Phenotyping From Space: Evaluating Very High-Resolution Satellite-Based Trait Estimation in Winter Wheat Microplot Trials

Remote sensing enables non-destructive, high-throughput monitoring of crop physiological traits and has become a key tool in field phenotyping. Over the past decade, sensing technologies and data-driven approaches have been widely adopted in crop improvement programs. Platforms ranging from proximal sensors to UAVs have been developed to enhance the efficiency and objectivity of trait measurements in field trials. Despite their advantages, proximal and UAV-based phenotyping present several limitations, including high operational costs, time-consuming deployment, and sensitivity to weather conditions—particularly for UAVs, which require low wind speeds and clear skies. Moreover, replicating multi-environment trials remains challenging due to the need for strict protocols and comparable equipment across sites. To address these limitations, we evaluated the potential of Very High-Resolution (VHR) satellite imagery for microplot-level phenotyping in a controlled field experiment. Cloud-free satellite images were acquired in April and May of 2022 and 2023 over a winter wheat trial in Mauguio, France, comprising 170 microplots (1.36 × 8 m) with ten varieties, two water treatments, and two sowing dates. The dataset included Sentinel-2 multispectral images (10 bands) super-resolved to 5 m and 1 m using deep learning, and WorldView-2 multispectral imagery (8 bands) at 1.6 m, pan-sharpened to a spatial resolution of 0.4 m. All acquisitions were within five days of in situ biomass sampling and phenomobile measurements. The latter included namely lidar-based Green Area Index (GAI), used as ground-truth. A state-of-the-art neural network (BV-NET) was applied to retrieve GAI from each satellite image, from which mean GAI and vegetation indices were extracted per microplot and date. This study addresses two key objectives: (1) evaluating the influence of high-resolution image pre-processing (i.e., super-resolution and pan-sharpening) on the retrieval of biophysical variables at the microplot level, and (2) assessing the performance of each sensor and resolution in estimating ground-based phenotyping data and capturing treatment effects. Results indicate that VHR imagery, when pan-sharpened to 0.4 m using suitable methods, effectively captures treatment-induced variability among microplots. These findings support the use of VHR satellite imagery as a scalable, consistent, and reliable tool for field phenotyping in multi-environment trials.