Hyperspectral Mineral Mapping
of the Mountain Pass REE Deposit

1. Introduction

Mountain Pass, California is the largest and highest-grade Rare Earth Element (REE) deposit in North America, and one of the most significant globally. The deposit is hosted within a Precambrian (~1.4 Ga) carbonatite intrusion that pierces Proterozoic gneisses and schists of the Clark Mountain Range. The dominant ore mineral is Bastnaesite [(Ce,La,Nd)FCO₃], a fluoro-carbonate that can carry 60–70% REE by oxide weight and is the world's primary source of light rare earth elements (LREE), including cerium, lanthanum, and neodymium.

This study applies hyperspectral remote sensing using AVIRIS-NG VNIR imagery (547–903 nm, 72 spectral channels) and the Spectral Angle Mapper (SAM) classifier to map Bastnaesite-bearing surface exposures. The ENVI software platform with custom IDL scripting was used for all processing steps. The workflow covers:

• Pre-processing and bad-band masking of the AVIRIS-NG reflectance cube
• SAM classification against a resampled Bastnaesite spectral library
• Rule-image thresholding to refine the classification
• Raster color-slice analysis to visualize spectral-angle distributions
• Classification statistics and geological interpretation

2. Dataset & Study Area

3. Classifier Selection

Several ENVI classifiers are suitable for hyperspectral mineral mapping. The principal options considered were:

4. Methodology

4.1 Bad Band Handling

The AVIRIS-NG image contains 72 spectral bands but bands 43 (762.39 nm) and 44 (767.40 nm) fall within the atmospheric O₂-A absorption window and carry unreliable radiance values. These bands are flagged in the image header (bbl = [..., 0, 0, ...]). The spectral library was pre-resampled to 70 bands (excluding these bad bands). For the SAM tool, a 72-element mean array was constructed by inserting 0.0 at positions 43 and 44, with the 70 library values filling the remaining slots. ENVI's SAM algorithm automatically excludes bad-band positions from the spectral angle computation.

4.2 SAM Classification

The SpectralAngleMapperClassification ENVI task was run with the following configuration:

4.3 Rule Image Analysis & Threshold Selection

The SAM rule raster was analysed to identify the distribution of spectral angles. Key findings:

• Minimum angle across valid pixels: 0.402 rad (~23°)
• Maximum angle: 0.798 rad (~46°)
• Mean angle: ≈ 0.479 rad
• A natural break was identified at 0.46 rad: pixel counts jump from ~2,938 to ~36,493 in consecutive 0.01 rad histogram bins — indicating a distinct spectral population boundary.

4.4 Classification Refinement — RuleRasterClassification

The RuleRasterClassification ENVI task was applied to the SAM rule raster with:

4.5 Mask Integration

The binary mask file (MountainPass_mask.dat, value 0 = no-data) was applied to assign all no-data pixels to the dedicated class Masked Pixels (326,642 pixels), producing the three-class final classification raster SAM_classification_final.dat. The HDR was written manually to ensure correct ENVI Classification metadata (file type, class lookup, class names, and UTM map info).

4.6 Raster Color Slice

To visualize the continuous distribution of spectral angles, the ColorSliceClassification ENVI task was run on the SAM rule raster:

5. Results

5.1 Scene Overview — CIR Composite & Final Classification Map

Figure 1. AVIRIS-NG Color Infrared (CIR) composite — NIR / Red / Green bands. The active open pit (dark, circular benches, centre-left) is clearly visible. Red tones = NIR-bright vegetation; gray/tan = exposed mine rock; brown = Mojave Desert alluvium. Scene extent ≈ 3.3 × 3.6 km.

Bastnaesite (Class 1) — 4,544 px

Unclassified (Class 0) — 696,862 px

Masked Pixels (Class 2) — 326,642 px

Figure 2. SAM final classification map. Orange pixels (Bastnaesite) have spectral angle < 0.46 rad to the reference spectrum. Clusters align with the active pit and ore stockpiles seen in Fig 1.

The 4,544 Bastnaesite-classified pixels (orange) are spatially concentrated in two principal zones: (1) the northern active open-pit mine, where the carbonatite ore body is exposed by mining benches; and (2) several smaller clusters in the eastern central area, corresponding to mine waste dumps and stockpiles. The dispersed pattern of orange pixels across terraced mine floors is consistent with heterogeneous exposure of the carbonatite ore and country-rock contacts.

5.2 SAM Rule Image & Raster Color Slice

Figure 3. SAM rule raster (spectral angle to Bastnaesite, radians). Brighter pixels indicate smaller spectral angle (higher similarity to Bastnaesite). The bright cluster in the northern sector corresponds to the active open-pit mine. Valid-pixel angle range: 0.402–0.798 rad.

Figure 4. Raster Color Slice of SAM rule image. 10 intervals × 0.01 rad, range 0.40–0.50 rad. Colors span the full IDL Rainbow color table (dark violet = 0.40 rad → purple → blue → cyan → green → yellow-green → yellow → orange → red = 0.50 rad). Black pixels are outside the 0.40–0.50 rad range (masked or unclassified). Most valid pixels fall in the yellow-green to orange bins (0.45–0.49 rad); the rare blue/cyan clusters in the open-pit area represent the most Bastnaesite-similar exposures (lowest spectral angle, 0.40–0.44 rad). The color bar on the right identifies each 0.01 rad interval.

5.3 Spectral Profile Analysis — Continuum-Removed Comparison

Figure 5. Continuum-removed reflectance of the Bastnaesite SLI reference spectrum (red dashed) versus the mean spectrum of all SAM-classified pixels with angle < 0.46 rad (blue solid, n = 4,544). Continuum removal was performed by dividing each spectrum by its upper convex-hull continuum, making both scale-invariant and enabling direct shape comparison. Red tick marks indicate known Bastnaesite VNIR absorption bands (REE crystal-field and Fe³⁺ charge-transfer transitions at ~570, 617, 670, 742, 800, 858 nm). The image mean is nearly flat after continuum removal (<6% band depth), confirming that the VNIR window (547–903 nm) lacks the diagnostic absorption depth of the reference; the SAM match is driven by spectral slope, not absorption features.

5.4 Spectral Angle Distribution Summary

6. Classification Statistics

6.1 Pixel Count Summary

6.2 Mean Spectral Reflectance by Class (Selected Bands)

Reflectance values are stored as integer × 10,000 in the image. Divide by 10,000 to obtain fractional reflectance. Bad bands (762, 767 nm) are shown in italics — values are unreliable.

⚠️ Bad band values in parentheses are unreliable due to O₂-A atmospheric absorption and were excluded from the SAM spectral angle computation.

7. Geological & Mineralogical Discussion

7.1 Geology of Mountain Pass

The Mountain Pass deposit is a Proterozoic carbonatite intrusion (~1.4 Ga) emplaced into Proterozoic gneisses and schists of the Mojave Desert terrane. The carbonatite body (locally called the "Sulfide Queen" body) consists dominantly of calcite and dolomite carbonatite with subordinate baritocelestite, barite, and accessory sulfides. The REE mineralization is concentrated in Bastnaesite, a fluoro-carbonate mineral with the formula (Ce,La,Nd)FCO₃. Cerium is the dominant rare earth cation (~49%), followed by lanthanum (~34%) and neodymium (~12%).

Bastnaesite forms yellowish-waxy to brownish hexagonal crystals, commonly intergrown with barite, calcite, and dolomite. The ore body has been mined intermittently since the 1950s. At its peak in the 1980s–1990s, Mountain Pass supplied ~25% of global REE production. After a period of closure, production restarted in 2017 under MP Materials Corp.

7.2 Spectral Properties of Bastnaesite

The spectral signature of Bastnaesite in the VNIR (400–1000 nm) is controlled by two mechanisms:

• Rare earth electronic transitions (f–f transitions): Neodymium³⁺ is the primary absorbing REE cation in the VNIR, producing characteristic absorptions at approximately 577 nm, 740 nm, 800 nm, and 870 nm. These are electric-dipole forbidden f–f transitions that produce relatively narrow, low-contrast absorption bands.
• Carbonate matrix: The fluoro-carbonate structural framework provides a high-reflectance continuum in the VNIR, giving Bastnaesite its characteristic bright yellowish-cream colour. The strong SWIR carbonate combination and overtone bands (at ~2.0, 2.16, 2.33 µm) are not accessible in this AVIRIS-NG sub-image (which covers only 547–903 nm).

The classified Bastnaesite pixels exhibit a steeply rising NIR reflectance (0.2733 at 547 nm → 0.4387 at 903 nm), consistently 1.5–1.7× brighter than the unclassified background terrain. This behaviour is a hallmark of bright carbonate rocks and is consistent with exposures dominated by carbonatite mineralogy (calcite, dolomite, barite, and Bastnaesite).

7.3 Interpretation of Large SAM Angles

The minimum SAM angle across all valid image pixels is 0.402 rad (~23°). This is unusually large compared to standard mineral mapping thresholds of <0.05–0.10 rad for confident pure-mineral identification. The elevated angles are attributable to the following factors:

1. Spectral mixing: At 3.4 m GSD, each pixel integrates reflectance from Bastnaesite ore, carbonate gangue (calcite/dolomite/barite), weathered soil, mine infrastructure, and desert pavement. No pixel is spectrally "pure" Bastnaesite.
2. Limited spectral range: The 547–903 nm VNIR window captures only the shorter-wavelength Nd³⁺ f–f transitions. The most discriminative carbonate absorption features (SWIR 1.5–2.5 µm) are outside the sensor range. AVIRIS-NG full-range data (covering to 2,500 nm) would dramatically improve classification accuracy.
3. Surface weathering: Chemical weathering transforms primary Bastnaesite into secondary phases such as cerianite (CeO₂) and REE-bearing clay minerals, which have different spectral signatures.
4. Mine infrastructure: Open-pit mines contain haul roads, conveyors, buildings, and water bodies that contaminate pixel spectra and introduce atypical spectral shapes.
5. Shadow effects: Deep mine benches and talus slopes create intra-pixel illumination variation that alters apparent reflectance — a factor SAM is designed to mitigate (angles are illumination-insensitive) but cannot fully eliminate when sub-pixel shading changes the continuum shape.

7.4 Spatial Significance of Classified Pixels

The 4,544 Bastnaesite-classified pixels cover approximately 5.25 hectares (52,509 m²), representing 0.65% of the valid image area. Their spatial distribution is geologically meaningful:

• Northern cluster (active open pit): The densest orange cluster in the northern portion of the scene corresponds to freshly exposed carbonatite faces on the active mine benches — precisely where unweathered Bastnaesite ore would be expected at the surface.
• Eastern clusters (dumps and stockpiles): Smaller clusters in the central-east coincide with mine waste dumps or ore stockpiles, consistent with REE-bearing material being concentrated by mining operations.
• Scattered southern pixels: Isolated orange pixels in the southern half of the scene may represent natural carbonate outcrops of the Clark Mountain Formation limestones, which are spectrally similar (though not geochemically identical) to the Bastnaesite-bearing carbonatite.

7.5 Recommendations for Improved Mapping

• Full AVIRIS-NG spectrum (to 2500 nm): Adding the SWIR spectral range would allow diagnostic carbonate combination bands (2.0, 2.16, 2.33 µm) to be included, dramatically reducing SAM angles and improving discrimination between Bastnaesite, barite, calcite, and dolomite.
• Matched Filter or Linear Spectral Unmixing: These methods can estimate sub-pixel mineral fractions, more appropriate for the spatially mixed environment of an open-pit mine.
• Integration of field spectra: Acquiring in-situ spectra with a field spectrometer (e.g., Spectra Vista HR-1024) from known ore benches would provide site-specific spectral endmembers, reducing the library-to-scene mismatch.
• Multi-date analysis: Change detection between pre- and post-mining flight lines could highlight newly exposed ore faces.

8. Conclusions

9. ENVI Modeler Workflow

The complete SAM classification pipeline has been encoded as a reusable ENVI Modeler workflow (MountainPass_Bastnaesite_SAM.model). This allows the analysis to be repeated on any compatible hyperspectral dataset with minimal configuration — only two inputs need to be provided at runtime.

Figure 6. ENVI Modeler workflow diagram for the Mountain Pass Bastnaesite SAM classification pipeline. Six processing stages are connected left-to-right. Teal nodes = user inputs / outputs; navy nodes = ENVI tasks; orange nodes = display views. White comment balloons annotate key parameters on each task node.

Workflow Structure

How to Use the Model

1. In the IDL for VS Code extension, open ENVI Modeler and load MountainPass_Bastnaesite_SAM.model via File > Open.
2. Click Run. A parameter dialog will appear with two user inputs:
   • Hyperspectral Image — select the AVIRIS-NG (or similar) reflectance raster.
   • Spectral Library — select an ENVI .sli spectral library containing the mineral reference spectrum.
3. The workflow automatically resamples the library to the image bands (via EndmemberCollection), runs SAM, refines with the 0.46 rad threshold, generates the color slice, and displays all three results in ENVI views.
4. To change the classification threshold, open the RuleRasterClassification node and edit threshold. To change the mineral target, supply a different spectral library and optionally add spectra_names to the EndmemberCollection node to select specific spectra from a multi-spectrum library.

Output Files