The Transferability of Spectral Grain Yield Prediction in Wheat Breeding across Years and Trial Locations

Abstract

Grain yield (GY) prediction based on non-destructive UAV-based spectral sensing could make screening of large field trials more efficient and objective. However, the transfer of models remains challenging, and is affected by location, year-dependent weather conditions and measurement dates. Therefore, this study evaluates GY modelling across years and locations, considering the effect of measurement dates within years. Based on a previous study, we used a normalized difference red edge (NDRE1) index with PLS (partial least squares) regression, trained and tested with the data of individual dates and date combinations, respectively. While strong differences in model performance were observed between test datasets, i.e., different trials, as well as between measurement dates, the effect of the train datasets was comparably small. Generally, within-trials models achieved better predictions (max. R2 = 0.27–0.81), but R2-values for the best across-trials models were lower only by 0.03–0.13. Within train and test datasets, measurement dates had a strong influence on model performance. While measurements during flowering and early milk ripeness were confirmed for within- and across-trials models, later dates were less useful for across-trials models. For most test sets, multi-date models revealed to improve predictions compared to individual-date models.

High-Throughput Field Phenotyping to Advance Precision Agriculture and Enhance Genetic Gain, Volume II

“…Furthermore, the VIs show a more consistent performance between dates compared to the single band reflectances. The NDRE was the best performing feature for yield assessment, which is in accordance with other studies (Prey et al, 2022) possibly due to its strong correlation to biomass (Argento et al, 2021). Many VIs have been screened by and few have been showing a consistent performance over the years, which makes a general selection difficult, similar to our study.…”

“…Primary traits such as grain yield and quality have been assessed by estimating the mentioned secondary traits during the growth season Vatter et al, 2022). A variety of sensors have been employed such as RGB cameras (Fernandez-Gallego et al, 2019), multispectral cameras (Prey et al, 2022), hyperspectral sensors (Bowman et al, 2015), thermal cameras (Elsayed et al, 2017) and active sensors such as Lidar to mention a few. Among these technologies, multispectral cameras offer a high work efficiency for a relatively low cost.…”

“…Second, the reflectance of single bands can be combined to calculate vegetation indices (VIs), which are often more sensitive to specific traits and less affected by environmental conditions during measurement (Tucker, 1979). This approach has been used by several studies for yield prediction Prey et al, 2022). However, single-band reflectance and VIs may suffer from saturation, particularly when the canopies are closed (Rischbeck et al, 2016).…”

“…Late stages such as the anthesis stage and the grain filling stage are often identified as the most suitable stages for yield prediction in wheat when using VIs (Bowman et al, 2015;. Earlier stages such the tillering stage are generally and naturally performing worse (Prey et al, 2022). Still, Walsh et al (2022) successfully predicted yield at the tillering stage and Marti et al (2007) at the stem elongation stage, both using the normalized difference vegetation index (NDVI).…”

“…For instance, Pinter et al (1981) suggested summing measurement dates after heading to improve yield prediction in wheat and barley whereas Raun et al (2001) suggested taking two spectral measurements after dormancy. Prey et al (2022) showed that models containing data from multiple dates could improve yield predictions and compensate if data could not be collected on the optimal date due to practical reasons and phenological shifts between years.…”