Temperature is one of the most important factors determining plant growth, development, and yield. Accurate data on crop responses to temperature are essential for predicting the potential impacts of a future warmer climate on crop production.
High-throughput field phenotyping represents the capacity to non-destructively and remotely-sense crop growth in a high-throughput fashion to accurately characterize hundreds of genotypes under field conditions. The resulting vast quantities of repeated, objective observations data can be used to quantify genotype-specific growth response to temperature.
Yet, the ability to determine the adequacy of genotype-specific response models from field-derived data is difficult, according to Dr. Lukas Roth from the Institute of Agricultural Sciences at ETH Zurich. “Field-based measurements are notoriously ‘noisy’ due to environmental and soil inhomogeneities and measurement errors. Therefore, we never know the ‘truth’ for field data and are therefore unable to judge if our model correctly predicts growth response to temperatures.”
Therefore, in a new paper published in in silico plants, Roth and colleagues used a model to generate data with seasonal field-condition distributions of temperatures. They then were able to verify the accuracy of a existing linear model approach compared to a proposed new asymptotic model to extract genotype-specific growth response to temperatures.
First, plant height data was generated for several wheat genotypes using a simulation based on the Wang-Engel response function. The Wang-Engel model simulates crop development based on the non-linear response of plant development to temperature. The genotypes characterized different growth responses to cardinal temperatures and maximum absolute growth rate. Seasonal canopy growth was simulated based on five years of temperature data. The authors simulated measurement intervals of 3, 7, and 14 days to determine what data collection interval was sufficient to quantify genotype-specific growth response to temperature.
Next, they compared the ability of the existing linear model compared to a new asymptotic model to predict temperature-response parameters from the simulated data.
“The linear approach is widely used in our field of research and promises great robustness, but the more accurate our phenotyping data became, the more often we saw evidence of a non-linear relationship,” Roth explains.
The authors found that the asymptotic model extracted the base temperature of growth and the maximum absolute growth rate with high precision, whereas the simpler linear model failed to do so. Additionally, the asymptotic model provided a proxy estimate for the optimum temperature. However, when including seasonally changing cardinal temperatures as plants develop, the prediction accuracy of the asymptotic model was strongly reduced.
Regarding the high throughput sampling resolution, the authors found that measurement intervals of about four days were sufficient to reliably extract minimum cardinal temperature and maximum absolute growth rate. This is good news for those times when regular measurements are not possible such as under poor weather conditions and long weekends.
READ THE ARTICLE:
Lukas Roth, Hans-Peter Piepho, Andreas Hund, Phenomics data processing: Extracting dose-response curve parameters from high-resolution temperature courses and repeated field-based wheat height measurements, in silico Plants, 2022; diac007, https://doi.org/10.1093/insilicoplants/diac007
Data and source code that support the findings of this study are openly available in the ETH GitLab repository at https://gitlab.ethz.ch/crop_phenotyping/htfp_data_processing.