Evaluating the Potential of Improving In-season Nitrogen Status Diagnosis of Potato Using Leaf Fluorescence Sensors and Machine Learning
1S. Wakahara, 1Y. Miao, 1K. Mizuta, 1J. Zhang, 3D. Li, 1S. Gupta, 1C. Rosen
1. Precision Agriculture Center, University of Minnesota, St. Paul, MN 55108 USA
2. Agricultural University of Hebei, Baoding, Hebei 071051 China
3. Guangzhou Institute of Geography, Guangdong Academy of Sciences, Guangzhou, Guangdong, 510070 China
Precision nitrogen (N) management is particularly important for potato crops due to their high N fertilizer demand and high N leaching potential caused by their shallow root systems and preference for coarse-textured soils. Potato farmers have been using a standard lab analysis called petiole nitrate-N (PNN) test as a tool to diagnose potato N status and guide in-season N management. However, the PNN test suffers from many disadvantages including time constraints, labor, and cost of analysis. Dualex Scientific, a leaf fluorescence sensor capable of chlorophyll, flavonol, and anthocyanin measurements, has the potential to be used to overcome these disadvantages. Studies on the potential of using this sensor for in-season potato N status diagnosis are still limited. The objective of this research was to determine the potential of improving PNN prediction for in-season potato N status diagnosis using the Dualex sensor and compared it with the SPAD chlorophyll meter. Field experiments were conducted at the Sand Plain Research Farm, in Becker, MN, USA in 2018 and 2019 on a Hubbard loamy sand using a randomized complete block design with three replications. Seven cultivars and three N rates were used. Dualex sensor and SPAD meter data as well PNN data were collected four times during the growing season. Simple regression, multivariate linear regression, and machine learning (Support Vector Machine and Random Forest) models were used to predict PNN using leaf sensor data and ancillary data. Coefficient of determination (R2), root mean square error (RMSE), percent error (PE), and Kappa statistic were used to assess model performance. The preliminary results indicated the Dualex sensor performed the best for non-destructive diagnosis of potato N status when genetics, environment, and management data were fused together using random forest regression (adjusted R2=0.89, RMSE=2579.70 ppm, PE=26.80%, and the Kappa statistics=0.65 with validation dataset). Using the data fusion approach, the Dualex sensor and SPAD meter performed similarly. More analyses are being performed to evaluate different approaches to use the Dualex sensor and SPAD meter for potato N status diagnosis and will be presented at the conference.
