Lume / Wastewater Dilution Test 50056

Wastewater Dilution Test — 50056

Enter dilution annotation windows below. Sensor data and regression update automatically.

Experiment Overview

On 20 May 2026, Lume sensor 50056 was tested in a controlled wastewater dilution series at eight concentration steps: 0%, 1%, 5%, 10%, 25%, 50%, 75%, and 100% raw wastewater in tap water. Each window lasted approximately 8–13 minutes. Two reference sensors were run in parallel: an Aqualog benchtop EEM fluorometer (IFE-corrected, treated as ground truth) and a Turner C3 in-situ fluorometer measuring CDOM (365 nm) and TLF (280 nm). Lab turbidity (NTU) was measured by grab sample.

The LUME sensor swept LED power (64–1024) and SiPM bias (2800–3400 V) continuously throughout the experiment, generating 108,173 raw readings across 44 LED×bias combinations. The calibration combo (LED 512 / bias 3000) was used for fluorescence analysis; TOF backscatter (SPS, kcps/SPAD) provided an independent turbidity channel.

8
Concentration steps
(0–100% WW)
108K
LUME SiPM readings
across 44 combos
3
Reference sensors
(Aqualog, Turner C3, NTU)
0.999
LOO-CV R² (3-channel
TLF + low-LED + SPS)
Loading…

LED/Bias Combo Comparison

Inner filter effect (IFE): At high concentrations, the sample absorbs its own excitation and emitted light, causing the fluorescence signal to plateau then decrease. The IFE onset is a property of the sample's optical density, not the detector — changing LED power scales all signal values proportionally but does not shift the concentration at which the peak occurs (the peak is at OD ≈ 0.05 regardless of I₀). Implication for this dataset: if 50% and 75% are already indistinguishable at LED 512 (flat top of the IFE curve), they will remain indistinguishable at any other LED power because the curve shape is unchanged — only its amplitude shifts. What the comparison below can reveal is whether any combo suffers from additional SiPM microcell saturation on top of IFE (signal plateaus earlier than expected), vs. combos where the SiPM is in its linear range and the true IFE shape is visible. Lines marked ⚠ are non-monotonic (IFE-dominated). Use Normalize to compare curve shapes independent of LED power. One combo per LED level shown (bias nearest to cal).

⚠
IFE is a fundamental optical constraint — no LED or bias setting avoids it
Every combo tested peaks at ~25–50% wastewater and then decreases. This is not a SiPM saturation artifact: it occurs even in combos where the detector is well within its linear range. The IFE onset concentration is set by the sample's UV absorbance (OD ≈ 0.05), not by LED power — scaling LED power shifts signal amplitude but not the peak location. Above ~50% wastewater, no fluorescence measurement alone can reliably rank-order concentration.

TLF Signal / Turbidity / Temperature — Sensor 50056

Turner C3 — In-Situ CDOM, TLF & Temperature

Turner C3 grab-sample time series collected in parallel with the dilution experiment. Timestamps are on the Turner’s own clock (not aligned with LUME UTC). Each concentration window has 180 readings at ~1 Hz; background shading shows the annotated windows. Note CDOM continues rising through 75–100% while TLF (tryptophan-like) peaks at ~25% and decreases at high concentrations — inner filter effect on the 280 nm excitation channel.

📋
Turner CDOM (365 nm) stays linear where TLF (280 nm) collapses — confirming the IFE mechanism
Turner TLF visibly peaks at ~25% and reverses, matching the IFE prediction for 280 nm excitation. Turner CDOM, excited at 365 nm (longer wavelength, lower sample absorbance), continues rising through 75–100% with far less attenuation. This wavelength-dependence is the fingerprint of IFE and validates the physical explanation: the problem is UV absorption by the sample, not detector saturation. A sensor using longer excitation wavelengths (CDOM range) would extend the linear regime, but at the cost of TLF-specific sensitivity to tryptophan-like compounds.

Wastewater Concentration vs Fluorescence Signal

Lab Turbidity vs TOF Backscatter

✔
Fluorescence is linear and sensitive at low concentrations; turbidity (SPS) is monotone across the full range
TLF fluorescence signal is a reliable predictor below ~25% wastewater, enabling detection at <1% dilution where turbidity changes are too small to measure. Above ~50%, fluorescence saturates due to IFE while SPS backscatter continues rising monotonically through 100%. The two channels are complementary by design: fluorescence for sensitivity at trace concentrations, turbidity for linearity at high loads. Neither channel alone spans the full dynamic range.

Low-end sensitivity (0%→1% step): LUME TLF signal increases by +141.6 counts (+36.8%) with a signal-to-noise ratio of 4.73 — compared to Turner C3 TLF which shows a similar +41.7% relative change but an SNR of only 2.70. LUME resolves the 0%→1% step ~75% more reliably than the Turner TLF on a per-reading basis (180 readings averaged here vs LUME’s continuous stream). Turner CDOM achieves SNR = 5.23 for the same step but measures organic carbon rather than tryptophan-like fluorescence.

Multivariate Regression — Predicted vs Actual Concentration (LOO-CV)

OLS fits using leave-one-out cross-validation: each point is predicted by a model trained on all other windows. R²LOO is an honest out-of-sample estimate — lower than in-sample R² (shown in parentheses). Left: TLF + TOF + temperature. Right: TLF + TOF only (temperature dropped — only 1.4 °C variation, costs a free parameter).

TLF + TOF + Temperature

TLF + TOF only

📌
Temperature adds no predictive power in this experiment — only 1.4 °C variation across all windows
The 3-predictor model (TLF + TOF + temperature) and the 2-predictor model (TLF + TOF only) achieve effectively identical LOO-CV R². Temperature was stable throughout the dilution series, making it a free parameter that costs model degrees of freedom without adding information. In a real deployment where temperature varies significantly, it may improve fluorescence correction — but it contributes nothing here.

Multi-Channel Model — Full-Range LOO-CV

The cal combo (LED 512 / bias 3000) saturates above ~25% — mon2 is flat at 50%, 75%, and 100% (SiPM hardware ceiling). At LED 64 / bias 3425 the SiPM stays below saturation, so the inner filter effect is visible: signal peaks at ~50% then decreases at 75% and 100% (SNR ≈ 7.6 between the three). Combined with SPS (TOF turbidity), this 3-predictor OLS achieves R²LOO = 0.999 across all 8 windows. Left: normalized (0–1) channel responses — shows where each channel saturates. Right: LOO-CV parity plot for the combined model.

Normalized Channel Responses

3-Predictor LOO-CV (cal TLF + low-LED TLF + SPS)

⭐
3-channel model achieves R²LOO = 0.999 across all 8 concentration windows — full-range prediction with no blind spots
Combining three complementary signals — cal TLF (LED 512 / bias 3000, sensitive at 0–25%), low-LED TLF (LED 64 / bias 3425, visible IFE shape at 25–75%), and SPS turbidity (monotone through 100%) — the OLS model predicts concentration with near-perfect accuracy across the entire 0–100% range. Each channel contributes information in the regime where the others saturate or invert. This validates the LUME multi-channel architecture: a single fluorescence channel cannot solve IFE, but three channels with different operating points together span the full dynamic range.

Reference Sensor Comparison — Normalized Signal vs Concentration

All signals normalized 0–1 for shape comparison. Aqualog (benchtop EEM, IFE-corrected): both TLF and CDOM are linear to 100%. Turner C3: CDOM (365 nm) nearly linear; TLF (280 nm) peaks near 25% and reverses. LUME in-situ TLF saturates above ~25–50% — same IFE behavior as all in-situ fluorescence sondes.

Sensor Performance vs Aqualog Truth

Each sensor signal (normalized 0–1) vs Aqualog TLF (normalized, ground truth). A perfect sensor lies on the 1:1 line with R² = 1.0. R² shown for full range, linear regime (0–25%), and IFE regime (50–100%).

✔
Key Finding: LUME TLF matches Turner C3 in the linear regime and is 3× more predictive at high concentrations
0–25% wastewater (linear regime): LUME TLF R² = 0.921 vs Turner C3 TLF R² = 0.908 — statistically equivalent. A continuously deployed in-situ sensor matches a dedicated grab-sample reference in the range where fluorescence is reliable.
50–100% wastewater (IFE regime): LUME TLF R² = 0.798 vs Turner C3 TLF R² = 0.289. Turner TLF signal fully reverses direction past ~25% (IFE inverts the response), destroying predictive power. LUME TLF also drops at 100% due to IFE, but attenuates more gradually — signal compression rather than reversal — resulting in nearly 3× higher R² in the high-concentration regime.
💡
Unexpected Finding: LUME Turbidity (SPS) tracks Aqualog TLF and CDOM almost exactly across the full range
LUME’s TOF backscatter signal (SPS kcps/SPAD) achieves R² = 0.982 vs Aqualog TLF across all concentrations — higher than every dedicated fluorescence sensor tested, including the benchtop Aqualog CDOM and the Turner C3. At 50–100% wastewater, SPS achieves R² = 1.000, a perfect rank-order match where all fluorescence sensors degrade. This suggests that in high-strength wastewater, particle loading (turbidity) is a robust proxy for organic concentration and may be preferable to fluorescence as a primary signal. The LUME multi-channel architecture — combining TLF for sensitivity at low concentrations with SPS for linearity at high concentrations — spans the full dynamic range that no single-channel sensor achieves.

Add Annotation

Concentration Annotations

No annotations yet. Add one using the form above.