The research team, Manuel Sampériz Sarvisé and Asunción Usón Murillo, focused on a deceptively simple statistic: the standard deviation of the Normalized Difference Vegetation Index, or NDVI. NDVI is calculated from the red and near-infrared light that plants reflect, and it has been a staple of agricultural remote sensing since the 1970s. Healthy, dense canopies absorb red light strongly and reflect near-infrared, pushing NDVI values higher. What makes this study different is that the authors did not chase the average greenness of a field. Instead, they tracked how much NDVI varied from pixel to pixel within each plot, treating that spread as a fingerprint of the soil beneath.

The study area was the 2,324-hectare Sardeta ravine basin in the Los Monegros region of Aragón, northeastern Spain, an arid landscape of hot summers, cold winters, and scarce rainfall. The soils there developed on limestone sediments and fall into distinct units, from poorly developed Xerorthents on valley bottoms and floodplains to moderately developed Calcixerepts on stable platforms and slopes. Barley dominates the winter cropping, covering more than 33 percent of the cultivated area in Robres and over 47 percent in Montesusín, while maize and alfalfa fill the irrigated summer calendar. Because agronomic practices such as sowing dates, fertilization, and irrigation were uniform within each plot, any spatial differences in crop development could be attributed primarily to soil properties rather than to management decisions.

To capture crop dynamics, the researchers drew on two complementary satellite sources. Sentinel-2 imagery, freely available at 10-meter resolution, provided the temporal backbone, allowing the team to build NDVI time series for every plot between 2019 and 2023. When finer spatial detail was needed, they turned to PlanetScope, a commercial constellation delivering 3-meter resolution imagery. The workflow was deliberately simple: plot boundaries defined in open-source GIS software were uploaded to the Sentinel Hub EO Browser, NDVI values were extracted for every cloud-free acquisition, and zonal statistics such as the mean, median, and standard deviation were computed for each field and date. Only images with zero cloud cover over a plot were retained, a strict filter that reduced the number of observations but guaranteed clean signals.

A crucial methodological innovation was the use of phenological markers to standardize comparisons. The team identified three key stages in each crop cycle: the start of the season, when the crop emerges and establishes; the period of maximum development, when canopy cover peaks; and the end of the season, when the crop senesces. Detecting these stages in irrigated semi-arid systems is not straightforward, because weeds, irrigation interruptions, and soil background effects can produce transient NDVI fluctuations that mimic real transitions. The researchers therefore applied a temporal consistency criterion, requiring several consecutive satellite observations to move in the same direction before declaring a stage change. For emergence, NDVI had to exceed roughly 40 percent of the seasonal amplitude across four consecutive images; for senescence onset, values had to fall below about 90 percent of the peak after three successive declines.

The first major finding came from comparing soil units across the basin. Median NDVI values barely differed among soil units or between summer and winter crops, a result the authors attribute to the well-known saturation of NDVI under dense canopies, where red reflectance maxes out and sensitivity to further biomass gains collapses. But the standard deviation told a very different story. The Glacis Slope unit, which covers nearly 72 percent of the basin with fine-silty Calcic Haploxerepts, showed the highest within-plot variability, while the Platform unit, a small area of petrocalcic soils, was remarkably homogeneous. Floodplain and redoximorphic floodplain soils fell in between. Spatial variability also increased as the season progressed, rising from 0.043 at emergence to 0.061 at senescence, which pointed to the end of the season as the most informative window for detecting soil-driven differences.

Emergence timing, by contrast, proved noisy. Start-of-season dates varied by up to three months among winter crops and about two months among summer crops, reflecting differences in sowing decisions and early establishment conditions rather than soil alone. To strip away this management noise, the team zoomed in on a single 21-hectare field in Montesusín that spans two contrasting soil units, Platform and Glacis Slope, and is managed uniformly. Within this field, they randomly delineated twelve sampling polygons of roughly 1,000 square meters each, six per unit, avoiding field edges and traffic zones. Sentinel-2 data identified the phenological stages, and PlanetScope imagery then resolved the spatial pattern at 3-meter resolution across every crop cycle from 2019 to 2023.

The controlled analysis sharpened the picture considerably. At emergence, NDVI medians were higher on the Platform and in winter cereals, possibly because wetter winter soils increase reflectance and because coarser Platform textures favor faster establishment, while widely spaced maize rows leave bare ground that drags early NDVI down. At peak development, the Platform again showed higher medians and lower variability, confirming more uniform canopy growth on that unit. But it was senescence that delivered the clearest signal: NDVI standard deviation responded significantly to soil unit, crop type, and their interaction, with the Glacis Slope showing far greater heterogeneity than the Platform, especially in winter crops. Once irrigation stops and evapotranspiration demand rises, differences in soil water-holding capacity become the dominant constraint, and the crops dry unevenly in ways satellites can see.

To understand the mechanisms behind these patterns, the researchers ran correlation and principal component analyses on a broad set of soil, terrain, and crop variables. Spearman rank correlations linked NDVI standard deviation strongly to terrain and soil moisture-related factors, while mean NDVI was inversely related to variability, meaning that lusher, more vigorous crops grew more uniformly. The first two principal components explained 37.33 percent of total variance. The first axis contrasted sand content and elevation against clay, capturing a gradient of texture and permeability that governs water movement. The second axis combined soil unit, slope, and organic matter, representing a gradient of heterogeneity and moisture retention. Notably, soil unit clustered with NDVI standard deviation in the loading plot, reinforcing the interpretation that dispersion metrics act as proxies for soil-driven crop variability.

The practical implications reach well beyond one Spanish basin. Because the method relies on freely available Sentinel-2 data, simple statistics, and open-source tools, it can be replicated almost anywhere without expensive sensors or specialized expertise, offering a realistic entry point for precision agriculture on farms that cannot justify LiDAR surveys or intensive soil sampling. Knowing where within-field variability concentrates, and recognizing that senescence is the stage when soil differences shout loudest, allows farmers to target variable-rate irrigation, fertilization, or seeding to the zones that need them. The authors acknowledge limitations, including NDVI saturation under dense canopies and the masking effect of differing management across farms, and they call for future work integrating synthetic aperture radar and additional vegetation indices. Still, the core message stands: the ground beneath a field writes its signature into the way crops green up and dry down, and satellites are now cheap enough to read it.

Subject of Research: Using NDVI dispersion metrics from Sentinel-2 and PlanetScope imagery to assess within-plot soil variability in irrigated cereal systems

Article Title: Remote Sensing for Within-Plot Soil Variability Assessment Using NDVI Dispersion Metrics

Article References: Sampériz Sarvisé, M., & Usón Murillo, A. (2026). Remote Sensing for Within-Plot Soil Variability Assessment Using NDVI Dispersion Metrics. Spanish Journal of Soil Science, 16, Article 14694. https://doi.org/10.3389/sjss.2026.14694

Image Credits: AI Generated

DOI: 10.3389/sjss.2026.14694

Keywords: remote sensing, NDVI, soil variability, precision agriculture, Sentinel-2, PlanetScope, phenology, crop senescence, soil moisture, irrigated cereals, Spain, satellite imagery