TSWIFT: Tower Spectrometer on Wheels for Investigating Frequent Timeseries for high-throughput phenotyping of vegetation physiology

BACKGROUND

Remote sensing instruments enable high-throughput phenotyping of plant traits and stress resilience across scale. Spatial (handheld devices, towers, drones, airborne, and satellites) and temporal (continuous or intermittent) tradeoffs can enable or constrain plant science applications. Here, we describe the technical details of TSWIFT (Tower Spectrometer on Wheels for Investigating Frequent Timeseries), a mobile tower-based hyperspectral remote sensing system for continuous monitoring of spectral reflectance across visible-near infrared regions with the capacity to resolve solar-induced fluorescence (SIF).

RESULTS

We demonstrate potential applications for monitoring short-term (diurnal) and long-term (seasonal) variation of vegetation for high-throughput phenotyping applications. We deployed TSWIFT in a field experiment of 300 common bean genotypes in two treatments: control (irrigated) and drought (terminal drought). We evaluated the normalized difference vegetation index (NDVI), photochemical reflectance index (PRI), and SIF, as well as the coefficient of variation (CV) across the visible-near infrared spectral range (400 to 900 nm). NDVI tracked structural variation early in the growing season, following initial plant growth and development. PRI and SIF were more dynamic, exhibiting variation diurnally and seasonally, enabling quantification of genotypic variation in physiological response to drought conditions. Beyond vegetation indices, CV of hyperspectral reflectance showed the most variability across genotypes, treatment, and time in the visible and red-edge spectral regions.

CONCLUSIONS

TSWIFT enables continuous and automated monitoring of hyperspectral reflectance for assessing variation in plant structure and function at high spatial and temporal resolutions for high-throughput phenotyping. Mobile, tower-based systems like this can provide short- and long-term datasets to assess genotypic and/or management responses to the environment, and ultimately enable the spectral prediction of resource-use efficiency, stress resilience, productivity and yield.

Bayesian modelling of phosphorus content in wheat grain using hyperspectral reflectance data

BACKGROUND

As a result of the technological progress, the use of sensors for crop survey has substantially increased, generating valuable information for modelling agricultural data. Plant spectroscopy jointly with statistical modeling can potentially help to assess certain chemical components of interest present in plants, which may be laborious and expensive to obtain by direct measurements. In this research, the phosphorus content in wheat grain is modeled using reflectance information measured by a hyperspectral sensor at different wavelengths. A Bayesian procedure for selecting variables was used to identify the set of the most important spectral bands. Additionally, three different models were evaluated: the first model assumes that the observations are independent, the other two models assume that the observations are spatially correlated: one of the proposed models, assumes spatial dependence using a Conditionally Autoregressive Model (CAR), and the other through an exponential correlogram. The goodness of fit of the models was evaluated by means of the Deviance Information Criterion, and the predictive power is evaluated using cross validation.

RESULTS

We have found that CAR was the model that best fits and predicts the data. Additionally, the selection variable procedure in the CAR model reveals which wavelengths in the range of 500-690 nm are the most important. Comparing the vegetative indices with the CAR model, it was observed that the average correlation of the CAR model exceeded that of the vegetative indices by 23.26%, - 1.2% and 22.78% for the year 2010, 2011 and 2012 respectively; therefore, the use of the proposed methodology outperformed the vegetative indices in prediction.

CONCLUSIONS

The proposal to predict the phosphorus content in wheat grain using Bayesian approach, reflect with the results as a good alternative.

Wheat physiology predictor: predicting physiological traits in wheat from hyperspectral reflectance measurements using deep learning

BACKGROUND

The need for rapid in-field measurement of key traits contributing to yield over many thousands of genotypes is a major roadblock in crop breeding. Recently, leaf hyperspectral reflectance data has been used to train machine learning models using partial least squares regression (PLSR) to rapidly predict genetic variation in photosynthetic and leaf traits across wheat populations, among other species. However, the application of published PLSR spectral models is limited by a fixed spectral wavelength range as input and the requirement of separate custom-built models for each trait and wavelength range. In addition, the use of reflectance spectra from the short-wave infrared region requires expensive multiple detector spectrometers. The ability to train a model that can accommodate input from different spectral ranges would potentially make such models extensible to more affordable sensors. Here we compare the accuracy of prediction of PLSR with various deep learning approaches and an ensemble model, each trained and tested using previously published data sets.

RESULTS

We demonstrate that the accuracy of PLSR to predict photosynthetic and related leaf traits in wheat can be improved with deep learning-based and ensemble models without overfitting. Additionally, these models can be flexibly applied across spectral ranges without significantly compromising accuracy.

CONCLUSION

The method reported provides an improved prediction of wheat leaf and photosynthetic traits from leaf hyperspectral reflectance and do not require a full range, high cost leaf spectrometer. We provide a web service for deploying these algorithms to predict physiological traits in wheat from a variety of spectral data sets, with important implications for wheat yield prediction and crop breeding.

A best-practice guide to predicting plant traits from leaf-level hyperspectral data using partial least squares regression

Partial least squares regression (PLSR) modelling is a statistical technique for correlating datasets, and involves the fitting of a linear regression between two matrices. One application of PLSR enables leaf traits to be estimated from hyperspectral optical reflectance data, facilitating rapid, high-throughput, non-destructive plant phenotyping. This technique is of interest and importance in a wide range of contexts including crop breeding and ecosystem monitoring. The lack of a consensus in the literature on how to perform PLSR means that interpreting model results can be challenging, applying existing models to novel datasets can be impossible, and unknown or undisclosed assumptions can lead to incorrect or spurious predictions. We address this lack of consensus by proposing best practices for using PLSR to predict plant traits from leaf-level hyperspectral data, including a discussion of when PLSR is applicable, and recommendations for data collection. We provide a tutorial to demonstrate how to develop a PLSR model, in the form of an R script accompanying this manuscript. This practical guide will assist all those interpreting and using PLSR models to predict leaf traits from spectral data, and advocates for a unified approach to using PLSR for predicting traits from spectra in the plant sciences.

High-throughput field phenotyping using hyperspectral reflectance and partial least squares regression (PLSR) reveals genetic modifications to photosynthetic capacity

Spectroscopy is becoming an increasingly powerful tool to alleviate the challenges of traditional measurements of key plant traits at the leaf, canopy, and ecosystem scales. Spectroscopic methods often rely on statistical approaches to reduce data redundancy and enhance useful prediction of physiological traits. Given the mechanistic uncertainty of spectroscopic techniques, genetic modification of plant biochemical pathways may affect reflectance spectra causing predictive models to lose power. The objectives of this research were to assess over two separate years, whether a predictive model can represent natural and imposed variation in leaf photosynthetic potential for different crop cultivars and genetically modified plants, to assess the interannual capabilities of a partial least square regression (PLSR) model, and to determine whether leaf N is a dominant driver of photosynthesis in PLSR models. In 2016, a PLSR analysis of reflectance spectra coupled with gas exchange data was used to build predictive models for photosynthetic parameters including maximum carboxylation rate of Rubisco (V _c,max ), maximum electron transport rate (J _max ) and percentage leaf nitrogen ([N]). The model was developed for wild type and genetically modified plants that represent a wide range of photosynthetic capacities. Results show that hyperspectral reflectance accurately predicted V _c,max, J _max and [N] for all plants measured in 2016. Applying these PLSR models to plants grown in 2017 resulted in a strong predictive ability relative to gas exchange measurements for V _c,max, but not for J _max, and not for genotypes unique to 2017. Building a new model including data collected in 2017 resulted in more robust predictions, with R² increases of 17% for V _c,max . and 13% J _max . Plants generally have a positive correlation between leaf nitrogen and photosynthesis, however, tobacco with reduced Rubisco (SSuD) had significantly higher [N] despite much lower V _c,max. The PLSR model was able to accurately predict both lower V _c,max and higher leaf [N] for this genotype suggesting that the spectral based estimates of V _c,max and leaf nitrogen [N] are independent. These results suggest that the PLSR model can be applied across years, but only to genotypes used to build the model and that the actual mechanism measured with the PLSR technique is not directly related to leaf [N]. The success of the leaf-scale analysis suggests that similar approaches may be successful at the canopy and ecosystem scales but to use these methods across years and between genotypes at any scale, application of accurately populated physical based models based on radiative transfer principles may be required.