Prediction of regrowth and biomass of perennial sorghum using unoccupied aerial systems

Abstract

Perennial grain sorghum [Sorghum bicolor (L.) Moench] has potential to produce grain and forage while improving soil health, ecosystem services, and carbon soil sequestration but requires further genetic improvement. Unoccupied aerial systems (UAS, also known as drones and unmanned aerial systems) provide opportunities to quickly evaluate plant traits on a large scale with precision. Unoccupied aerial system flights were used to evaluate biomass yield and rhizome characteristics of 100 diverse sorghum hybrids, most being from an interspecific hybridization program, in the establishment year and first year of regrowth. Twenty‐one vegetation indices (VIs) with canopy height measurements (CHMs) were processed from seven UAS flights made temporally during each growing season. Regression of the temporal data (VI and CHM) and phenotypic traits, including rhizome characteristics based on plant stand count (PSC), rhizome‐derived shoots (RDS), and fresh and dry biomass yields, showed useful predictions when combining temporal VI with CHM and machine learning. Blue chromatic coordinate index (BCC) best predicted all measured traits. If predictions could be generalized, UAS would reduce field evaluation time for perennial sorghum or breeding perennial grasses in general and allow breeders to evaluate additional genotypes. In this study, we found that optimizing flights to specific dates after planting could minimize resource requirements and costs in prediction of regrowth and biomass yield of perennial sorghum.

Plant phenotyping relevance

UAS画像から植生指数・ canopy height と機械学習を用いてソルガムのバイオマスや根茎関連形質を予測する手法が研究の中心であり、植物表現型の取得・推定を実質的に評価している。

Code and data availability

The supplied blocks describe UAS flights, field phenotyping, and regression modeling for perennial sorghum, but contain no data availability statement, public phenotype dataset, image repository, or author analysis code with a public URL. All URLs in the text are references to prior work or generic software (AgiSoft,LA

No evidence-backed public reproduction asset is currently recorded.