Quantifying crop nitrogen (N) and disease stress at actionable spatial and temporal scales is essential for improving nitrogen use efficiency, reducing N losses and targeted use of pesticides in intensive Danish and worldwide agricultural systems. Yet current approaches often rely on sparse sampling or models that do not transfer robustly across crops, seasons, regional and national scale. Here, we develop a cross-scale remote-sensing framework to (i) quantify crop leaf and canopy N content, and (ii) provide early warning of emerging nitrogen stress and crop disease infection using time-series change detection and multi-sensor feature fusion. The targeted crop systems reflect dominant production and management contexts, focusing on winter wheat, potato, sugar beet, and cover crops. Preliminary evaluation indicates strong predictive power for crop N quantification (cross-validated R2 = 0.84 for canopy N content) and moderate power for the Verticillium wilt disease detection (R2 = 0.67), using Xbost  as machine learning algorithms, respectively. ​

The framework integrates ground truth measurements (leaf sampling), UAV-based multispectral retrievals, satellite time-series from Sentinel-2 and PlanetScope, weather conditions from ERA5, soil properties, and other agricultural related information.  We will develop spectral-index and full-spectrum retrieval models for crop N content and disease infection, then couple them with process-guided machine learning (PGML), domain adaptation, and uncertainty-aware upscaling to translate UAV-calibrated relationships to satellite pixels across crop types and across the Danish national scale.