Validation Guide Sec. 9.2 — smoke concentration (Fig. 9.49) vs. obscuration (Fig. 9.51) summary plots

[mcgratta]

Discussed in #16296

^{Originally posted by ManuelOsburg May 21, 2026}
Hi all,

I've been working through Section 9.2 Smoke of the Validation Guide and wanted to raise a question about the two smoke summary scatter plots — partly to make sure I'm reading them correctly.

Figure 9.49 ("Summary of smoke concentration predictions") reports mass concentration in mg/m³ and combines the NIST/NRC and FM/FPRF Datacenter datasets, with a Model Bias Factor of 2.46. Figure 9.51 ("Summary of smoke obscuration predictions") reports obscuration in %/m for the FAA Cargo Compartments dataset, with a Bias Factor of 1.02. Both figures for reference:

Looking at the experiment data files, all three datasets appear to be based on optical (light-extinction) measurements — the FAA experiment files use LT_* channels in %/m, NIST/NRC has a "Smoke Conc." channel from a laser extinction photometer, and the FM/FPRF data center work also relies on a laser signal. Mass concentration and extinction are linked through $K = K_m · c$, so for a single sample they're essentially interchangeable.

My concern is that the way the two summary plots are set up can leave a reader with the impression that "mass concentration is hard to predict, optical extinction is easy" — when the actual finding (Sec. 9.2.3) is that FDS overpredicts smoke in the closed-door NIST/NRC tests because agglomeration and wall deposition aren't modeled. That's an error in the transported soot inventory, and it would show up under either unit. The two figures differ in two ways at once — the reported quantity and the experiment set — so the unit and the physics end up confounded.

A couple of questions:

1. Is there a specific reason the FAA data is summarized as obscuration while NIST/NRC and FM/FPRF are summarized as mass concentration? I'd guess it's because a reliable $K_m$ isn't available for the FAA fuels, whereas it is for heptane/propylene — but I wanted to confirm.
2. Would it be worth either harmonizing the quantity across both plots, or adding a sentence near the figures clarifying that the difference in bias reflects the experimental configurations (well-ventilated vs. closed/under-ventilated) rather than the choice of unit?

Happy to be corrected if I've misunderstood the intent here. Thanks for all the work on the Guide.

Best regards,
Manuel

[drjfloyd]

1. I think this is just because the data was reported as soot density in the experiments for NIST/NRC and FM/FPRF and obscuration in FAA, and we just made the FDS outputs reflect that.

2. I think it makes sense to do all the plots as one unit. Given that these are all obscuration measurements of some kind rather than a gravimetric soot measurement, I would agree that it makes sense to use %/m. However, in cases where we know the actual extinction coefficient we should use that in lieu of the default value of 8700.

For the NIST/NRC closed cases there are multiple factors at play that include deposition, agglomeration possibly changing the extinction coefficeint over time, and soot oxidation (soot being re-entrained into the fire). The NIST/NRC cases do have deposition for a single assumed soot size. The cases lack agglomeration and oxidation. At one point I had played with that some. Something to return to at some point.

We (FSRI) did do a short series of tests where we measured soot size distributrion, deposition, gravimetric soot density, and obscuration. That report is in the queue for this year. Wuppertal is doing a visibility workshop this fall before the Juelich summer school which I hope will be an opportunity to discover other recent work we might be able to use.

I go on leave tomorrow and get back a couple days before I will need to fly to IAFSS.

[ManuelOsburg]

Thanks Jason — that all makes sense, and I agree on plotting everything in %/m and using the measured extinction coefficient where known.

On new data — we ran full-scale compartment fire tests in Leipzig, together with the Wuppertal and Jülich teams, that should also be a solid basis for smoke/visibility validation; posted as a round-robin call in #16279.

I'll be at the Wuppertal workshop and the Jülich summer school this fall as well, and look forward to the exchange there.

Enjoy your leave, and safe travels to IAFSS.

[mcgratta]

I am going to convert the NIST/NRC results into %/m.

[mcgratta]

If I've done my math right, I get this

I switched back to the linear plot format because now the ranges are 0 to 100 and there's no need for the log-log. This plot is a bit more clear.

We'll add the FM results to this plot and do away with the "Smoke Concentration" category. The obscuration plot makes more sense in practice.

[mcgratta]

I updated the FDS Validation Guide. I'll leave the issue open until we convert the FM results to %/m.

[ManuelOsburg]

One small terminology point from reading the diff: the NIST/NRC section text and figure captions now say "transmission", while the dataplot Quantity entry is Smoke Obscuration. Transmission and obscuration are complementary (obscuration = 100 − transmission) rather than the same thing — it might be worth using one term consistently so the section text, captions, and scatter plot all match. Doesn't affect the results, just the wording.

[ManuelOsburg]

A follow-up on how the bias factor in the new obscuration summary should be read.

Part of it is a scale effect. Since FDS computes the extinction coefficient as $K = K_m \rho Y_s$ with constant $K_m$, $K$ is linear in soot mass — the $K$ bias equals the concentration bias and is independent of the measured value. The obscuration bias is not. With optical depth $\tau = K L$, an overprediction of $K$ by a factor $\mathrm{Bias}_K$ means the predicted obscuration is $1 - e^{-\mathrm{Bias}_K \tau}$ against a measured $1 - e^{-\tau}$, so:

$$\mathrm{Bias}_O(\tau) = \frac{1 - e^{-\mathrm{Bias}_K \tau}}{1 - e^{-\tau}}$$

For optically thin points ($\tau \to 0$), $\mathrm{Bias}_O \to \mathrm{Bias}_K$; for optically dense ones ($\tau \to \infty$), $\mathrm{Bias}_O \to 1$ — any bias is compressed toward apparent agreement. The closed-door NIST/NRC cases sit at optical depths around 1 and above, where the compression is already substantial (at $\tau \approx 1$, roughly $2.5 \to 1.45$). So a given soot-mass error shows up as a smaller number in obscuration than in concentration, without the model having changed.

This doesn't argue against the %/m switch. As a representation-independent alternative, a $K$-based bias factor would be invariant to the scale effect — it stays equal to the concentration bias — and is also a natural fire-safety quantity, since $K$ is the input to Jin's visibility correlation.

[mcgratta]

TRANSMISSION is an FDS output in units of %/m while PATH OBSCURATION is in units of %. So now in the Validation Guide, you will see plots comparing measured and predicted transmission. The scatterplot is just 1-trans.

As far as the bias factor, this is just calculated automatically by the scatterplot routine. If we want to express the bias in terms of K rather than obscuration, we'd have to plot K, or have two plots, one for K and one for obs. Given that we are going to get rid of one scatterplot for smoke density, I'd rather not introduce another quantity.

[ManuelOsburg]

Fair enough — thanks for working through this.