Index definition and intended use

GREW-OI measures the relative spatial co-location of hydroclimatic opportunity, acid-soil co-benefit opportunity and mapped feedstock-access opportunity under an explicit decision structure. Each component is dimensionless on a 0–1 scale; the displayed composite is scaled to 0–100.

Analytical workflow

The model keeps environmental calculation, uncertainty analysis and administrative reporting as distinct operations.

Spatial resolution and input datasets

The national analytical grid follows WorldClim 2.1 at 2.5 arc-minutes, approximately 4.6 km north–south in Ghana. SoilGrids predictions are sampled and depth-weighted to 0–15 cm, but their finer native resolution does not make the composite a 250 m product.

Membership functions and reference parameters

For any value x, clip(x) truncates values below 0 to 0 and above 1 to 1. Breakpoints are analytical assumptions and are varied in the scenario ensemble.

Opportunity components

Composite aggregation

Geometric aggregation reduces compensation between strongly contrasting components relative to an additive mean. The weights express a declared decision structure; they are not fitted causal coefficients.

Epistemic scenario analysis

The ensemble uses random seed 20260808. Each draw perturbs soil predictions, climate inputs, continuous-membership breakpoints, mapped source quality, distance decay and composite weights before recalculating every cell.

Structural-sensitivity analysis

Structural tests change complete weight structures or distance-decay lengths rather than sampling near the reference configuration.

Area-weighted district reporting

District statistics are calculated after the national grid. Each cell–district intersection contributes in proportion to overlap area with a cosine-latitude correction.

Spatial P10, median, P90 and IQR summarize variation among cells inside a district. They are different from the cell-level epistemic P10–P90 scenario envelope. Boundary areas requiring nearest-complete-cell evidence are quantified in the underlying reporting table.

Reporting and reproducibility

• Application version, analytical build and access date
• Selected layer and whether reference or custom weights were used
• Grid-cell coordinates and district reporting overlay
• H, B, F and composite values
• Cell-level P10–P90 width and structural minimum
• Any threshold and its non-empirical reporting status
• Evidence-source limitations and outstanding field measurements
• Statement that opportunity is not verified CDR or deployment approval

Analytical scope and validation requirements

GREW-OI is a national reconnaissance index, not a mechanistic estimate of weathering rate or verified carbon dioxide removal. Grid scores do not establish project feasibility, material safety, regulatory approval, carbon credits or investment returns. Site-level measurements and independent review are required for these determinations.

The opportunity grid is the established GREW-OI analytical build. The geochemical catalogue and source–field scenarios use separately documented published observations, empirical material distributions and explicit user inputs. They support evaluation of the interface and source–field workflow only. Measured, quality-assured source and batch records are required before scientific or operational use of feedstock outputs.

The primary input resolution, climate baseline, predictive soil inputs and reconstructed geology constrain interpretation. The scenario envelope represents declared epistemic assumptions rather than a calibrated probability of deployment success. Administrative summaries do not replace cell-level investigation.

Publications supporting KNUSTRocks

KNUSTRocks draws on NCEL research in national ERW screening and scenario-dependent opportunity assessment. The publications below document the development of the research framework. The application build and the published analytical versions are identified separately for reproducible use.

Application-build correspondence

The current application retains the GREW-OI 7.0 analytical archive: 400 epistemic draws, the reference and seven alternative weighting/distance-decay structures, and an 18.3-point median P10–P90 width. These are the values described by the equations and controls on this page.

The published Frontiers analysis uses a revised 3,200-member Latin-hypercube parameter ensemble, with geological-class, soil-support and district-support analyses treated separately; it reports a 10.30-point median parameter-envelope width. The two uncertainty products are not interchangeable. The publication is cited as a research foundation, not as a claim that every current application output reproduces its final analytical archive. Cite the application version and access date when reporting application-derived values.

Material evidence and application scenarios

The feedstock workspace supports geological reconnaissance, material comparison and sampling design. The source–field workspace evaluates inventories and idealized soil mixing for a specified composition, application mass and receiving layer. National opportunity, rock composition and receiving conditions remain separately identifiable analytical products.

Data sources, licensing and coverage

The composition catalogue derives from Gard, Hasterok and Halpin’s global whole-rock compilation, version 1.1.0, archived as 10.5281/zenodo.3359791 under CC BY 4.0. The associated database paper describes its structure and sources. Each displayed record includes original identifiers, analytical metadata and the underlying study reference. Missing citation fields for eight records were restored using their original reference keys, the archive bibliography and publisher metadata; corrections are preserved separately from the raw data. NCEL’s processing creates a harmonized subset; it does not transfer authorship of the observations.

The immutable archive contains 1,022,092 rows. An initial geographic/country extraction retained 10,209 candidate rows, including a bounding-box audit; country-based West African harmonization retained 5,420 records after filtering and deduplication. The application’s composition distributions use the curated Ghana literature set. The full catalogue remains available through the Record set control. Of 353 Ghana-labelled input rows, 81 had no supported analytical values or represented marine/biogenic material; 58 repeated rows were consolidated, yielding 214 analysis records with 14 original reference identifiers. The curated set contains 193 analysis records, 191 inferred sample families and 11 literature references in 10 conservative source groups. Twenty records retain unresolved USGS analytical attribution. One further record has geographic attribution inconsistent with its cited study and is excluded from the Ghana map and primary analyses. Two repeated sample families connect related publications; alternative oxide analyses remain separate records. No conflicting shared concentrations were found within the original same-reference merges at the declared tolerance. This does not establish record independence across references.

The source manifest stores SHA-256 checksums of the raw files. The whole-rock archive also reproduces the publisher’s MD5 checksum, 9b97b54887ee7184c6650c845b4e92d4. Original download files, retained rows, duplicate decisions, model metrics and processing scripts are preserved in the release dossier.

GEOROC and additional Ghana studies remain eligible for future additions after record-level provenance and reuse review. The selected whole-rock archive already contains GEOROC-derived records. The Hou et al. global soil-metal product reports exceedance probabilities rather than mg/kg baselines; it is not used in the mixing model. Its download was unavailable during this release audit. The coarse GLiM archive was reviewed but was not used to assign fine-scale chemistry.

Record processing and lithology assignment

The pipeline preserves source rows before transformation. It retains supported non-negative concentrations and original metadata, separates geographic eligibility from chemistry availability, and identifies repeated analyses by normalized sample identifier and original DOI (or source reference identifier). Missing fields may be completed from a duplicate; overlapping values must agree within 1% relative or 0.01 mg/kg absolute tolerance. Conflicting observations are retained in the audit and excluded from empirical model fitting. Related records stay in the same study group during evaluation.

Source rock names, protoliths and descriptive text determine the catalogue material classes: mafic/metamafic, intermediate/metaintermediate, felsic/metafelsic, ultramafic, alkaline, carbonate-rich, other metamorphic and unclassified. This descriptive crosswalk takes precedence over the archive’s computed whole-rock classification. For example, basaltic-andesitic material is included in the broad mafic/metamafic class; classifications and original names are both retained.

Records are plotted at their reported coordinates. The archive does not consistently specify datum, positional uncertainty, moisture preparation, weathering state or sampling date. Shared study localities are identified where four or more records have identical coordinates within a study. Such records contribute to the catalogue and empirical distributions but not to spatial evaluation. Duplicate coordinate assignments differing by more than 0.03° also exclude a record from spatial evaluation. The country audit uses a 2 km tolerance around the generalized Ghana land mask for coastal agreement.

A rock sample is a geochemical observation. A production source and a batch require separately documented operator, processing, represented mass and sampling support. The catalogue does not create these production attributes. Users can enter a batch identifier, laboratory/method and sampling/report reference for project-specific scenarios.

Lithology distributions and model release criteria

The operational composition model uses the curated Ghana literature set within each descriptive lithology class. Inferred sample families have equal weight; alternative analyses within a family share that weight. The model reports medians, 5th–95th percentiles and record, family and source-group counts. Quantiles use the generalized inverse of the weighted empirical distribution. They describe the observed sample distribution, rather than confidence limits for a class mean or a calibrated prediction interval for an unsampled production batch.

A fixed spatial candidate was evaluated for Ni and total Cr: the inverse-distance-weighted mean log concentration of the 12 nearest eligible training records, with distance floored at 1 km and weights proportional to inverse squared distance. The baseline is the training-fold median for the held-out sample’s lithology, using the training-wide median where fewer than three class observations exist. There is no random-row split: evaluation leaves out entire conservative source groups and, separately, 0.5° spatial blocks. Coordinates alone are candidate predictors; laboratory major elements are not used as unobserved map covariates.

Candidate evaluation requires at least 30 eligible observations across three source groups. Cobalt has 11 eligible observations from one source and therefore remains unevaluated spatially. These project-specific release criteria are development safeguards, not universal statistical thresholds. The performance rule requires at least 100 eligible observations across four source groups, a reduction of at least 10% in mean absolute log error relative to the baseline in both holdout schemes, and 80–98% empirical coverage for the candidate’s nominal 90% intervals. A national surface additionally requires audited geological predictors and an evaluated area of applicability. Neither Ni nor total Cr meets all release criteria. The application releases the observations and empirical class distributions; no continuous geochemical concentration surface is published.

The empirical class model remains useful for material comparison and scenario sensitivity. Expanded observations and geological attribution can subsequently support hierarchical or spatial models. Future evaluations must preserve study and spatial independence and distinguish sample variability from uncertainty in geological class assignment.

Units, chemical species and censored results

For compatible solid-mass bases, 1 ppm equals 1 mg/kg and 1 wt% equals 10,000 mg/kg. The interface displays major oxides in wt%; the calculation engine converts every composition to mg/kg. Oxide inventories remain oxide-equivalent masses, with the chemical formula stated. They are not presented as elemental Ca, Mg or Si. Total chromium and Cr(VI) occupy separate fields throughout input, simulation and comparison.

The source compilation does not consistently retain censoring flags for every historical value. Its numerical concentrations are presented as reported; zero concentrations are excluded from log-model evaluation. The uniform treatment of user-entered censored values is an explicit sensitivity assumption, not an estimate of the concentration distribution below detection. Laboratory values and baselines must use a compatible dry-mass basis; the historical archive’s variable preparation metadata are part of the observational scope.

Applied inventories and soil mixing

Application rate R is dry tonnes per hectare per application. The number n is an application count, with no imposed interval between events. The initial model is cumulative and assumes no loss, transformation or removal during the application sequence. Incorporation depth z is expressed in metres; the interface accepts centimetres.

A blank baseline yields applied inventory and, where specified, released inventory; it does not yield a mixed concentration. Negative ΔC describes dilution by material with a lower concentration. For independently drawn application batches, the calculation first sums the applied analyte mass across applications; the mass-weighted average composition is equivalent because each application has the same specified rate.

Joint panel resampling and input ranges

The main scenario uses 1,200 pseudorandom draws with seed 20261015. A published sample provides a fixed composition. A lithology scenario gives each inferred sample family equal probability and then selects a reported analysis within that family, preserving within-row multielement association and missing-value patterns. Repeated fixed-material applications reuse the same sampled panel. The alternative “independent panel” mode draws one panel per application. An analyte is summarized only for draws with all required application-panel values. If measured-panel availability is p, expected complete-draw frequency is p for fixed material and p to the power n for n independent applications. Results condition on complete measured panels. Percentiles and reference frequencies are displayed for at least 100 valid draws as a numerical reporting convention; valid counts accompany every analyte.

Optional lower and upper laboratory-composition, baseline, density, depth and release-fraction inputs define uniform scenario distributions. These user-defined ranges are sampled independently unless linked through the material-panel resampling. A single value is fixed. Release fractions are fixed values or bounded sensitivity ranges tied to a stated method or sensitivity basis. The interface reports medians and 5th–95th percentiles of the resulting scenario ensemble. These intervals express the specified empirical/input variation; they are not automatically calibrated probabilities for a future field or batch.

Rate-response curves use 400 draws per point, the same fixed seed, nine equally spaced rates and otherwise unchanged inputs. Reusing the seed makes comparisons reproducible. Scenario records include input panels, units, source identifier, data/calculation version, seed, selected mass basis, location and summaries. Up to four scenarios may be compared within the current browser tab.

Soil-grid support and river proximity

Coordinates identify a receiving location within Ghana. District selection navigates to an existing grid centre near the district’s grid-centre mean; users may then enter their actual location. Lookup requires the point to lie within a retained 2.5 arc-minute context cell. Values describe the sampled context grid centre, not a newly measured field.

SoilGrids 2.0 mean bulk density and coarse-fragment maps were retrieved from the official WCS at native 250 m support for 0–5 and 5–15 cm. The pipeline samples them at the established application-grid centres. Bulk density in cg/cm³ is multiplied by 10 to obtain kg/m³; coarse-fragment volume in per mille is divided by 1,000. Effective fine-earth density is calculated separately for each depth layer and then weighted by layer thickness:

The “Apply soil estimate” action sets this density, the fine-earth mass basis and 15 cm depth together. It applies the coarse-fragment correction once. Changing these inputs creates a user-defined scenario. No SoilGrids Ni or Cr baseline is supplied: those concentrations are not standard properties in this product. pH, CEC and organic carbon context retain the established 0–15 cm preprocessing described in the GREW-OI methods.

HydroRIVERS v1 provides the river network and linked reach/basin identifiers. Straight-line distances are computed in EPSG:32630 against the original line geometry, including nearby transboundary reaches. Display lines are simplified separately; this does not change distance calculations. The product represents rivers meeting its upstream-area or flow criteria, derived from 15 arc-second hydrography. Proximity is geographical context, not a model of flow routing, chemical transport or exposure.

Operationally released inventory

Release fractions are entered explicitly for the selected method or scenario. The user supplies fractions only where relevant to the selected analysis, with their laboratory or sensitivity basis. A total concentration, lithology distribution or probability of soil-threshold exceedance does not determine this fraction. Its field interpretation depends on mineralogy, particle size, pH, redox, solution chemistry and test duration.

Named references and portable scenario records

Reference comparison is optional and requires a name/version, jurisdiction or research purpose, analyte/species, feedstock or mixed-soil domain, numerical value and source URL. References use the declared dry-mass basis. The comparison reports the median and the frequency of valid scenario draws above that reference. This is a scenario frequency conditional on the declared input distributions, not a calibrated environmental-risk probability. It does not combine unrelated product and soil limits into an overall safety index.

Laboratory CSV imports and manually entered values are processed in the browser. No uploaded laboratory file is transmitted to the server. Scenario export is an explicit user action and contains the values entered in that tab. Source downloads, model artefacts and the site’s approved-access register remain separate systems. Normal sign-out and session expiry continue to clear the analytical browser session through page reload.

Sampling, model support and field interpretation

The compilation is a heterogeneous literature sample rather than a probability sample of Ghana’s rock resources. Study density, lithology coverage, analytical methods, censoring information and location precision vary. The catalogue includes altered and metamorphic materials where described; freshness and processing state are not consistently recorded. Broad class distributions may therefore combine material variability and study effects. They do not establish reserves, supplier availability, batch representativeness or universal composition for a mapped geological unit.

Empirical resampling balances inferred sample families; sources containing more families still contribute more strongly. The current distributions do not include an estimated hierarchical study effect, a calibrated batch-prediction model or a validated national geochemical surface. No Cr(VI) inference is made from total Cr. Historical moisture-basis and detection-limit metadata may require original-source review before comparison with a specific laboratory panel.

The mixing model assumes uniform incorporation and cumulative mass conservation. It does not predict mineral transformation, sorption, transport, erosion, plant uptake, drainage concentration, exposure or ecological effects. Context-grid values do not resolve within-field heterogeneity. SoilGrids means are fixed context estimates here; their full prediction uncertainty and cross-depth correlation are not propagated automatically. Optional ranges and empirical resampling form scenario intervals, not a validated environmental-risk probability.

Operational decisions may require representative batch sampling, mineralogy, species-specific assays, laboratory QA, release studies, local baseline measurements and independent assessment against the applicable framework. These evidence needs are documented here so that the analytical workspaces can present concise, quantitative results with clear provenance.