Advancing early-stage plant phosphorus assessment for winter rye via hyperspectral data: A model-based approach harnessing feedforward neural networks

Context

Phosphorus (P) deficiency is a critical limiting factor in crop production, significantly impacting growth and yield, particularly in winter rye. Traditional methods for detecting P deficiency often face challenges in terms of accuracy and timeliness, especially during the early stages of crop development. Hyperspectral remote sensing presents a promising alternative for monitoring nutrient stress, while feedforward neural networks (FNNs) offer robust predictive capabilities for data analysis.

Objective

This study seeks to develop an innovative method for detecting phosphorus deficiencies in winter rye during early growth stages by integrating hyperspectral data with feedforward neural networks. The primary objective is to improve the efficiency and scalability of phosphorus (P) deficiency detection, with a particular focus on reducing root mean square error (RMSE) and enhancing the responsiveness of detection compared to traditional phosphorus detection methods, eliminating the need for time-consuming sample collection and chemical analysis.

Methods

Field experiments were conducted at two distinct locations: the Professor Marian Górski Experimental Station in Skierniewice, Poland, and experimental fields in Połczyn, West Pomeranian Voivodeship, during the months of April and May 2023. Hyperspectral data were acquired through drone-based, FieldSpec ground-based, and satellite measurements. Plant and soil samples were analyzed to assess nutrient content. The hyperspectral reflectance data were processed using feedforward neural networks (FNNs), which were trained to predict phosphorus levels based on spectral data collected from both ground-based and drone-based sensors. Additionally, spectral channels were aligned with the Sentinel-2 and PlanetScope satellite bands for broader applicability.

Results and conclusions

The integration of hyperspectral data with FNNs significantly improved the accuracy and timeliness of phosphorus deficiency detection. The spectral ranges of 500–550 nm and 950–1000 nm were identified as crucial for accurate prediction. The model achieved a root mean square error (RMSE) of 0.84 g/kg for phosphorus content and 64 % accuracy in identifying deficiency points, demonstrating the efficacy of this approach for early-stage nutrient monitoring in winter rye.

Significance

This study contributes to the advancement of precision agriculture by providing an efficient, accurate method for early-stage phosphorus deficiency detection in winter rye. The findings hold practical implications for optimizing fertilization strategies, improving crop management practices, and promoting sustainable farming.