Raju et al. (2026) Temporal trend analysis and multi-temporal satellite feature integration for mango orchard acreage estimation using machine learning algorithms approach

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-09-22
• Authors: V. Raju, Yogesh A. Garde, Dr. V. S. Thorat, V.T. Shinde, Nitin Varshney, Alok Shrivastava, A. P. Chaudhary
• DOI: 10.1038/s41598-026-65828-3

Research Groups

• Department of Agricultural Statistics, N.M. College of Agriculture, Navsari Agricultural University, Navsari, Gujarat, India
• ICAR-Indian Agricultural Statistics Research Institute, Pusa, New Delhi, India
• ICAR - National Soybean Research Institute, Indore, Madhya Pradesh, India
• Department of Agricultural Engineering, N.M. College of Agriculture, Navsari Agricultural University, Navsari, Gujarat, India

Short Summary

This study developed a multi-temporal remote sensing framework integrating Sentinel-1 SAR, Sentinel-2 optical imagery, and machine learning to accurately estimate mango orchard acreage in Navsari District, India, achieving 99.80% overall accuracy with Random Forest and a 5.05% estimation error compared to official statistics. It also analyzed 23-year trends in mango cultivation, finding a linear increase in area and cubic variability in production.

Objective

• To analyze trends in mango area and production over 23 years using polynomial regression models.
• To develop a multi-temporal Sentinel-1 and Sentinel-2 feature stack integrating spectral, SAR, vegetation index, and texture information.
• To evaluate the performance of Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM), and Multinomial Logistic Regression (MNLR) algorithms for mango orchard classification.
• To estimate mango orchard acreage and compare model outputs with official statistics.

Study Configuration

• Spatial Scale: Navsari District, South Gujarat, India (approximately 2,211 square kilometers).
• Temporal Scale:
  • Trend analysis: 23 years (2001–02 to 2023–24).
  • Remote sensing data acquisition: Multi-temporal observations from November 2023 to March 2024.

Methodology and Data

• Models used:
  • Polynomial regression models (Linear, Quadratic, Cubic, Exponential) for temporal trend analysis.
  • Machine learning algorithms for classification: Random Forest (RF), XGBoost (XGB), Support Vector Machine (SVM), Multinomial Logistic Regression (MNLR).
• Data sources:
  • Satellite imagery: Multi-temporal Sentinel-2 Level-2A optical imagery (10 meter spatial resolution), Sentinel-1 Ground Range Detected (GRD) SAR imagery (VV polarization, VH polarization, VV/VH ratio).
  • Derived features: Vegetation indices (NDVI, GNDVI, NDRE, SAVI, EVI, NDMI), texture metrics (contrast, entropy, variance, inverse difference moment from Gray Level Co-occurrence Matrix), temporal difference features (NDVI temporal difference, EVI temporal difference).
  • Ground truth data: 190 ground reference points for mango and sapota orchards, and other land-use/crop classes (Mango, Sapota, Sugarcane, Rice, Forest, Mixed plantation, Coconut, Water bodies, Barren land, Pond/Fish Pond) collected via field surveys (2023-24) and visual interpretation.
  • Historical data: Mango area (hectares) and production (tonnes) for 23 years (2001–02 to 2023–24) from Directorate of Horticulture, Govt. of Gujarat, and Directorate of Economics and Statistics, Govt. of Gujarat.
  • Official statistics: Horticultural statistics (2022-23) for validation.

Main Results

• Temporal Trends: Mango cultivation area showed a consistent upward trend, best represented by a linear regression model (Adjusted R² = 0.969). Mango production exhibited greater variability, best captured by a cubic regression model (Adjusted R² = 0.634), indicating climatic and seasonal influences.
• Classification Performance: Random Forest (RF) achieved the highest classification performance with an Overall Accuracy of 99.80% and a Kappa coefficient of 0.997. Support Vector Machine (SVM) followed with 99.30% Overall Accuracy, while Multinomial Logistic Regression (MNLR) achieved 98.21%. XGBoost performed poorly with an Overall Accuracy of 9.20%.
• Key Features for Classification: SWIR bands (B11, B12), red-edge indices (NDRE), and temporal vegetation dynamics were identified as the most influential variables for mango orchard discrimination.
• Acreage Estimation: The RF model estimated mango orchard area at 36,099.94 hectares, showing the closest agreement with official horticultural statistics (34,363 hectares) with an estimation error of 5.05%. SVM and MNLR overestimated orchard extent by 16.03% and 43.37%, respectively.

Contributions

• Developed a novel phenology-guided multi-temporal classification framework integrating Sentinel-1 SAR, Sentinel-2 optical imagery, vegetation indices, and texture features with machine learning for accurate mango orchard acreage estimation in tropical horticultural regions.
• Demonstrated the superior performance of Random Forest for accurate mango orchard mapping compared to XGBoost, SVM, and MNLR, achieving high accuracy and low estimation error against official statistics.
• Provided a reliable and scalable methodology for operational horticultural monitoring, crop inventory generation, and precision agricultural planning, particularly in environments with spectral confusion and climate-induced variability.
• Utilized freely available satellite datasets and cloud-based processing (Google Earth Engine), enhancing reproducibility and applicability for large agricultural regions and resource-constrained environments.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Raju2026Temporal,
  author = {Raju, V. and Garde, Yogesh A. and Thorat,  Dr. V. S. and Shinde, V.T. and Varshney, Nitin and Shrivastava, Alok and Chaudhary, A. P.},
  title = {Temporal trend analysis and multi-temporal satellite feature integration for mango orchard acreage estimation using machine learning algorithms approach},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-65828-3},
  url = {https://doi.org/10.1038/s41598-026-65828-3}
}