6.4.3 Predicting all traits

Reflectance spectra were also used to predict Vcmax25 and Vcmax25/Narea. The genotypic range

for each character is compared against data collected from the Mex experiments elite wheat genotypes from CIMCOG Subset II before anthesis (CB_Mex) and after anthesis (CA_Mex), see Chapters 3 and 5. The L_Mex2 set of wheat genotypes showed the biggest variation in all traits. For Vcmax25, J and Vcmax25/Narea, the Mex experiments showed less

143 LMA and different ranges for Narea and SPAD. Also shown are the ranges for the selected

subsets LS_Mex2 and CCS_Mex2. Mean values for Vcmax25, J and Vcmax25/Narea all increased

while the range narrowed for LS compared to L. By contrast, mean values for Vcmax25, J and Narea all decreased for CCS compared to CC.

J increased in the second measurement, LS_Mex2 (Figure 6.7), and the predictions reveal

that for most of these genotypes, Vcmax25, J and Vcmax25/Narea also increased, while LMA, Narea

and SPAD remained similar in both measurements (Figure 6.10).

Elite genotypes from LS_Mex2 showed a decrease in J (Figure 6.7.b), which agrees with predictions from CCS_Mex2. These elite genotypes reduced Vcmax25 and J in the second measurement, and interestingly Vcmax25/Narea remained similar between occasions. Narea was

also lower in the second measurement, but LMA and SPAD increased (Figure 6.10).

Figure 6.10 Ranges of Vcmax25, J, Vcmax25/Narea, LMA, Narea and SPAD predicted from

reflectance measurements and the models generated in Chapter 5 for 6 different groups of wheat genotypes (CB, CA, L, LS, CC, CCS).


This chapter demonstrates that hyperspectral reflectance can be used in the field to rapidly screen for photosynthetic, leaf structure and composition traits. Importantly, the models generated in Chapter 5 were able to predict traits for novel genotypes that were not used in their construction. This screening method allowed us to rapidly select genotypes in the


same season in which they were measured, while biochemical analysis and gas exchange can take much longer and require many more resources to produce similar results.

6.5.1 Predicting traits for novel wheat genotypes that were not used for model derivation

Models derived from a combination of several sets of wheat genotypes were tested on genotypes which had not been used to develop the models in Chapter 5. Results revealed that it is possible to screen for Vcmax25, J, Vcmax25/Narea, LMA, Narea and SPAD using

reflectance. The best predictions were obtained for LMA and Narea. Interestingly, models

derived from aspen and cotton were able to predict leaf nitrogen concentration and LMA from reflectance measured on soybean (Ainsworth et al., 2014). It would be useful to compare the models derived here for wheat with those derived from aspen and cotton. It is still difficult to define the size and composition of the germplasm training set required to build a robust model with 2,000 wavelengths while balancing good prediction against ‘over fitting’. In this project with wheat, 282 measurements were used to build the model to predict LMA, Narea and SPAD, and 262 measurements to predict Vcmax25, J, Vcmax25/Narea.

These models performed well at predicting traits in 223 novel elite wheat genotypes and 235 novel wheat landraces. In another experiment with wheat, Ecarnot et al., (2013) used reflectance to predict LMA and Narea, using a calibration obtained from a diverse collection

of wheats measured under multiple conditions and environments (Ecarnot et al., 2013). By contrast, it seems that the calibration for aspen required fewer observations (Serbin et al., 2012). However, in this study a strong driver of variation was environmental treatment rather than genetic variation. Further analysis comparing different sizes of training sets to construct the models is required.

In the second measurement, gas exchange was used to validate predictions of J. The correlation between observed J and predicted J in Mex2 was relatively poor with R2=0.4-

0.5 (Figure 6.4 and 6.8). During the second measurement, mean leaf temperature was 32 ºC and many plants showed low gs (average of 0.23 mol H2O m-2 s-1) suggesting that plants

were stressed on the day of measurements. Both of these factors could influence

calculations of J with gas exchange. There is likely to be genetic variation for gm between

elite wheats and wheat landraces as found by Jahan et al., 2014 or between leaves with different photosynthetic capacity (von Caemmerer and Evans, 1991). We assumed a constant gm of 0.55 mol m-2 s-1 bar-1 for all wheat genotypes in these surveys, but because

the measurements were made at high ambient CO2 concentrations (800 ppm), the error

145 from chlorophyll fluorescence, the need to surround the leaf with a high CO2

concentration would mean that each measurement was more complicated and time consuming. Wheat landraces are a source of diversity that needs to be explored more intensively in the future.

At present we are satisfied with the calibration of the models, which provide adequate estimates for six different traits from a single hyperspectral reflectance measurement. Other instruments target only one trait such as SPAD for chlorophylls (Konica Minolta, 2009- 2013) or FluorPen to estimate electron transport rate from chlorophyll fluorescence. However, choosing the best method to screen genotypes for photosynthetic traits will depend on the objectives of the experiment, and whether the data is used simply to rank genotypes or provide more precise quantitative data.

In document Screening genetic variation for photosynthetic capacity and efficiency in wheat (Page 142-145)