Pixel size and spatial scale in phenomics and enviromics modeling for precision breeding

Knowing how much information a single pixel can provide is not just a technical issue; it shapes what we learn, and what we will learn, from the field. This article examines how pixel size governs the capture and interpretation of environmental and phenotypic information in remote sensing, phenomics, enviromics, and precision breeding. Spatial resolution, expressed as Ground Sample Distance (GSD), connects image scale to biological meaning, with direct implications for data quality, model accuracy, and the interpretation of genotype × environment (GEI) interactions. The discussion spans multiple observation platforms: ( i ) satellites provide meter-scale data for regional and temporal monitoring; ( ii ) drones offer centimeter-level imagery suited to within-plot variation and high-throughput phenotyping; and ( iii ) intermediate resolutions connect detail, comparability, and computational efficiency. When spatial resolution is mismatched with the biological process, information loss occurs through spectral mixing, aggregation biases, and distortion of derived indices such as NDVI, canopy temperature, and biomass. Spatial scale can also be transformed through superpixel aggregation, which converts fine raster grids into homogeneous spatial objects, and through subpixel refinement, which disaggregates coarse pixels to recover local heterogeneity. Empirical evidence shows that predictive ability improves when the spatial scale of environmental covariates aligns with the ecological and experimental scale under study. These concepts extend to experimental design, Target Population of Environments (TPE) mapping, and recommendation zoning through GEI mapping. This review treats spatial resolution strategically as a modeling parameter, not as a fixed technical constraint. Choosing pixel size through sensitivity analysis and study objectives helps align environmental representation, predictive modeling, and comparability across experiments. We also discuss perspectives involving autonomous sensing, multiscale data processing, computational infrastructure, and the use of different spatial scales for genomic and genomic–enviromic prediction