Genomic enabled two-stage analysis of high-throughput phenotyping data in crop variety trials using adaptive splines

Crop variety trials increasingly incorporate high-throughput phenotyping (HTP) data, such as normalized difference vegetation index (NDVI), collected over time. In addition to HTP data, marker or pedigree information may be available to account for genetic structure. These datasets require spatio-temporal modelling, which poses challenges when integrating spatial, temporal, and genetic components in a single analysis. A two-stage modelling approach was adopted. In the first stage, data for each plot at each time point were adjusted to remove design and spatial effects. In the second stage, the adjusted plot-level effects were used for temporal modelling. Seven temporal modelling strategies were evaluated: natural cubic smoothing splines including reduced-rank models, B-splines, P-splines, and adaptive variants of B- and P-splines with two forms. These methods were applied at both the plot and genetic levels at the second stage of analysis. The analysis focused on NDVI measurements recorded across 19 time points, representing a stay-green trait. Model comparison using the Akaike Information Criterion (AIC) indicated that adaptive B-splines provided the best fit across both levels of analysis. This suggests that adaptive spline-based methods offer improved flexibility and accuracy for modelling longitudinal HTP data in crop trials.