Evaluating the Potential of Improving In-season Nitrogen Status Diagnosis of Potato Using Leaf Fluorescence Sensors and Machine Learning

The Best Management Practices (BMPs) guideline recommends potato growers split-apply
nitrogen (N) fertilizer according to the results of petiole nitrate-nitrogen (PNN) test for improved
profitability and sustainability. However, the PNN test is a wet chemistry analysis which suffers
from destructive and laborious sampling campaign, high laboratory analysis cost, and long
laboratory turnaround time against a short window for responsive in-season N management. A
latest leaf sensor, Dualex Scientific (Dualex), is expected to overcome the disadvantages of the
PNN test. The objectives of this study are 1) to investigate how well Dualex can estimate the
PNN concentrations across different genetic, environmental, and management (GxExM)
conditions, and 2) to evaluate the PNN concentration-based potato N status classification
accuracy, and 3) to identify the best model for the PNN concentration estimation. The study was
conducted at the Sand Plain Research Farm, Becker, Minnesota in 2018 and 2019 using a
randomized complete block design with three replications. Six cultivars and three N rates were
included. This study showed that Dualex could identify in-season potato N status non- destructively at 76% accuracy by estimating PNN concentrations with GxExM information using
random forest regression. It is important to note that accumulated growing degree days and asapplied N rates were selected as two of the most important variables for PNN prediction using
the random forest regression. Dualex was also compared to a traditional leaf sensor, SPAD-502, in the capability of the PNN concentration estimation, and they were shown to be equally
capable. Further analyses and research are required to evaluate Dualex sensor under diverse
on-farm conditions and develop in-season site-specific N management strategies.