Table 1.
Hyperspectral vegetation indices that were compared to estimated Vcmax and Jmax values.
Fig 1.
Temporal variability in meteorological and physiological conditions throughout the study period.
A) shows mean daytime temperature in °C (orange line; 06:00 to 18:00), peak VPD in kPa (purple line; 10:00 to 14:00), precipitation in mm and supplemental watering throughout the study period with dark blue bars representing precipitation and light blue representing supplemental watering days. B) Shows the time series of predawn water potential (MPa) throughout the study period. C) and D) show the time series of estimated Vcmax (μmol m-2 s-1) and Jmax (μmol m-2 s-1), respectively.
Fig 2.
Leaf reflectance spectra and wavelength importance in PLSR models.
(A) Pooled reflectance spectra (λ = 450–2500 nm) for all plants studied throughout the course of the study period (5/24/2016 to 7/05/2017). The mean reflectance at each wavelength is shown by the red line, the maximum and minimum reflectance at each wavelength are shown by the blue and green lines, respectively. (B)&(C): selectivity ratio of each wavelength in the PLSR models tested on variable water potential. (B) Shows the selectivity ratio of each wavelength in the Vcmax PLSR model, and (C) shows the selectivity ratio of each wavelength in the Jmax PLSR model. The red lines in B&C are LOESS smoothers for visualization purposes.
Fig 3.
Temporal stability of PLSR models.
All figures show observed vs. predicted values from PLS regression models. A&B: leave-one-out cross-validation procedure for (a) maximum rates of RuBP carboxylation (Vcmax, μmol m-2 s-1) and (b) RuBP regeneration Jmax (μmol m-2 s-1), n = 86. C&D: representative 80/20% split of the data for PLSR (C) Vcmax (μmol m-2 s-1) and (D) Jmax (μmol m-2 s-1). Each model was trained on 69 observations and tested on 17 observations. E & F: models tested on the period with the greatest between-individual variation in pre-dawn water potential (Ψpd) for (E) Vcmax (μmol m-2 s-1) and (F) Jmax (μmol m-2 s-1). Each model was trained on 74 observations and tested on the 12 observations on 6/23/2016 and 6/24/2016 when the population varied widely in Ψpd. Statistical tests of increase in rigor moving downward.
Fig 4.
Comparison of predictive abilities of hyperspectral indices and PLSR models.
Predictive capability for (A) maximum rates of RuBP carboxylation (Vcmax, μmol m-2 s-1) and (B) RuBP regeneration Jmax (μmol m-2 s-1). Pairwise correlations between Vcmax and Jmax estimated using standard gas exchange techniques and spectral indices were tested and R2 is reported. The R2 reported for the PLSR model is from the leave-one-out cross-validation approach. Indices with non-significant correlations are represented with “ns”.