Quantification of in-season crop traits and post-harvest residues with hyperspectral remote sensing

Increasing global food demand due to climate change impact on production and rapid population growth can only be met through more sustainable, resilient and smart agricultural practices. Monitoring crop conditions through Earth Observation (EO) data is crucial for the development of sustainable intensification strategies. In particular, spaceborne hyperspectral imaging (HIS) from current and future satellite missions enhances traditional multispectral data by providing detailed information on field status and plant condition. This chapter demonstrates the value of HSI in estimating in-season crop traits and post-harvest crop residues (CR), using machine learning regression algorithms (MLRA). Case studies, exploiting images acquired by the hyperspectral satellite PRISMA, launched by the Italian Space Agency (ASI), are presented. The estimation and mapping of green vegetation crop traits was successfully achieved through the implementation of an active learning (AL) procedure in conjunction with a hybrid approach (radiative transfer model simulation + MLRA). Canopy-level traits were predicted with Gaussian Process Regression (GPR) models on independent datasets with satisfactory to high accuracy (e.g., Chlorophyll/Nitrogen Content: R²>0.50, nRMSE ∼ 16%; LAI: R²>0.75, nRMSE ∼ 12%). Generated maps enabled the identification of within-field patterns related to crop nutritional status. The quantification of crop residues (CR) was achieved using a Random Forest (RF) model, trained with a comprehensive spectral library, after proper spectroscopic features modeling. The model, validated against an independent data set, demonstrated a high level of accuracy (R²>0.7, RMSE 0.028) and allowed to map and quantify CR distribution at a farm scale. The high degree of accuracy and consistency observed in these results suggests the potential of spaceborne HSI for crop monitoring providing a complementary data source to generate value-added geo-information to support agriculture.