Carbon Credit Marketplace with Multi-Modal Verification for Smallholder Agroforestry – Satellite NDVI, Ground-Photo AI and Soil Carbon

Uncategorized

Authors: Rudramurthy V C, Chiranthan S V, Dheemanth J Nirvani, Samarth Kumble

Abstract: Smallholder agroforestry plots rarely market, because the measurement, reporting and verification (MRV) step that backs each credit still rests on field visits that are too expensive to run on a one- or two-acre holding, in many respects; indeed, the platform is a FastAPI backend, a React web app and an Expo mobile app, with 139 automated tests covering it, to a significant degree. A rule-based fraud engine, role-separated farmer, auditor and buyer portals. The check is fusing satellite vegetation data (NDVI computed from real Sentinel-2 imagery inside the boundary the farmer draws), a ground-photo classifier (MobileNetV3-Small, 98.4% accuracy on 1,420 held-out real photos) and a soil organic carbon (SOC) reading into one weighted score, 0.40 NDVI + 0.35 photo + 0.25 SOC. A score of 75 or above is permitted, 55 to 75 is routed to a human auditor; indeed, we built a working prototype of a carbon-credit marketplace in which a plot can mint credits only after it clears an automated, explainable three-part check. And anything under 55 is rejected; if any single piece of indication is missing the plot stays pending, and no value is ever invented, and an Ethereum-style ledger (Ganache) are logging each credit from issuance, through the transfer to retirement. In an end-to-end run on live data, a real areca plot in Shivamogga, Karnataka created an NDVI of 0.794 from a Sentinel-2 scene dated 27 September 2026, was authorized at 84.9/100, minted a 20 tCO2 credit; and thus, as, well, that credit was then bought and retired.

DOI: https://doi.org/10.5281/zenodo.24391807

× How can I help you?