Rice Yield Mapping via Sentinel-1 , SAR-Based Crop Yield Estimation Under Indian Agricultural Conditions
Budget: ₹1,500 – ₹12,500 INR
I need a complete, end-to-end workflow that turns Sentinel-1 SAR scenes into reliable, field-level yield estimates for rice grown across Tamil Nadu. The goal is to generate season-long yield maps and a final quantitative report that I can hand straight to agronomists and policy planners.
Here’s what I expect you to handle:
• Source and preprocess the full Sentinel-1 stack for the latest kharif season (orbit selection, radiometric calibration, speckle filtering, terrain correction and conversion to σ°).
• Extract backscatter-based indicators that are proven proxies for crop biophysical parameters, then build and train a predictive model—statistical, machine-learning or hybrid—suited to rice phenology in South-Indian conditions.
• Validate the model against whatever ground-truth you can access (government statistics, published trials or any in-situ data I can help secure) and report the accuracy clearly.
• Deliver georeferenced raster yield maps (GeoTIFF), the trained model, a concise methodology document and a short slide deck summarising findings and limitations.
Feel free to code in Python with SNAP, Google Earth Engine, QGIS or any equivalent SAR-friendly toolchain, as long as the process is fully reproducible and I receive the scripts/notebooks. If you have prior work on rice or humid-tropic crops, let me know; that experience will be a big plus.
needs to workable in google collab. requirements can be negotiable
Here’s what I expect you to handle:
• Source and preprocess the full Sentinel-1 stack for the latest kharif season (orbit selection, radiometric calibration, speckle filtering, terrain correction and conversion to σ°).
• Extract backscatter-based indicators that are proven proxies for crop biophysical parameters, then build and train a predictive model—statistical, machine-learning or hybrid—suited to rice phenology in South-Indian conditions.
• Validate the model against whatever ground-truth you can access (government statistics, published trials or any in-situ data I can help secure) and report the accuracy clearly.
• Deliver georeferenced raster yield maps (GeoTIFF), the trained model, a concise methodology document and a short slide deck summarising findings and limitations.
Feel free to code in Python with SNAP, Google Earth Engine, QGIS or any equivalent SAR-friendly toolchain, as long as the process is fully reproducible and I receive the scripts/notebooks. If you have prior work on rice or humid-tropic crops, let me know; that experience will be a big plus.
needs to workable in google collab. requirements can be negotiable