Şimşek et al. (2025) Phenology aware agricultural boundary extraction using segment anything model and planet scope imagery (zero shot learning approach)

Identification

• Journal: Advances in Space Research
• Year: 2025
• Date: 2025-12-29
• Authors: Fatih Fehmi Şimşek, Melih ALTAY
• DOI: 10.1016/j.asr.2025.12.086

Research Groups

• TÜBİTAK Space Technologies Research Institute, Ankara 06800, Türkiye
• Department of Geomatics Engineering, Hacettepe University, Ankara 06800, Türkiye

Short Summary

This study integrates phenology-driven multi-temporal image selection with the zero-shot segmentation capabilities of the Segment Anything Model (SAM) to automatically delineate agricultural parcel boundaries. The approach significantly improves segmentation performance, demonstrating that phenologically-based multi-temporal imagery enhances zero-shot models for accurate and operationally feasible boundary extraction.

Objective

• To integrate phenology-driven multi-temporal image selection with the zero-shot segmentation capability of SAM for the automatic delineation of agricultural parcel boundaries.

Study Configuration

• Spatial Scale: Agricultural parcel level, utilizing high-resolution PlanetScope imagery.
• Temporal Scale: Multi-temporal analysis covering a growing season (May 25, July 19, and September 22, 2022) based on phenological stages derived from Sentinel-2 NDVI time series.

Methodology and Data

• Models used: Segment Anything Model (SAM)
• Data sources:
  • Sentinel-2 satellite data (for NDVI time series to determine phenological stages)
  • PlanetScope imagery (high-resolution, multi-spectral, for segmentation input)
  • Farmer-declared parcels (used to create characteristic spectral profiles for crops)
  • Inputs to SAM included combinations of true color, false color, and NDVI time series (NDVI TS).

Main Results

• The NDVI TS approach significantly outperformed single-date segmentation.
• Multi-temporal segmentation achieved an Intersection over Union (IoU) of 0.89 and an F1 score of 0.93.
• Geometric segmentation errors for multi-temporal analysis remained low (Geometric Object-based Segmentation Error (GOSE): 0.12; Geometric Under-Segmentation Error (GUSE): 0.13).
• Single-date analyses yielded IoU values ranging from 0.73 to 0.81, with increased errors, particularly during periods of poor vegetation cover.
• The study demonstrated that phenologically-based multi-temporal image selection significantly improves segmentation performance for zero-shot models like SAM, enabling highly accurate and operationally feasible boundary extraction without extensive training data.

Contributions

• Demonstrated that phenologically-based multi-temporal image selection significantly enhances the performance of zero-shot segmentation models (SAM) for agricultural parcel boundary extraction.
• Provided an efficient, cost-effective, and directly applicable method for identifying cultivated land patterns and types using high-resolution satellite data.
• Emphasized that imagery content characteristics and acquisition time are crucial determinants of segmentation success, alongside the model itself.
• Showcased the operational feasibility of using zero-shot models for accurate boundary delineation without the need for large-scale training data.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Şimşek2025Phenology,
  author = {Şimşek, Fatih Fehmi and ALTAY, Melih},
  title = {Phenology aware agricultural boundary extraction using segment anything model and planet scope imagery (zero shot learning approach)},
  journal = {Advances in Space Research},
  year = {2025},
  doi = {10.1016/j.asr.2025.12.086},
  url = {https://doi.org/10.1016/j.asr.2025.12.086}
}