  * Registration and participation in ConnectOre events and activities \n  * Information about service usage\n  * Cookies\n  * Authentication data\n  * Location information\n  * Author and peer review information\n  * Other information you upload or provide to us\n\n#### How do we use your information?\n\nConnectOre uses (and, where specified, shares) your personal information for the following purposes:\n\n  * To provide support or other services. ConnectOre may use your personal information to provide you with support or other services that you have ordered or requested. ConnectOre may also use your personal information to respond directly to your requests for information, including registrations for webinars, or other specific requests, or pass your contact information to the appropriate ConnectOre supplier or reseller for further follow-up related to your interests.\n  * To provide information based on your needs and respond to your requests. ConnectOre may use your personal information to provide you with notices of new product releases and service developments.\n  * To administer products. ConnectOre may contact you if you make use of (digital) products we offer, to confirm certain information (for example, that you did not experience problems in a download process). We may also use this information to confirm compliance with licensing and other terms of use and may share it with your company/institution.\n  * To assist in your participation in ConnectOre activities. ConnectOre will communicate with you, if you are participating in certain ConnectOre activities such as ConnectOre Summit, authoring or reviewing a ConnectOre article, or ConnectOre humanitarian activities. ConnectOre may send you information such as update messages related to those activities (such as but not limited to the event's content, and event logistics)\n  * To update you on relevant ConnectOre events and opportunities. ConnectOre may communicate with you regarding relevant ConnectOre events and opportunities.\n  * To protect ConnectOre content and services. We may use your information to prevent potentially illegal activities and to enforce our terms and conditions.\n  * To get feedback or input from you. In order to deliver products and services of most interest to our stakeholders, from time to time, we may ask you to provide us input and feedback (for example through surveys).\n\n#### How can you control your information?\n\nYou can control the information we have about you and how we use as follows:\n\n  * If you are a registered guest for ConnectOre Annual Summit 2021, any request for review, revise or correction of your personal data can be sent to john.fennell@copper.com.au specifying your request.\n\n#### Personal data about minors and children\n\nConnectOre does not knowingly collect data from or about children under 16 without the permission of parent(s)/guardian(s). If we learn that we have collected personal information from a child under 16, we will delete that information as quickly as possible. If you believe that we might have any information from or about a child under age 16, please contact us.\n\n#### How will you know if the Privacy Policy is changed?\n\nConnectOre may update its Privacy Policy from time to time. If we make any material changes you will be notified by means of a notice on our website prior on the date the change becomes effective. We encourage you to periodically review this page for the latest information on our privacy practices.\n\n## Technical And Regulatory Information\n\n#### Logging practices\n\nConnectOre automatically records the Internet Protocol (IP) addresses of visitors. The IP address is a unique number assigned to every computer on the internet. Generally, an IP address changes each time you connect to the internet (it is a \"dynamic\" address). Note, however, that if you have a broadband connection, depending on your individual circumstance, the IP address that we collect may contain information that could be deemed identifiable. This is because, with some broadband connections, your IP address doesn't change (it is \"static\") and could be associated with your personal computer.\n\nAs well as recording the IP addresses of users, ConnectOre may also keep track of sites that users visited immediately prior to visiting ConnectOre's website and the search terms they used to find it. We keep track of the pages visited on ConnectOre's website, the amount of time spent on those pages and the types of searches done on them. Your searches remain confidential and anonymous. ConnectOre uses this information only for statistical purposes to find out which pages users find most useful and to improve the website.\nConnectOre also captures and stores information that you transmit. This may include:\n\n  * Browser/Device type/version\n  * Operating system used\n  * Media Access Control (MAC) address\n  * Date and time of the server request\n  * Volume of data transferred\n\n#### External links behaviour\n\nSome of the links on ConnectOre's websites link to other sites created and maintained by other public- and/or private-sector organizations. ConnectOre provides these links solely for your information and convenience. When you transfer to an outside website, you are leaving ConnectOre domain, and ConnectOre's information management policies no longer apply. ConnectOre encourages you to read the privacy statement of each external website that you visit before you provide any personal data.\n\n#### Cookies and web beacons\n\nCookies and web beacons are electronic placeholders that are placed on your device by websites to track your individual movements on that website over time. ConnectOre uses both session-based cookies (which last only for the duration of the user's session) and persistent cookies (which remain on your device and provide information about the session you are in and waits for the next time you use that site again).\n\nThese cookies and web beacons provide useful information to ConnectOre, enabling us to recognize repeat users, facilitate the user's access to and use of our sites, allows us to track usage behavior, and to balance the usage of our websites on all ConnectOre web servers.\nTracking cookies, third-party cookies, and other technologies such as web beacons may be used to process additional information, enable non-core functionalities on ConnectOre website and enable third-party functions (such as a social media \"share\" link). We may also include web beacons and other similar technology in promotional email messages to determine whether the messages have been opened.\n\n#### Do Not Track (DNT)\n\nThe online advertising industry has self-regulatory initiatives designed to provide consumers a choice in the types of ads they may see online and to conveniently opt-out from online behavioral ads served by some or all of the companies participating in these programs. Our websites do not respond to DNT consumer browser settings.\n\n#### Responses to legal requests\n\nConnectOre reserves the right to share your information to respond to duly authorized information requests of governmental authorities or where required by law.\n\n#### Your data rights\n\nConnectOre complies with all applicable data privacy laws and regulations including, but not limited to, the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). Under these laws and regulations, you may have certain rights to your data. Should you wish to exercise any of these rights, please send an email request to john.fennell@copper.com.au with \"Data Privacy Request'' in the subject line and in the email please identify the specific privacy right you request assistance with. Please note additional information may be requested prior to fulfilling a request and that ConnectOre reserves the right to charge a fee, where permitted, to cover the cost of certain requests.\n\n#### How do I contact you if there is an issue?\nIf you have any questions or concerns about this Privacy Policy or about the use of your personal information, please feel free to contact us by email at john.fennell@copper.com.au"},"welcomeUrl":"https://connectore.org/app/welcome","welcomeTitle":"Welcome to ConnectOre and thanks for joining us! ","invitationAction":"join the ConnectOre community","setupCompletionMessage":null,"languages":[],"brandColors":[],"showLogin":true,"showShareMenu":null,"showFollowMenu":true,"showPlatformLogin":null,"showContactUs":null,"loginLabel":"Explorer Login","platformLoginLabel":"Community Platform","headerButtons":[]},"includeProfile":true,"disableProfileEditing":false,"dynamicEntityTypeNames":["Person","Organisation"],"profilePath":"/app/profile","welcomePath":"/app/welcome","id":"65d2edd7ebf7bb9388ffb16d","slug":"heterogeneity-modelling","entityPath":"/sub-topics/heterogeneity-modelling","openGraph":{},"entityTypeName":"SubTopic","name":"Heterogeneity Modelling","updated":"2024-12-20T01:01:47.75","__type":"SubTopic"}},
    "staticQueryHashes": ["1044227382","1158597448","1242646999","1427075558","1452322194","1520036161","1545705019","1586309863","1606754935","1816168740","1864381666","1903214493","1945056504","1989845544","2034981229","2181044613","2409034939","2463401854","2526615987","2642656775","2862279633","2867243664","2910164142","2912920178","296816903","3073584486","3150399266","3172506128","3320076387","3447680494","3453148481","3624873332","364221563","3692255024","3778988535","3782604890","4091857177","4216505212","591477963","701411134","758936535","80004486","857137463","881103158"]}