Enhancing photosynthesis for increased sorghum grain yield has become a key focus in sorghum breeding efforts. Phenotyping, involving the measurement of various morpho-physiological and physical traits associated with photosynthesis and grain yield, is a time-intensive process. However, the potential of non-invasive leaf-level hyperspectral imaging to swiftly detect plant performance, optimizing photosynthesis and grain yield, is promising. This study aimed to evaluate the feasibility of utilizing hyperspectral reflectance in the 350–950 nm range for the rapid estimation of these traits in intact sorghum leaves. Multiple machine learning regression algorithms were developed using leaf-level hyperspectral reflectance data from nearly 400 sorghum accessions within an association panel. The best-performing prediction models were then considered as potential methods for constructing a prediction model targeting multiple other physiological and yield traits in sorghum accessions. The results indicate that this approach enables the early detection of leaf photosynthetic and yield traits through leaf-level hyperspectral reflectance without the need for a full-range, high-cost leaf spectrometer.

Modern maize hybrids prolong the period that they photosynthesize and accumulate Nitrogen (N) out of the soil which has helped them produce more yield per unit of N fertilizer. However, the increase in post flowering activity is inversely correlated with N remobilization from the leaves. Further gains in N response could be achieved by breaking this association, but doing so requires an in-depth understanding of the temporal dynamics of maize canopy traits and plant N mobilization. Leaf nutrient samples were collected at five time points and remote sensing phenotypes were extracted from Unoccupied Aerial System (UAS) imagery (orthomosiacs and point clouds). Spectral indices and point-cloud based metrics were used to investigate the relationship between changes in N storage dynamics and yield among hybrids grown in low and high N treatments. From these combined phenotypes, it is possible to dissect how rate of growth and canopy health help to describe hybrid response N and also provide clues for how to break the negative relationship between yield and N remobilization.

Hyperspectral based prediction of nutrient content in maize leaves Hyperspectral imaging is a promising method to predict crop traits in a high-throughput manner and unlock quantitative genetic studies. A single hyperspectral image can be used to predict several unrelated traits at once using spectral data from 350nm - 2500nm. Researchers have successfully modeled different physiological traits in maize such as vegetative Nitrogen content but the effect of different development stages, genotypes, and treatments on modelling power remains unclear. Here, I explore the ability to model leaf macro- and micro- nutrient content and leaf water content from hyperspectral transmittance data collected with a LeafSpec imaging device. I will compare three different machine learning algorithms; Partial Least Squares Regression, Random Forest and a Convolutional Neural Net to model nutrient content collected from twenty hybrids throughout the 2020 field season in fertilized and not fertilized blocks. Genotypes and development stages excluded from model training are used to externally validate models. Sulfur, Nitrogen, Calcium, Copper, and Iron leaf concentrations were the most amenable nutrients to prediction with coefficient of determination scores from 0.78 - 0.73, respectively. Models trained on samples from a collection of time points were able to accurately predict new time points and genotypes. The findings demonstrate the ability to predict nutrient content in field grown maize over a variety of developmental stages, genotypes, and treatments from a handheld hyperspectral imaging device.

Hyperspectral imaging is a promising method to predict traits in a high-throughput manner with the potential to unlock quantitative genetic studies. Researchers have successfully modeled physiological traits such as vegetative Nitrogen content, but scope of methodology and lack of truly novel testing data hinder large scale trust in the process. Here, I explore the ability to model leaf Nitrogen content from hyperspectral reflectance data collected with a LeafSpec imaging device on 22 maize hybrids. Three broad strategies based on different input feature sets are undertaken. Strategy one mines data for the most informative hyperspectral channels and then constructs a normalized index similar to NDVI as input features. Strategy two considers all 364 channels of hyperspectral data and makes predictions using various machine learning techniques; partial least squares regression(PLSR), random forest regression, and a feed-forward neural net regression. Strategy three aims to take advantage of the spatial distribution of hyperspectral data on the leaf surface by training a convolutional neural net(CNN). A normalized visual index constructed from bands most correlated with nutrient content out-performed established NDVI. PLSR was the most accurate algorithm, followed by feed-forward neural net and then CNN, based on coefficient of determination score. PLSR is well established as a robust method for hyperspectral prediction which is further evidenced by this study. This is one of the first applications of CNN for hyperspectral data. Despite not being the most accurate algorithm there remains room for hyper-parameter optimization.