UAV multispectral to hyperspectral reconstruction based on deep learning and radiative transfer models for crop nitrogen monitoring

Abstract

Accurate and timely monitoring of canopy nitrogen (N) status is essential for precision farming and understanding N dynamics in vegetation systems. Uncrewed aerial vehicle (UAV)-based multispectral sensors are widely used for canopy monitoring due to their high spatial and temporal resolutions. However, their broad spectral bandwidths and the lack of short-wave infrared (SWIR) bands limit their effectiveness in advanced canopy nitrogen estimation models. In this study, we propose a novel framework of multispectral to hyperspectral reconstruction method (M2H–SWIR) by integrating the physically-based radiative transfer modeling (PROSAIL-PRO) with convolutional neural networks (CNN) to reconstruct full-spectrum spectral reflectance (400–2500nm) from the visible to near infrared broad bands (400–900nm) multispectral UAV images. The reconstructed hyperspectral data were validated against ground-based spectrometer-measured canopy reflectance data, where averaging strategies were used to reduce footprint and scale mismatches between UAV multispectral imagery and ground-based spectrometer-measured canopy reflectance observations. Our results demonstrated that the LUT-M2H–SWIR approach outperformed traditional PROSAIL-PRO inversion methods, particularly in reproducing the nitrogen-sensitive SWIR bands. Despite this success, extrapolating the short-range multispectral data to the long-range SWIR bands remains challenging, primarily due to spectral absorption overlapping from water absorption and canopy structural effects. Additional validation through canopy nitrogen estimation using the reconstructed hyperspectral reflectance reinforced the robustness of our proposed approach, suggesting its potential for operational canopy nitrogen mapping.

Jie Wang

PhD student

Using remote sensing to study the impact of climate change and cropping practices on global crop yields and nitrogen use efficiency

Dr. Anirudh Belwalkar

Postdoctoral Researcher

My research interests include hyperspectral remote sensing of vegetation and precision agriculture.

Prof. Dr. Kang Yu

Professor of Precision Agriculture

My research interests include precision crop farming, hyperspectral remote sensing, and AI in agriculture.