Maffei et al. (2025) Monitoring soil substrate influence in vineyards using Sentinel-2 time series and land surface phenology

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2025
• Date: 2025-11-30
• Authors: Thomas Maffei, Marco Moretto, Pietro Franceschi
• DOI: 10.1016/j.jag.2025.104977

Research Groups

Unit of Digital Agriculture, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

Short Summary

This study investigates the potential of Sentinel-2 time-series and Land Surface Phenology (LSP) to differentiate vineyards based on varying soil substrate types. It found that both growing season and off-season metrics, particularly vegetation water content indices like Global Vegetation Moisture Index (GVMI), effectively distinguish soil substrate effects with high temporal stability, with off-season metrics showing better transferability across years.

Objective

• To explore the potential of Sentinel-2 time series imagery to assess differences in vineyard responses to various soil substrates by integrating Land Surface Phenology (LSP), growing season (GS), and off-season metrics.

Study Configuration

• Spatial Scale: Piana Rotaliana wine-growing area, Autonomous Province of Trento, Italy, spanning approximately 80 square kilometers, encompassing 1142 Pergola-trained vineyards.
• Temporal Scale: Sentinel-2 satellite data acquired annually from 2017 to 2023.

Methodology and Data

• Models used:
  • Random Forest classifier for soil substrate classification.
  • Weighted Whittaker method for time series reconstruction.
  • Beck double logistic model for Land Surface Phenology (LSP) curve fitting.
  • Threshold Method (TRS) and Gu Method for LSP metrics extraction.
• Data sources:
  • Sentinel-2A and Sentinel-2B Multi-Spectral Instrument (MSI) Level-2 atmospherically corrected surface reflectance time series (bands B02, B03, B04, B05, B06, B07, B08, B11, B12, and Scene Classification Layer).
  • Detailed soil maps (1:25000 scale or finer) from the PICA project, classifying Alluvial Group (AL) and Conoid Group (CO) soils.
  • Cadastral parcels for vineyard delineation.
  • Aerial orthophotos and Planet basemaps for vineyard selection and plantation renewal assessment.
  • Vegetation indices: Enhanced Vegetation Index 2 (EVI2), Normalized Difference Vegetation Index (NDVI), Global Vegetation Moisture Index (GVMI), green Normalized Difference Vegetation Index (gNDVI), Normalized Difference Moisture Index (NDMI).

Main Results

• Enhanced Vegetation Index 2 (EVI2) demonstrated better stability than Normalized Difference Vegetation Index (NDVI) for Land Surface Phenology (LSP) estimation in vineyards.
• The Beck double logistic model was selected for LSP assessment due to its lower variability and continuity, aligning with vineyard growth patterns.
• Random Forest models achieved high and comparable classification accuracies (approximately 0.85) in distinguishing between Alluvial Group (AL) and Conoid Group (CO) soil substrates.
• Models based solely on LSP metrics were less effective than those incorporating growing season (GS) or off-season metrics, indicating the importance of spectral intensity.
• Vegetation water content indices, particularly Global Vegetation Moisture Index (GVMI), emerged as the most effective and temporally stable predictors for differentiating soil substrates.
• Off-season datasets consistently provided better results and higher temporal stability when models trained on one year were tested on other years, outperforming GS and LSP metrics.
• A linear mixed model confirmed a significant fixed effect between the two substrates across years, with the substrate:year interaction having no contribution to variance, indicating a constant soil effect over time.
• Spatial analysis showed lower GVMI values predominantly over Conoid Group soils and higher values over Alluvial Group soils, consistent with their respective non-hydromorphic and hydromorphic water dynamics.

Contributions

• Presents a novel methodology for characterizing vineyard responses to soil substrate types using Sentinel-2 time series and Land Surface Phenology at a regional scale.
• Identifies EVI2 and the Beck double logistic model as optimal for vineyard LSP extraction, addressing a gap in specific literature.
• Demonstrates the effectiveness of satellite data for large-scale (over 1000 vineyards) soil substrate characterization, complementing traditional field surveys.
• Highlights the superior temporal stability and transferability of off-season vegetation water content metrics (e.g., GVMI) for distinguishing soil types, suggesting applicability to other crops and trellis systems.
• Provides evidence for the consistent influence of soil water retention capacity on grapevine vigor over time, even amidst inter-annual weather variability.
• Suggests that satellite information can be used to optimize soil sampling strategies by identifying small-scale variability often overlooked in regional soil maps.

Funding

• Autonomous Province of Trento’s IRRITRE project: territorial information system for precision irrigation in Trentino - CUP D43C23001950003.

Citation

@article{Maffei2025Monitoring,
  author = {Maffei, Thomas and Moretto, Marco and Franceschi, Pietro},
  title = {Monitoring soil substrate influence in vineyards using Sentinel-2 time series and land surface phenology},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2025},
  doi = {10.1016/j.jag.2025.104977},
  url = {https://doi.org/10.1016/j.jag.2025.104977}
}