National-Scale Orchard Mapping and Yield Estimation in Pakistan Using Object-Based Random Forest and Multisource...

Area of Science:

• Agricultural Remote Sensing
• Geospatial Analysis
• Food Security Monitoring

Background:

• Pakistan lacks comprehensive geospatial data for fruit orchards, hindering precision horticulture and food security efforts.
• Existing remote sensing methods struggle with accurate delineation and yield estimation at national and sub-national levels.

Purpose of the Study:

• To develop and validate a national-scale, object-based Random Forest (RF) framework for orchard delineation and yield estimation.
• To integrate multi-temporal Sentinel-2 and high-resolution Pakistan Remote Sensing Satellite-1 (PRSS-1) data for improved accuracy.
• To establish a scalable and transferable methodology for fruit orchard mapping and yield forecasting.

Main Methods:

• Utilized Google Earth Engine (GEE) for processing multi-temporal Sentinel-2 imagery.
• Implemented an object-based Random Forest (RF) classifier for orchard delineation, comparing its performance against Support Vector Machines and Gradient Boosting Decision Trees.
• Integrated RF with Object-Based Image Analysis (OBIA) using PRSS-1 data to enhance boundary precision.
• Developed regression-based models for yield estimation using vegetation index aggregations.

Main Results:

• The RF classifier achieved the highest accuracy (OA = 79.0%, κ = 0.78) on Sentinel-2 data, significantly outperforming other classifiers.
• Integrating RF with OBIA on PRSS-1 data further improved accuracy (OA = 92.6%, κ = 0.89) and boundary precision (IoU increased from 0.71 to 0.86).
• Mean and median vegetation index aggregations yielded the most stable yield predictions (R² = 0.77-0.79), with significantly lower errors compared to extreme-value models.

Conclusions:

• The developed multisensory framework provides a scalable, transferable, and operationally viable solution for orchard mapping and yield forecasting in Pakistan.
• This approach demonstrates the strategic value of national satellite assets for food security monitoring, particularly in data-scarce regions.
• The study establishes a robust methodology for creating essential geospatial inventories for precision horticulture and agricultural resource management.