Skip to main content
ARS Home » Southeast Area » Canal Point, Florida » Sugarcane Field Station » Research » Publications at this Location » Publication #386275
Research Project: Development of High-Yielding, Stress Tolerant Sugarcane Cultivars Using Agronomic, Genetic, and Molecular Approaches

Location: Sugarcane Field Station

Title: Sugarcane yield prediction and genotype selection using unmanned aerial vehicle-based hyperspectral imaging and machine learning

Author
item CHIRANJIBI, POUDYALA - Texas A&M University
item LUCAS FIDELES, COSTA - University Of Florida
item SANDHU, HARDEV - University Of Florida
item YIANNIS, AMPATZIDAS - University Of Florida
item DENNIS CALVIN, ODERO - University Of Florida
item Coto Arbelo, Orlando
item CHERRY, RONALD - University Of Florida

Submitted to: Agronomy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/6/2022
Publication Date: 5/30/2022
Citation: Chiranjibi, P., Lucas Fideles, C., Sandhu, H., Yiannis, A., Dennis Calvin, O., Coto Arbelo, O., Cherry, R.A. 2022. Sugarcane yield prediction and genotype selection using unmanned aerial vehicle-based hyperspectral imaging and machine learning. Agronomy Journal. https://doi.org/10.1002/agj2.21133.
DOI: https://doi.org/10.1002/agj2.21133

Interpretive Summary: Sugarcane is a high biomass perennial crop in which manual data collection for early yield prediction through its growth cycle is labor intensive and a time consuming task. Aerial imagery can be explored to predict yield parameters and high throughput phenotyping for genetic selection. In this study, aerial imagery was used to predict sucrose content, cane and sugar yields during final selection stage of the breeding program in Florida. Results showed that yield was predicted with greater accuracy in July in plant cane and April in the first ratoon. Sucrose percentage was predicted with greater accuracy than cane and sugar yields. Results showed that aerial imagery may be useful in making genotype selection based on cane and sugar yields.

Technical Abstract: Sugarcane (Saccharum spp. hybrids) is a high biomass perennial crop in which manual data collection for early yield prediction, through its growth cycle is labor intensive and a time consuming task. Alternately, aerial imagery can be explored to predict yield parameters and high throughput phenotyping for genetic selection. In this study, aerial imagery and ground data were collected in the final stage of genotype selection of a sugarcane cultivar development program in Florida to evaluate the use of unmanned aerial vehicles (UAVs) in yield prediction (tons of cane per hectare [TCH], sucrose concentration, and tons of sugar per hectare [TSH], and genotype selection in 12 new genotypes in plant cane and 9 new genotypes in first ratoon. The Gradient Boosting Regression Tree model was chosen as the best algorithm based on a low mean absolute percentage error (MAPE). Results showed that yield was predicted with greater accuracy in July in plant cane and April in the first ratoon. Sucrose percentage was predicted with greater accuracy (94% and 93% in plant cane and first ratoon, respectively) than TCH and TSH yields. Although only two out of top five genotypes were common in both selection methods (measured versus predicted yields) the high accuracy in TCH and sucrose percentage predictions shows that aerial imagery may be useful in making genotype selection in sugarcane.