🎓 Lesson 15
D5
Calibrating Multispectral Bands for Iron Oxide Mapping in Waste Dumps
Calibrating multispectral bands means adjusting drone camera measurements so they accurately show how much iron oxide (like rust) is present in waste dumps.
🎯 Learning Objectives
- ✓ Explain the physical basis of iron oxide spectral absorption features in the visible-near infrared (VNIR) range
- ✓ Apply empirical line calibration using Spectralon® reflectance panels to convert DN to top-of-canopy reflectance
- ✓ Calculate normalized difference iron index (NDII) and compare results against laboratory-determined Fe₂O₃ concentrations
- ✓ Analyze calibration residuals to diagnose atmospheric or sensor-induced errors in waste dump surveys
- ✓ Design a field calibration protocol compliant with ASTM E2795–23 for drone-based multispectral mineral mapping
📖 Why This Matters
Waste dumps at iron ore mines often contain residual hematite and goethite—valuable indicators of processing inefficiency or environmental risk (e.g., acid rock drainage potential). Uncalibrated drone imagery may misclassify iron-rich zones as barren material, leading to missed recovery opportunities or flawed geochemical risk assessments. In one Pilbara operation, uncalibrated NDVI-like indices overestimated goethite content by 42%, delaying remediation planning by 8 weeks. Calibration isn’t just academic—it’s the bridge between pixels and pounds of recoverable iron.
📘 Core Principles
Iron oxides exhibit distinct electronic transition absorptions: hematite (α-Fe₂O₃) shows strong, narrow absorption near 860 nm due to Fe³⁺ charge transfer; goethite (α-FeOOH) has broader features at 490 nm (Fe³⁺ d-d transition) and 920 nm (OH⁻ combination band). Multispectral sensors capture discrete bands straddling these features—but raw digital numbers (DN) are affected by solar irradiance, atmospheric scattering (Rayleigh & aerosol), sensor nonlinearity, and vignetting. Calibration removes these confounders in three tiers: (1) Radiometric: converting DN → radiance (W·m⁻²·sr⁻¹·nm⁻¹) using sensor gain/offset; (2) Atmospheric: converting radiance → surface reflectance via dark object subtraction (DOS) or MODTRAN-based correction; (3) Spectral: aligning band centers and widths to match laboratory spectral libraries (e.g., USGS SPECIMEN) using empirical line method (ELM) with calibrated reflectance panels.
📐 Empirical Line Calibration (ELM)
ELM uses known reflectance targets (e.g., 5% and 50% Spectralon® panels) to derive linear gain and offset per band. It corrects for illumination and atmospheric path effects under uniform conditions and is preferred over DOS for mining applications where panel placement is feasible.
Empirical Line Calibration (ELM)
ρ = m × DN + bConverts raw digital number (DN) to unitless surface reflectance (ρ) using linear regression derived from reference panels.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| ρ | Surface reflectance | unitless (0–1) | Calibrated reflectance value for a given pixel and band |
| m | Gain coefficient | DN⁻¹ | Slope of linear relationship between DN and reflectance |
| DN | Digital number | counts | Raw sensor output value for a pixel |
| b | Offset coefficient | unitless | Y-intercept of linear relationship |
Typical Ranges:
Altum-PT Band 5 (840–880 nm): 0.00010 – 0.00015 DN⁻¹
Altum-PT Band 3 (550–570 nm): 0.00008 – 0.00012 DN⁻¹
💡 Worked Example
Problem: A DJI M300 RTK with MicaSense Altum-PT captures a waste dump at 10:30 AM local time. Raw DNs for Band 5 (840–880 nm, targeting hematite) are: 5% panel = 1,240 DN, 50% panel = 4,890 DN. Measured panel reflectances: ρ₅% = 0.052, ρ₅₀% = 0.508. Calculate reflectance for a pixel with DN = 3,150.
1.
Step 1: Compute gain (m) = (ρ₅₀% − ρ₅%) / (DN₅₀% − DN₅%) = (0.508 − 0.052) / (4890 − 1240) = 0.456 / 3650 = 0.0001249
2.
Step 2: Compute offset (b) = ρ₅% − m × DN₅% = 0.052 − (0.0001249 × 1240) = 0.052 − 0.1549 = −0.1029
3.
Step 3: Apply ρ = m × DN + b = 0.0001249 × 3150 − 0.1029 = 0.3934 − 0.1029 = 0.2905
Answer:
The calibrated reflectance is 0.291 (29.1%), which falls within the typical hematite-bearing waste dump range of 0.15–0.45.
🏗️ Real-World Application
At Rio Tinto’s Yandicoogina mine (Western Australia), engineers deployed an Altum-PT on a Matrice 300 RTK to map iron oxide distribution across a 120-ha waste dump. Using four 30×30 cm Spectralon® panels (5%, 20%, 50%, 99%) placed along the dump’s toe, they performed ELM calibration before each flight. Calibrated reflectance was used to compute the Normalized Difference Iron Index (NDII) = (R₈₆₀ − R₅₅₀) / (R₈₆₀ + R₅₅₀), validated against 42 core samples analyzed by XRF. Calibration reduced RMSE between drone-predicted and lab-measured Fe₂O₃ from 8.7 wt% (uncalibrated) to 1.3 wt% (calibrated), enabling accurate zoning for selective reprocessing.