Skip to content

Water-stress steady-state fluorescence (): Flexas et al. (2002) benchmark ​

Open the dedicated interactive Figure 2 & SIF formulation explorer, run the Julia benchmark, or view the summary panels in the interactive offline gallery. Model assembly lives in setup/flexas_2002.jl, and every grapevine parameter, unit, and subpixel-digitized Figure 2 observation (55 main-panel points across all six curves plus 6 Figure 2 Inset points) is isolated in flexas_2002.toml.

Flexas et al. (2002, Physiologia Plantarum 114:231–240, doi:10.1034/j.1399-3054.2002.1140209.x) measured simultaneous gas exchange (LI-6400) and steady-state chlorophyll fluorescence (Walz PAM-2000,   long-pass detector) on Vitis vinifera L. cv. Tempranillo at (298.15 K) and relative humidity across six progressive drought stages indexed by maximum daily stomatal conductance to water vapor,   . Normalizing steady-state fluorescence by dark-adapted basal fluorescence eliminates leaf-to-leaf variation in chlorophyll content and sensor geometry, exposing three diagnostic features in Figure 2:

  1. Peak suppression: maximum declines from 1.448 in well-watered vines ( ) to 1.160 under severe drought ( ).

  2. Leftward irradiance shift: in the digitized curves the photosynthetic photon flux density (PPFD) at which peaks shifts from 729 ( ) and 735 ( ) down to 331 ( ) and 108 μmol photons m⁻² s⁻¹ ( ).

  3. Sub- high-light quenching & curvilinear conductance inset: under severe drought ( ), non-photochemical quenching pulls high-irradiance below dark ( ), producing a positive curvilinear vs. relationship (0.900 → 1.449) at    (Figure 2 Inset).

Evaluating every steady-state fluorescence formulation in LeafPhotosynthesis.jl against these experiments separates structural mathematical constraints from parameter calibration.


1. Why single-variable empirical fits fail on Figure 2 ​

In RateCompetitionFluorescence paired with SaturationQuenching (van der Tol et al. 2014, JGR Biogeosciences 119:2312–2327, Eq. 19), energy-dependent non-photochemical quenching is parameterized solely as a Hill function of relative PSII light saturation    , where     is dark-adapted maximum PSII photochemical yield:

Because is a single-variable function with no independent dependence on plastoquinone redox state or stromal sink capacity, any irradiance sweep from darkness ( ) to strong Rubisco limitation (  near ) maps PPFD through the exact same scalar curve :

  • Peak-height invariance across drought: changing or rescales the horizontal coordinate (shifting where occurs along the PPFD axis), but reaches the identical global peak   at every conductance—1.461 (at  ,    for  ) for the Cotton parameter set ( ,  ,  ) and 1.451 (at  ,    for  ) for the Flexas Drought parameter set ( ,  ,  ). Single-variable models therefore cannot reproduce the progressive 1.448 → 1.160 peak suppression in Figure 2.

  • High-light sign inversion in Cotton : because Cotton   saturates near   by  , further increases in increase the numerator    faster than the denominator, turning upward at high light and predicting higher under severe drought than under well-watered conditions (1.22 stomatal-only / 1.28 combined at   vs. 1.18 at   at  , and 1.35 vs. 1.31 at  —opposite in sign to the Figure 2 Inset). Switching to the empirical Drought fit (  , which van der Tol et al. 2014 fitted directly to Flexas et al. 2002) restores sub- quenching at high irradiance (0.90 at  ), yet still leaves peak height fixed near 1.45 across all conductances.


2. Two-state Cytochrome mechanics and Walz PAM-2000 PSI leakage ​

CytochromePhotosynthesis + CytochromeFluorescence (Johnson & Berry 2021, Photosynthesis Research 148:101–136) resolves both limitations by separating two distinct physical state variables:

  1. Thylakoid electron-supply poise : open PSII trap fraction    is governed by absorbed PSI excitation and Cytochrome plastoquinol oxidation capacity , independent of stromal . At low irradiance, carbon assimilation is light-limited ( ), dynamic feedback quenching remains near zero, and progressive reduction of the plastoquinone pool ( decreasing with PPFD) drives the initial rise in .

  2. Stromal sink saturation : once absorbed light exceeds the Rubisco + photorespiratory ceiling , realized linear electron yield   falls below the unquenched connected-units envelope . Closing the PSII exciton budget induces lumen-pH feedback quenching  , truncating the -driven fluorescence rise right at the   transition. Lowering draws down chloroplast , lowering so the   crossover occurs at lower PPFD where is still largely open—simultaneously lowering peak , shifting peak PPFD leftward, and driving high-light  .

Apparent PAM fluorometry (PAMFluorescenceObserver) ​

Flexas et al. (2002) recorded with a Walz PAM-2000 (>700 nm long-pass filter). The model weights far-red PSI emission twice as strongly as PSII emission ( , observation model of Johnson & Berry 2021, Eqs. 38–42). This weighting is carried over from Johnson & Berry (2021), who fitted 2.0 (interquartile range 1.6–2.5) for a LI-6800 on Populus fremontii; it is not a calibration of the PAM-2000:

Because oxidized quenches closed PSI reaction centers at the same rate as open photochemistry (   ), PSI contributes a redox-invariant baseline ( of dark ). With Johnson & Berry's Table 1 decay rates and   used unchanged (neither was fitted to Flexas et al. 2002), dark_pam_levels predicts dark-adapted apparent  , inside the 0.80–0.83 measured by Flexas et al. (2002, p. 234). This agreement depends on carrying the LI-6800 / P. fremontii estimate of the detector weighting over to the PAM-2000 and grapevine. Pure PSII emission ( ) has   and over-predicts well-watered (model 1.66 at  ; observed 1.44).


3. Photorespiratory electron-sink floor and non-stomatal drought transition ​

Separating pure stomatal diffusion (enable_nonstomatal = false, fixed and ) from coordinated stomatal + biochemical downregulation (enable_nonstomatal = true) in the Figure 2 Inset (  ) reveals a fundamental C3 stoichiometric boundary. In ambient air (  ,   ), even at the photorespiratory compensation point ( ), gross Rubisco carboxylation and photorespiratory release    cancel in net carbon uptake while both carboxylation ( ) and photorespiration + PSI cyclic flow ( , giving    for  ,  ) continue consuming electrons:

  • Mild-to-moderate drought (  ): pure stomatal drawdown quantitatively tracks both Figure 2 (RMSE = 0.049–0.061) and the Figure 2 Inset (1.41 → 1.39 → 1.35 stomatal-only and 1.41 → 1.36 → 1.23 combined vs. observed 1.449 → 1.369 → 1.229).

  • Severe drought (  ): pure stomatal closure hits the photorespiratory sink floor (   at   for  , and   as   where day respiration   holds    ), preventing the model's Figure 2 peak from shifting below ~350–400 μmol m⁻² s⁻¹ or from dropping below ~1.08. For the model to move the severe-drought peaks down to 200 μmol m⁻² s⁻¹ ( ; digitized: 108) and to reach sub- high-light   (0.73–0.88 at  ), it needs a non-stomatal biochemical downregulation. Such downregulation is documented in Vitis vinifera when prolonged drought reduces below 80–100 mmol m⁻² s⁻¹ (Escalona et al. 1999; Flexas et al. 2002, Fig. 3A & p. 239). The model represents it by a smooth Hill transition retaining 60% and 85% as  ; its half-conductance, exponent and floors were chosen by the author, were not fitted to the digitized points and have no Tempranillo-specific source in this repository (see inputs chosen, not fitted).


4. Physiological controls: high- reversion, qT state transitions, and DTT {#4.-Physiological-controls:-high-\mathrm{CO_2}-reversion,-qT-state-transitions,-and-DTT} ​

The fourth benchmark panel tests three experimental perturbations at severe water stress (  ):

  1. High- / -depleted reversion (Flexas et al. 2002, Fig. 3C): switching ambient air (  ,   ) to    (1500 ppm) in balance (  ) bypasses stomatal diffusion limitation, restores carboxylation (   at   and   at  , matching Fig. 3A's ~12.5), relieves feedback , and restores to 1.45–1.52 across high irradiance—confirming that the severe-drought   depression is driven primarily by stromal electron-sink restriction rather than irreversible PSII photoinhibition.

  2. Closed-form LHCII state transitions (StateTransitionRule, qT): in a static connected antenna (  ), exciton sharing boosts PSII efficiency as traps close and induces small positive feedback   even in the Cytochrome-limited low-light regime (Case 9), damping the low-light rise (1.30). Allowing StateTransitionRule (Case 10) to migrate LHCII from PSII to PSI (  ) maintains   up to Rubisco saturation while increasing PSI emission (weighted 2× by the assumed detector weighting), raising the model's low-light apparent peak to 1.76 before decaying under high-light feedback (1.18 at  ). Flexas et al. (2002) observed a damped peak (1.219) at  , not a spike like the model's 1.76. This misfit suggests the hypothesis that water-stressed grapevine leaves operate closer to static antenna partitioning than to full   state-transition rebalancing; it is not a finding, since other model inputs (for example , the detector weighting or ) could also cause the misfit.

  3. DTT xanthophyll-cycle inhibition (Flexas et al. 2002, Fig. 5): evaluating RateCompetitionFluorescence with ConstantQuenching(0) ( ) removes energy-dependent thermal dissipation, causing to climb monotonically past 2.3–4.5 as traps close—matching dithiothreitol-infiltrated leaves in Figure 5.


Connectivity , the PSI weight and the dark level ​

Two quantities set the size of the modeled signal and have no direct calibration for these leaves or this instrument: the exciton transfer rate between PSII units, , and the PSI detector weight . Johnson & Berry (2021) estimated both, together with five other parameters, from one Populus fremontii leaf; the benchmark carries their values (2 ns⁻¹ and 2.0) over. This section is a sensitivity analysis with every other input of the benchmark held fixed: the drought schedule, , , mesophyll conductance, the static antenna (no state transitions), zero sustained quenching and the Table 1 rate constants.

For fixed rate constants, zero sustained quenching and a static antenna, the PSII fluorescence relative to depends on three numbers only: the photochemical yield , the open fraction and . Gas exchange and electron transport fix the first two without reference to , which then decides how much of the closure of reaction centers appears as fluorescence and how much the model must assign to . Where light limits and excitation is balanced,   , and the two limits are

with   for the Table 1 rate constants: separate units need no quenching and show the full rise, a lake shows none.

Size of the signal. For the well-watered leaf ( ) the peak PSII is 2.53, 1.72 and 1.22 for  , 2 and 10 ns⁻¹.

What the assumed observer predicts. These are PSII yields. A fluorometer adds the PSI level to and to ; with the benchmark's observer that level is constant and equal to   times the dark PSII level, so a PSII ratio is reported as   , with     . The peaks above become 1.94, 1.44 and 1.13; only these compare with the measured 1.45. The constant level holds for this observer: PSI yield independent of P700 state ( ), static antenna fractions, fixed optics, and no PSI fluorescence dynamics from acceptor-side closure. A PSI level that differs between the dark reference and the light (state transitions) would also move the crossing of .

Below . Under the same assumptions and whatever , the PSII yield is below its dark level exactly when      : a low realized photochemical yield relative to the absorbed excitation, with strong quenching. Low net assimilation alone does not suffice, because photorespiration keeps consuming electrons. Well watered, the curves stay above up to PPFD 2100 for all three transfer rates. Under severe drought ( ) they are below it from PPFD 450, 375 and 250 on (first sampled points of the 25-unit grid) and end at 0.58, 0.55 and 0.49 (0.75, 0.73 and 0.68 as the observer predicts them: a constant PSI level compresses the signal around but leaves the crossing where it is). No drought parameterization is needed for this: in our reproduction of the Case 9 simulation of Johnson & Berry (their Fig. 8c; chloroplast CO₂ prescribed at 20 Pa, static antenna) the yield at    peaks at 1.53 , is below from PAR 1640 on and ends at 0.88 at PAR 2400; at 10 ns⁻¹ it is below from PAR 940 on and ends at 0.72. These crossings are our calculations, not values stated in the paper. Extending the same simulation to other prescribed CO₂ levels, it stays above at 40 Pa and ends at 0.59 at 10 Pa (tested).

A trade-off, bounded by dark . RMSE of over the 49 lit points of Figure 2 on a grid of the two quantities:

(ns⁻¹)weight 02.02.5smallest on the grid 0–6 (weight)
0.50.5880.2870.2460.088 (6.0)
10.3600.1510.1230.047 (6.0)
1.50.2470.0880.0700.049 (4.0)
20.1800.0620.0550.054 (3.0)
30.1150.0700.0750.068 (1.5)
50.1070.1130.1180.102 (0.5)
100.1670.1640.1650.163 (1.0)

Judged by the 49 lit points alone, a lower transfer rate with a larger PSI weight does as well as the carried-over pair: the smallest error per transfer rate lies between 0.047 and 0.068 from   to 3 ns⁻¹. No error model for the digitized points is available, so these differences are not shown to be significant either way.

The PSI weight is constrained by a second measurement, however. With the PSI level relative to PSII in the dark, the dark-adapted apparent yield is     for the PSII : 0.882 without PSI, 0.821 at weight 2.0, 0.807 at 2.5, 0.794 at 3.0, 0.768 at 4.0 and 0.721 at 6.0. Flexas et al. (2002, p. 234) report 0.80–0.83 for all leaves examined, which admits weights of about 1.7 to 2.8 (PSI share of dark of 35–47 %) and excludes the low- end of the valley. At the two grid weights inside that range (2.0 and 2.5),    has the smallest error (0.062, 0.055), 1.5 and 3 follow (0.070–0.088), and 1 and 5 are clearly worse (0.11–0.15).

Both constraints are conditional on the fixed inputs listed above, which shape , and the dark reference, and on the PSII rate constants: the bound uses  , and a different would move it. The grid varies the detector weight and keeps at the actinic dark-balanced values ( ), although the measuring beam of a fluorometer has its own absorbed fractions. Under the observer's assumptions these enter the normalized signal only through the product , which is 0.94 times the weight tabulated here: the range above corresponds to a product of about 1.6 to 2.6. Within those conditions the benchmark supports that range of the product and between about 1.5 and 3 ns⁻¹; it is not a calibration of either. A direct constraint on connectivity would need controlled fast-induction measurements interpreted with an explicit connectivity and kinetic model.


5. Summary across SIF / fluorescence yield formulations ​

Inputs chosen, not fitted ​

No parameter was fitted to the digitized fluorescence points. The following model inputs were chosen by the author and have no Tempranillo-specific source in this repository (flexas_2002.toml):

  • the non-stomatal Hill schedule [nonstomatal_drought]: half_conductance_mmol = 56.0 (mmol H₂O m⁻² s⁻¹), hill_exponent = 4.0, vcmax_severe_floor = 0.60 and vqmax_severe_floor = 0.85;

  •    (site_density = 0.58 μmol sites m⁻² at   );

  •    (mesophyll_co2 = 0.20);

  •    at 298.15 K and 21 kPa O₂ (gamma_star_ref_Pa = 3.70).

The schedule matters: with stomatal diffusion alone the lit Figure 2 RMSE is 0.097 and the Inset RMSE 0.144; with the schedule they are 0.062 and 0.038 (the two Cytochrome + PAM rows below). The detector weighting   is carried over from Johnson & Berry (2021) (LI-6800, P. fremontii), and their Table 1 rate constants are used unchanged.

PSII yield and detector signal ​

A PSII fluorescence yield is not what a fluorometer reports. The assumed observer adds a PSI level that is constant here, so its share of each level changes as PSII is quenched: 39 % of dark and 7 % of dark ; among the 49 lit digitized points, 27 % (well watered, at the peak) to 51 % (severe drought, high light) of and up to 48 % of . The indices built from these levels are apparent quantities, biased in known directions: with a PSII level and the PSI level,     and     are both too low, while comes out too high. At   and PPFD 1222 the model has 0.093 against an apparent 0.048, NPQ 11.5 against 6.0, and 0.17 against 0.21. The PSII yields themselves are unchanged.

Each formulation is scored twice against the 49 lit points: as the PSII ratio, and through the observer (detector_level, with the same weights, measuring-beam fractions and PSI level for all):

FormulationRMSE, PSII ratioRMSE, with PSI levelwith PSI,  with PSI,  
CytochromeFluorescence0.1800.0620.0490.092
Cotton + prescribed PSI0.1520.1320.2100.161
Drought + prescribed PSI0.1180.1180.2050.065

For the mechanistic model the observer is the proper comparison, and it changes the picture. For the empirical fits the second column is a sensitivity case, not a correction: van der Tol et al. (2014) estimated their constants from fluorometer signals in which PSI was not corrected out, so their yields already represent both photosystems, and the unaugmented column follows that calibration more closely. Adding the prescribed PSI level on top mixes the two interpretations: it lowers their dark from 0.80 to 0.71, below the measured 0.80–0.83, and pulls their well-watered peak (1.46) down to about 1.28.

The observer is an assumption throughout: weights carried over from another instrument and species, the measuring-beam fractions set equal to the actinic dark-balanced ones for every formulation, and fixed weights standing in for wavelength-dependent emission, reabsorption and detector response. Its 39 % PSI share of dark lies inside the range of published estimates for detection beyond 700 nm, which is wide: 24 % in Arabidopsis (Pfündel et al. 2013), about 30 % in C3 leaves (Pfündel 1998), 41 % computed for an RG9 long-pass filter from isolated spinach PSI and PSII spectra (Bos et al. 2023) and 45 % in cherry laurel (Pfündel 2021). It should not be read as a general correction for this instrument. The lake-model rows of the Gu et al. comparison pass through the same observer.

Peak positions ​

The digitized Figure 2 curves peak at PPFD = 729, 735, 516, 322, 331, 108 μmol m⁻² s⁻¹ for   (at   the points at 331 and 441 tie). The model (CytochromeFluorescence + PAMFluorescenceObserver, stomatal + non-stomatal) peaks at 1050, 675, 525, 475, 325, 200 μmol m⁻² s⁻¹ on its 25 μmol m⁻² s⁻¹ grid. Its well-watered peak lies ~320 μmol m⁻² s⁻¹ above the digitized one, a mismatch.

Formulations ​

Formulation in LeafPhotosynthesis.jlDark Lit Fig. 2 RMSE ( )Inset RMSE ( )Peak vs. High-light   under severe drought
ConstantQuenching(0) (DTT /  )0.800— (DTT control)—None (climbs to 4.47)No (matches DTT-infiltrated leaves only, Fig. 5)
SaturationQuenching — Cotton 0.8000.1520.224No (invariant 1.45–1.46)No (1.35 at 1400, inverted drought sign)
SaturationQuenching — Drought 0.8000.1180.157No (invariant 1.45)Yes (0.90 at 1400, empirical fit to Flexas)
CytochromeFluorescence (pure PSII,  )0.8820.1800.168Yes (1.72 → 1.36), over-amplifiedYes (0.60 at 1400)
CytochromeFluorescence + StateTransitionRule (qT) + PAM-20000.8210.4020.493Overshoots (2.30 → 1.55)Yes (0.91 at 1400)
CytochromeFluorescence + PAMFluorescenceObserver ( , stomatal only)0.8210.0970.144Partial (1.44 → 1.29, floor at )Floor near 0.95 (1.08 at 750)
CytochromeFluorescence + PAMFluorescenceObserver ( , stomatal + Escalona/Flexas non-stomatal)0.8210.0620.038Yes for the height (1.44 → 1.22 vs. obs. 1.45 → 1.16); mismatch in position at   (model 1050, digitized 729)Yes (0.75 at 1400, 0.82 at 750)

6. Scope limits ​

  • All simulations are steady-state point evaluations; slow dark-to-light induction and photoinhibitory recovery timescales (qI relaxation) are not modeled.

  • In CytochromeFluorescence, regulated thermal dissipation is solved algebraically to close the PSII exciton budget without an upper xanthophyll/PsbS saturation ceiling ; at extreme drought and high light ( ,   ), unbounded slightly over-quenches (0.75–0.77 vs. observed 0.91–0.94).

  • Pre-escape photosystem yields and PAM detector ratios () describe leaf-level fluorometry; canopy-leaving solar-induced chlorophyll fluorescence (SIF) additionally requires spectral PSII/PSI emission shapes, within-leaf chlorophyll reabsorption (685 nm vs. 740 nm), and canopy radiative transfer in CanopyOptics.jl.