Remote Sensing–Based Monitoring and Yield Forecasting

Project Overview

This FAO-funded initiative established a replicable satellite-based crop monitoring system for the Gezira wheat belt. Support Vector Machine (SVM) classifiers were trained on multi-temporal Sentinel-2 imagery (20 bands, 6 seasonal composites) using 600+ GPS-verified field training samples. Wheat yield forecasting combined WaPOR NPP biomass estimates with a harvest index calibrated against historical yield records from the Gezira Board, delivering pre-harvest estimates 8 weeks before official reports.

Methodology

1. 01SVM multi-class crop classification on Sentinel-2 composites (10 & 20 m bands)
2. 02Training sample collection using ODK Collect mobile field surveys (600+ GPS points)
3. 03Accuracy assessment using Olofsson (2014) area-adjusted estimators — 94.2% overall accuracy
4. 04WaPOR NPP × harvest index yield estimation model development and calibration
5. 05Time-series NDVI phenology profiling per administrative block
6. 06R (tidyverse, ggplot2) statistical analysis and visualization

Key Outputs

• Pre-harvest wheat yield forecast model (validated within 7% of actual)
• 15% improvement in crop area monitoring accuracy over previous method
• 9% water productivity gain identified through spatial analysis
• GeoAccuRate QGIS plugin released implementing Olofsson accuracy assessment
• Full technical report submitted to FAO Sudan country office

15% monitoring accuracy improvement, 9% productivity gain