Home / Methodology

The science underneath the platform.

REALM Krishi makes scientifically defensible claims. This page documents the underlying methodologies — data sources, processing pipelines, MRV protocols, and validation procedures — so that government, investor and registry counterparties can evaluate the platform on technical grounds.

Satellite Crop Intelligence

Sentinel-2 + Landsat-8/9 time-series processing.

All field-level crop intelligence is derived from publicly available Copernicus Sentinel-2 (10–30 m, 5-day revisit) and USGS Landsat-8/9 (30 m, 16-day revisit) optical imagery, processed through a standardised pipeline.

IndexBandsUsed For
NDVI — Normalised Difference Vegetation IndexNIR, Red (B8, B4)Crop vigour, biomass proxy, season tracking
NDMI — Normalised Difference Moisture IndexNIR, SWIR (B8, B11)Canopy water content, irrigation timing
NDWI — Normalised Difference Water IndexGreen, NIR (B3, B8)Surface water, waterlogging detection
SAVI — Soil-Adjusted Vegetation IndexNIR, Red (B8, B4) + LSparse-canopy / early-season vigour
EVI — Enhanced Vegetation IndexNIR, Red, Blue (B8, B4, B2)Dense-canopy & high-biomass crops

Processing pipeline

  1. Acquisition: Daily ingestion from Sentinel Hub and Copernicus Open Access Hub for Maharashtra and Karnataka tiles.
  2. Cloud masking: Sentinel-2 Scene Classification Layer (SCL) applied; pixels classified as cloud, shadow, snow or cirrus are excluded.
  3. Spatial aggregation: Per-plot statistics (mean, median, p10/p90, standard deviation) computed using geo-fenced FPO boundaries.
  4. Temporal smoothing: Savitzky–Golay filter applied to per-plot NDVI time series to suppress residual noise.
  5. Baselining: 5-year historic baseline computed per plot; anomaly score is current value vs. baseline percentile.
  6. Alerting: Anomalies beyond ±1.5 σ trigger language-localised farmer alerts via WhatsApp and dashboard surfacing.
Carbon MRV

Verra VM0042 + Gold Standard pathway.

REALM Krishi's carbon programme is designed to qualify under two leading registries. Final methodology selection depends on the activity mix, baseline data availability and partner preference per FPO. Both pathways are technically supported.

LayerApproachVerification
MethodologyVerra VM0042 (ALM) primary; Gold Standard SDG Impact for Smallholder Agriculture secondaryThird-party VVB-accredited
BaselineRemote-sensing carbon stock estimation (NDVI, EVI, SAR biomass) + farmer historic activity surveyStratified ground-truth sampling
Activity trackingFPO-aggregated digital activity log (cover crop, residue retention, biochar, agroforestry, reduced tillage)Random plot audits + satellite verification
QuantificationIPCC Tier 2 emission factors blended with locally calibrated soil-carbon modelsAnnual MRV reporting cycle
IssuanceVCU (Verra) or VER (Gold Standard) — held in FPO-controlled buffer pool until verifiedPublic registry retirement

Smallholder aggregation principle

Most carbon registries impose minimum-project-size thresholds that exclude individual smallholders. REALM Krishi aggregates plots at the FPO level — typically 500–10,000 farmers — so the project meets registry economics while every contributing farmer retains traceable, individually attributable activity data and revenue share.

Farmer revenue split

Target split (to be confirmed per pilot MoU): 70% farmer, 20% FPO operating cost, 10% platform fee. Verification, registry and trader margins are paid out of the platform fee. All splits are published per project on the public dashboard.

Weather Intelligence

IMD-primary, ensemble-blended forecasting.

Indian smallholders need forecasts they can trust enough to act on. We treat IMD as the source of record — its products are calibrated to Indian climatology — and use global ensembles to communicate forecast confidence.

SourceResolutionUse
IMD GFS / NCMRWF~12 km, 6-hourlyPrimary 7-day forecast
ECMWF IFS~9 km, 6-hourlyEnsemble confidence + week 2
NOAA GFS~13 km, 3-hourlyConvective / severe weather signal
IMD radar / nowcast1 km, 10-min0–6 hour storm alerts

Forecast packaging

Raw model output is translated into farmer-actionable advisories: sow / don't sow this week, irrigate today, spray window open, harvest now to avoid forecast rain, etc. Advisory wording is generated by the REALM GPT Lab platform, locked to deterministic templates per crop and validated against agronomist-reviewed criteria.

AI Advisory

Multilingual reasoning, agronomist-validated outputs.

Knowledge sources

  • ICAR (Indian Council of Agricultural Research) crop manuals and package-of-practices
  • State Department of Agriculture extension materials — Maharashtra, Karnataka
  • Krishi Vigyan Kendra (KVK) advisory archives
  • Victorian Department of Energy, Environment and Climate Action on-farm trial data (where licensable)
  • FAO and CGIAR crop-system references for cross-validation

Languages supported at launch

  • Hindi — pan-India default
  • Marathi — Maharashtra primary
  • Kannada — Karnataka primary
  • English — institutional and partner-facing

Safety & guardrails

The REALM GPT Lab platform applies layered guardrails: a domain-restricted knowledge base, deterministic templates for time-bound advisories (e.g. dosing recommendations are not generated freeform — they're surfaced from validated tables), and a human-in-the-loop escalation pathway routing low-confidence answers to a district agronomist within four working hours.

Data Governance

Compliant by construction. Sovereign by default.

REALM Krishi processes personal and geospatial data covered by India's Digital Personal Data Protection Act 2023 (DPDP) and, where Victorian partners are involved, by the Australian Privacy Principles. The platform is designed to satisfy both regimes simultaneously.

Core principles

  • Data residency: All farmer and field data hosted in ap-south-1 (Mumbai); no personal data exits India by default.
  • Consent: Multilingual opt-in at onboarding; granular toggles for advisory, carbon, dashboards, marketing.
  • Right of erasure: Self-service deletion via WhatsApp command and FPO dashboard; 30-day full removal SLA.
  • FPO data trusteeship: FPOs control aggregated data on behalf of members; institutional partners receive only aggregated layers unless individual consent is explicit.
  • Carbon revenue chain of custody: All carbon-related data logged immutably; auditor-accessible via secured registry interface.

Certification roadmap

SOC 2 Type II readiness in pilot Year 1; ISO 27001 certification targeted for Year 2 as district-scale deployment begins.

"Victoria offers a depth and breadth of expertise across the entire agricultural supply chain… an opportunity to share knowledge."
Victoria's India Strategy 2025-30, p.20 — Food & Agribusiness

Want the deep-dive?

The full technical whitepaper extends each of these sections with architecture diagrams, sample data flows, and the complete Vic Strategy alignment mapping.

Standards & Frameworks
  • — Verra VCS / VM0042
  • — Gold Standard SDG Impact Tools
  • — IPCC AR6 Tier 2
  • — India DPDP Act 2023
  • — UN SDGs 2, 5, 13, 15