PHENOSPAT-N: Uncertainty-Aware Multi-Season Sentinel-2 Spectral–Spatial Optimization for Reproducible Fertilizer-Management Zoning and Targeted Nitrogen Assessment
Year: 2026
Type: article
Abstract: Purpose: Inefficient nitrogen (N) management in heterogeneous agricultural landscapes reduces nitrogen use efficiency and increases environmental risks, as uniform fertilizer applications cannot match spatially variable crop demand driven by soil properties, moisture, and management history. This study aimed to determine whether multi-season Sentinel-2 (S2) observations can be transformed into spatially coherent, reproducible fertilizer-management zones (FMZs) suitable for site-specific nutrient management while maintaining strict independence from soil-fertility and crop-yield data during zone formation. Methods: The framework was implemented over 1227.59 ha in the semi-arid Pilani region (Jhunjhunu district, Rajasthan, India) using S2 Level-2A imagery from the 2024–2025 agricultural cycle grouped into five phenological periods (pre-kharif to late rabi). Five indices (NDVI, NDRE, GNDVI, NDMI, BSI) were composited by seasonal medians to create 25 multitemporal layers, aggregated to an operational 100 × 100 m grid (with 1220 model cells after completeness and area filtering). Principal component analysis with permutation-based parallel analysis retained five components (95.17% variance), with PC1–PC3 (82.83%) used for zoning. Five FMZs were delineated via multi-objective spectral–spatial optimization using Differential Evolution, penalizing within-zone spectral dispersion, weak inter-zone separation, spatial dispersion, fragmentation, and zone-size imbalance under a minimum zone-size constraint. Results: Minimum mapping unit (MMU) regularization quantified a spectral–spatial trade-off; MMU3 reduced the number of connected components from 112 to 32 (71.4%) while maintaining a silhouette coefficient = 0.233 and DBI = 1.097, whereas stronger MMU5–MMU10 weakened spectral separation. Across 20 repeated optimizations, reproducibility was high (ARI = 0.817; NMI = 0.817), with 91.66% mean cell agreement and uncertainty concentrated in transitional areas (24.10% variably assigned). Independent validation using 250 georeferenced samples showed strong FMZ differentiation for soil organic carbon (H = 63.489, FDR-adjusted p = 9.10 × 10−12, ε2 = 0.243; highest median in FMZ 5 = 0.560 g kg−1; lowest in FMZ 2 = 0.310 g kg−1), while soil N (adjusted p = 0.224) and multi-crop productivity (Yield PC1; adjusted p = 0.224) were not significant after FDR correction. Conclusions: Overall, multi-season S2 data, coupled with uncertainty-aware spectral–spatial optimization, produced stable, operational FMZs that captured meaningful soil-environment variability (organic carbon); however, these should be interpreted as management-priority domains for zone-stratified soil-N sampling, targeted verification of uncertain zone boundaries, and replicated N-rate response trials to support subsequent zone-specific fertilizer recommendations rather than direct soil-N classes or fertilizer-rate prescriptions.
Source: Nitrogen
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,207
OpenAlex ID: W7225655806