Mapping landscape canopy nitrogen content from space using PRISMA data.

Area of Science:

• Remote Sensing
• Agricultural Science
• Ecology

Background:

• Satellite imaging spectroscopy missions like PRISMA and EnMAP are maturing, with CHIME planned for global hyperspectral data.
• Hyperspectral data enables new vegetation trait retrieval models, crucial for applications like sustainable agriculture and biodiversity management.
• Canopy nitrogen content (CNC) is a key vegetation trait for agricultural monitoring.

Purpose of the Study:

• To develop and validate the first hybrid Canopy Nitrogen Content (CNC) retrieval model for operational use with spaceborne imaging spectroscopy data.
• To combine physically-based radiative transfer models with machine learning and active learning for robust CNC estimation.
• To demonstrate the model's capability for large-scale agricultural monitoring.

Main Methods:

• Coupled PROSPECT-PRO and SAIL radiative transfer models to generate diverse vegetation training data.
• Employed dimensionality reduction and active learning with Gaussian Process Regression (GPR) for model refinement using in-situ data.
• Incorporated non-vegetated spectra to enhance model performance across varying landscapes.

Main Results:

• The hybrid CNC model achieved a Root Mean Square Error (RMSE) of 3.4 g/m² and R² of 0.7 against field data.
• Applied to PRISMA data over Munich, the model produced reliable CNC estimates and uncertainty maps for agricultural areas.
• Demonstrated the feasibility of operational, routine CNC quantification from space.

Conclusions:

• The developed hybrid model offers a robust method for estimating Canopy Nitrogen Content (CNC) from hyperspectral satellite imagery.
• This approach is suitable for operational applications, supporting precision agriculture and environmental monitoring initiatives.
• The findings support the potential of future missions like CHIME for global vegetation trait assessment.