What an attenuated reading actually means
A darker location receives less light in the measured direction. That does not establish that cells absorbed the entire difference. Photons can be absorbed, scattered into other directions, reflected at interfaces or transported laterally outside the observed region. Low transmission and high biological utilisation are therefore not equivalent claims.
This chapter deliberately analyses a bounded model: a directional signal decreases exponentially along a planar optical path. The model has no lateral input, back-reflection or spatial biomass variation. Effective attenuation coefficient k describes the measured pattern; it is not automatically called a pure absorption coefficient. All numerical observations are synthetic values generated exactly from that model.
The original paper by Cornet and Dussap relates light transport to biological activity through absorption, scattering, spectrum and geometry. It demonstrates why the full relationship is more demanding than a single exponential fit. No species-specific literature coefficient or claimed prediction accuracy is transferred to other cultures here.
Build the exponential model with explicit units
For constant effective attenuation coefficient, dI/dz = −kI. Integrating from z = 0 to depth z gives I(z) = I₀ exp(−kz). I and I₀ share units, here µmol m⁻² s⁻¹; z is in m and k in m⁻¹. Only the dimensionless product kz belongs in the exponent. Combining a path expressed in cm with k in m⁻¹ introduces a factor-of-100 error.
The teaching case assumes I₀ = 300, path length L = 0.05 m and k = 5 + 60X. X is dry biomass concentration in g/L, numerically equal to kg/m³. The intercept 5 has units m⁻¹, while the slope 60 has units m²/kg when X is in kg/m³. Both are invented inputs, not optical constants of an alga. X = 0 represents a separately assumed background from the medium and optical arrangement.
A linear dependence of k on X can be tested as a local approximation. It is not universal merely because a dilution series fits well in one apparatus. Cell size, pigmentation, aggregation and the scattered fraction reaching the detector can change. Different species, growth phases or measurement geometry require renewed assessment.
Four concentrations produce different light profiles
| X / g L⁻¹ | k / m⁻¹ | kL | I(L) / µmol m⁻² s⁻¹ | Depth mean / µmol m⁻² s⁻¹ |
|---|---|---|---|---|
| 0.25 | 20 | 1.00 | 110.36 | 189.64 |
| 0.50 | 35 | 1.75 | 52.13 | 141.64 |
| 1.00 | 65 | 3.25 | 11.63 | 88.73 |
| 1.50 | 95 | 4.75 | 2.60 | 62.61 |
At X = 1 g/L, k = 65 m⁻¹ and I(L) = 300 exp(−65·0.05) = 11.63 µmol m⁻² s⁻¹. The exiting directional signal is approximately 3.88% of its inlet value. The remaining 96.12% initially represents loss from that measurement channel. Without additional optical accounting, it cannot be equated to cellular absorption or photosynthetic output.
The table also shows why inlet intensity cannot fully describe a culture. Every case has the same front illumination and geometric path, yet rear readings and depth means differ markedly. Increasing biomass during cultivation can change the internal field while lamp current and front sensor remain constant. Controlled incident light therefore does not establish constant exposure of individual cells.
Normalise the spatial integral by path length
The depth mean is Ī = (1/L)∫₀ᴸI₀ exp(−kz)dz = I₀[1 − exp(−kL)]/(kL). The factor 1/L is essential. The unnormalised integral has units of photon flux density multiplied by length, not units of a mean intensity. Dimensional checking is straightforward: numerator and denominator of the ratio are dimensionless, leaving Ī with the units of I₀.
At kL = 3.25, Ī = 300[1 − exp(−3.25)]/3.25 = 88.73. Averaging inlet and outlet values instead gives (300 + 11.63)/2 = 155.82, considerably larger. That approximation follows a straight line between endpoints, whereas the assumed profile is exponential. Two boundary readings therefore do not automatically give the spatial culture mean.
Even a correct spatial mean is not automatically the mean biological stimulus. In an idealised two-zone example, cells spend equal time at I = 50 and I = 550. Mean light is 300. For the invented dimensionless response f(I) = I/(100 + I), f(300) = 0.75, whereas [f(50) + f(550)]/2 ≈ 0.58974. Nonlinearity makes the calculations differ. The function demonstrates mathematics only and is not a model of a specified organism.
Estimate attenuation from observations
If I₀ is independently known and signals are positive, ln[I(z)/I₀] = −kz. Fitting a line through the origin can estimate k under suitable error assumptions. Uncertain I₀ may require a free intercept, but that intercept can also absorb errors in the incident reference. High goodness of fit establishes neither correct I₀ measurement nor the proposed optical mechanism.
Log transformation changes error structure. A roughly constant absolute measurement error becomes relatively large at low readings. Near the dark signal, taking logarithms can strongly distort results or become impossible. Negative dark-corrected readings must not be replaced by arbitrary small positive values merely to permit a fit. Alternatives include fitting the original measurement scale, appropriate weighting or an explicit measurement model.
Every reference-data row is generated exactly from the stated exponential model. Apart from CSV rounding, fitting therefore returns the input coefficients. This verifies arithmetic, not empirical model validity. A genuine test needs independently measured profiles, adequate instrument-range coverage, replicates and deliberately varied concentrations. Assess fit together with residuals and the optical arrangement.
An optical threshold is not a biological compensation point
For calculation only, choose a signal threshold Ithr = 20 µmol m⁻² s⁻¹. It is not an experimentally established biological compensation point. Solving I₀ exp(−kz) = Ithr gives zthr = ln(I₀/Ithr)/k. With I₀ = 300 and k = 65, the threshold occurs at 0.04166 m. Approximately 83.3% of the 0.05 m model path lies above that directional signal threshold.
At X = 1.5 g/L and k = 95, the fraction falls to approximately 57.0%. This is a geometric depth fraction under specified optical assumptions, not the fraction of living cells or productive culture. Unequal residence probabilities, local cell densities, scattered light and time-dependent movement can change biological interpretation.
The study by Bernard and Lu on optimal optical conditions relates light distribution, growth and respiration within a defined model. Such compensation requires an explicitly defined net biological response. An arbitrary light threshold cannot replace biological measurements. The threshold used here exists solely to make the geometry calculation reproducible.
When the simple approximation becomes inadequate
| Assumption | Possible violation | Implication |
|---|---|---|
| Constant k along the path | Settling, aggregates or pigment gradients | One coefficient hides local differences |
| No lateral illumination | Reflective surroundings or adjacent modules | One-dimensional attenuation does not represent the apparatus |
| Same optical properties at equal dry mass | Photoacclimation, changing cell size or pigment loss | An old X-to-k calibration can shift |
| Comparable directional measurement | Detector collects different scattered fractions | k also depends on measurement geometry |
| Stable front illumination | Source drift or changing wall transmission | Profiles from different times need a reference |
| Mixing gives every cell the same history | Different circulation paths and residence times | A spatial mean cannot replace exposure distributions |
The original study by Zhao and colleagues is useful for time-dependent optical properties. It investigates specified cultures and measurement conditions; its parameters are not transferred to other algae. For moving cells, cell lifelines add exposure histories to a spatial light map. Geometrically similar equipment need not provide equivalent optical or temporal conditions.
Estimate background and biomass contributions separately
The synthetic relationship k = 5 + 60X contains a background of 5 m⁻¹ and a biomass contribution, with X entered in g/L. Since 1 g/L equals 1 kg/m³, the slope in this expression is 60 m²/kg. Using X in g/m³ instead requires a slope of 0.060 m²/g. A numerical slope is therefore fully specified only together with its concentration unit.
The supplied profiles include X = 0 to identify the prescribed background. Forcing a fit through the origin would assign part of that background to biomass at positive concentrations. The fitted coefficient might look adequate over a narrow investigated region but would have a different interpretation and systematically differ particularly at low concentrations.
An actual blank must match the optical arrangement. Water, complete medium without biomass and a used reactor after cultivation are different reference states. A wall deposit may change the entry value without producing the same spatial profile as an absorber distributed throughout the medium. One logarithmic straight line does not automatically distinguish those causes.
The original study by Zhao and colleagues on time-dependent radiative properties demonstrates changing absorption and scattering properties in its investigated cultures. No universal coefficient is imported here. The practical consequence for this independent analysis is to re-examine an optical relationship fitted at one culture state before using it at another.
A logarithmic fit changes the error model
From I(z) = I₀ exp(−kz), ln[I(z)/I₀] = −kz when both intensities are positive. Linear regression after taking logarithms is convenient, but it treats errors differently from fitting on the original signal scale. With constant absolute sensor error, relative error becomes large in dark regions; logarithmic transformation can make those points influential.
An original two-point example uses I₀ = 300 and I(0.05 m) = 30 µmol m⁻² s⁻¹, giving k = ln(10)/0.05 = 46.0517 m⁻¹. Increasing the second reading by 1 to 31 lowers k to about 45.396 m⁻¹. With a darker endpoint of 3 and the same absolute increase to 4, k changes from 92.103 to approximately 86.350 m⁻¹. Identical absolute measurement discrepancy therefore affects coefficient estimation much more strongly in the darker case.
Check dark offset before taking logarithms. Negative or nearly zero corrected signals cannot meaningfully be made “logarithmable” by replacing them with an arbitrary tiny positive number. Such values require treatment consistent with detection limits and the measurement model. Simply deleting all dark points can instead restrict the analysis to a different depth region.
An auditable fit specifies included depths, exclusions with reasons, treatment of I₀ and the error model. The synthetic CSV profiles are generated ideally and contain no actual sensor variability. Recovering their prescribed coefficient almost exactly checks arithmetic, not equivalent identifiability in a real multiply scattering culture.
Compare two-sided illumination at equal photon input
An independent model variant distributes the same combined input of 300 µmol m⁻² s⁻¹ across two opposite panel faces, each receiving 150. Keeping the linear effective-attenuation model and excluding additional reflection gives I(z) = 150 exp(−kz) + 150 exp[−k(L − z)]. Here I(z) denotes the sum of the two opposing directional contributions, not net passage through a plane or the reading of one single-sided PPFD sensor. With k = 65 m⁻¹ and L = 0.05 m, optical depth is 3.25.
At the mid-plane, summed intensity is approximately 59.074 µmol m⁻² s⁻¹. At either wall it is about 155.816. The original one-sided case instead has 300 at the front and only 11.632 at the back. Illumination becomes more uniform in this variant while combined entering flux remains unchanged. The spatial mean also remains 88.729 µmol m⁻² s⁻¹ under these assumptions because the two half-sized exponential contributions add linearly.
This does not establish identical biomass production. A non-linear biological response can react to redistribution. Actual two-sided equipment may also need additional lamps, mounting area, electrical drivers or changed spacing. Equal photon input therefore guarantees neither equal electricity demand nor equal investment.
The variant exposes an unfair comparison: supplying the full original intensity to both sides doubles total incoming flux. Any resulting growth increase would not by itself establish a benefit of spatial distribution. Define the comparison question and held quantity before experimenting.
Why mean light does not determine mean biological response
Take an explicitly abstract dimensionless saturation response r(I) = I/(100 + I) in two equal regions with I = 50 and I = 550. Mean light is 300, but mean local response is [50/150 + 550/650]/2 = 0.58974, whereas response at the mean is 300/400 = 0.75. Non-linearity alone creates the difference. These numbers describe no particular alga or actual growth rate.
This tests a common modelling step: inserting spatially averaged intensity into a non-linear growth function and calling the result mean growth. That is generally an additional approximation. A spatially resolved approach evaluates local response before weighting it over volume or cell residence. Dynamic adaptation may make even this quasi-steady averaging insufficient.
Residence weighting must match the question. Uniform volume distribution does not imply every individual cell spends exactly the same fraction of a short observation interval in each zone. Circulation, poorly exchanged regions and preferred flow paths can change individual exposure histories. The chapter on cellular exposure histories explains why flow and illumination should be interpreted together.
The original work of Bernard and Lu investigates optimisation conditions within an explicit optical growth model. Its mathematical conclusions are conditional on that model. The compendium therefore distinguishes a calculated optical threshold region, a measured biological compensation condition and an actually confirmed productive operating region.
Working case: the same outlet value can hide different internal profiles
Consider two hypothetical transparent layers, each 0.025 m thick. In the first arrangement, effective attenuation is initially 100 m⁻¹ and then 30 m⁻¹. Reverse the layers in the second arrangement. Reflection, scattering between directions and light-induced property changes are excluded from this purely exponential exercise.
Total optical depth is 100 × 0.025 + 30 × 0.025 = 3.25 in both cases. With I₀ = 300, both consequently have an outlet of 11.632 µmol m⁻² s⁻¹. Inlet-and-outlet measurements alone cannot distinguish the arrangements within this model. Each would yield an effective coefficient of 65 m⁻¹ across the complete path.
Spatial means nevertheless differ. Placing the stronger attenuating layer first gives approximately 63.737 µmol m⁻² s⁻¹ after piecewise integration; reversing the order gives 131.542. Directed flux lost early is unavailable across more of the path. Commuting exponential factors for the endpoint does not make internal profiles identical.
This deliberately simple case explains a two-point measurement's limits. One outlet value can be compatible with several internal distributions. Actual cultures may additionally vary through spatial biomass concentration, local deposits or other optical components. The layer model identifies none of these causes; it demonstrates why internal measurement positions or independent information can be valuable.
To recalculate, integrate each layer separately. The second layer's entry value is the already attenuated exit of the first. Add both profile integrals and divide by total depth of 0.05 m. Averaging two local means without accounting for layer thickness would be wrong when thicknesses differ. The same weighting principle applies to finer spatial discretisation.
Check limiting cases before handing over an optical model
Limiting cases often test a model more effectively than a long table of plausible numbers. For I(z) = I₀ exp(−kz), k = 0 must give I = I₀ everywhere. At very small optical depth τ = kL, the correctly normalised depth mean I₀(1 − exp(−τ))/τ also approaches I₀. The apparent zero divided by zero at the exact origin therefore has a well-defined limit.
The calculation uses a numerically stable expression for small τ. The expm1 function evaluates exp(x) − 1 with less cancellation loss for small differences. The mean can be written −I₀ expm1(−τ)/τ, assigning I₀ directly at exact τ = 0. This changes no physics; it prevents an avoidable numerical error.
At large positive τ, outlet intensity approaches zero. The depth mean stays positive and decreases approximately as I₀/τ in this simple model. Negative mean intensity or increasing outlet intensity with larger k clearly indicates a sign or unit issue under these assumptions. Negative fitted k may instead signal noise or model inadequacy and should not automatically be interpreted as physical light amplification.
Geometric limits help too. Halving L at constant k halves τ. Doubling k simultaneously preserves τ and the dimensionless profile over z/L. Actual biomass mass, volume and illuminated area can nevertheless differ. Equal optical depth consequently provides limited optical similarity, not complete process similarity.
Handover therefore separates equations with units, actual parameter evidence and limiting-case results. An appropriate equation without an appropriate parameter source is incomplete. A good fit with incorrect limits can instead reveal numerical or structural errors. Both are needed for a useful, clearly bounded decision model.
From a profile to the next defensible conclusion
The principal conclusion is that incident light and biomass concentration determine an internal profile only when the optical model is justified for that combination. Even a correct profile does not establish biomass productivity. Biological response, material balance and time basis are also required. The next chapter separates biomass formation from useful harvest.
The downloadable file contains a blank and four biomass concentrations at seven depths. First recover k from each positive depth reading, then integrate analytically and compare with sufficiently fine numerical integration. Next change only L. This reveals which results follow from geometry and which depend on additional biological assumptions.
Terms in this chapter
24 terms explained in the glossary
- Attenuation coefficient
- Biomass
- Calibration
- CSV file
- Density
- Equipment utilisation
- Experimental factor
- Exposure
- Growth phases
- Harvest
- Light scattering
- Linear regression
- Material balance
- Measurement model
- Membrane flux
- Optical depth
- Parameter identifiability
- Photoacclimation
- Photon
- Photon flux density
- Pigment
- Response variable
- Reynolds number
- Statistical residual