Multi-angular spectroscopic detection of winter wheat nitrogen fertilizer utilization status using integrated feature selection and machine learning

Abstract

Nitrogen (N) utilization rate is a key index used to assess whether N fertilizer is applied rationally. In addition, it can reflect crop growth. However, research on multi-angular spectral real-time monitoring of physiological indexes of N efficiency (photosynthetic N-use efficiency [PNUE]) during wheat growth and advance prediction of final N-use efficiency (NUE) at maturity is scant. Consequently, the accuracy of existing methods is estimating N fertilizer utilization status with remotely sensed data is low, and the mechanisms underlying the relationship between reflectance and PNUE remain unclear. To address the knowledge gap, in the present study, two wavelength variable-selected algorithms, competitive adaptive reweighted sampling (CARS) and feature selection learning (ReliefF), were used to identify wavebands sensitive to PNUE. The screened feature bands were used as inputs in the input layer of four multivariate algorithms (Partial Least Squares Regression [PLSR], Support Vector Regression [SVR], Artificial Neural Network [ANN], and Random Forest [RF]) to determine the best model for monitoring PNUE and predicting NUE before wheat ripening. Compared to all machine learning methods, the PLSR-based CARS (CARS-PLSR) algorithm predicts PNUE with an accuracy >90 % at 13 observation angles. At last, we predicted the NUE according to the PNUE-NUE correlation and the CARS-PLSR-PNUE correlation. The lack of significant differences in slope and intercept across the five growth stages indicates that the CARS-PLSR model is a better PNUE tracker and NUE predictor in diverse field conditions. The combination of remote sensing techniques and integrated evaluation approaches provides accurate and timely information on crop N fertilizer utilization status, which could facilitate tailoring N fertilizer management to wheat requirements, thus maintaining N fertility for high photosynthetic yield, while minimizing N losses to the environment.

Plant phenotyping relevance

多角度スペクトル計測と特徴選択・機械学習を用いて、コムギのPNUEおよびNUEを推定・予測する手法が研究の中心であり、植物生理形質の取得・抽出に該当する。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.