Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
177 lines (134 loc) · 6.1 KB

ARCHITECTURE.md

File metadata and controls

177 lines (134 loc) · 6.1 KB

OpenFarm Strategic Architecture

OpenFarm is an open, modular field intelligence platform that fuses satellite, weather, soil, and time to explain what is happening in a field — and why.

OpenFarm follows a 3-layer strategic architecture. Each layer has a distinct role: the Observation layer creates data gravity, the Intelligence layer is the moat, and the Delivery layer drives distribution.

┌─────────────────────────────────────────────────────────────────┐
│  Layer C — Delivery Surfaces                   (Distribution)  │
│  Map UI · Reports · API · Webhooks · MCP · Mobile scouting     │
├─────────────────────────────────────────────────────────────────┤
│  Layer B — Intelligence Engine                       (Moat)    │
│  Phenology · Anomaly detection · Stress signals · Yield        │
│  Risk models · Soil-derived insights · Explainability          │
├─────────────────────────────────────────────────────────────────┤
│  Layer A — Observation Infrastructure         (Data Gravity)   │
│  Satellite · Weather · Soil · Field boundaries · Sensors       │
└─────────────────────────────────────────────────────────────────┘

Layer A — Observation Infrastructure

"What do we observe about the field?"

Collects, standardizes, and stores raw signals about every field. Source-agnostic, reproducible, and extensible.

Satellite Intelligence

  • Sentinel-2 (initial), extensible to Landsat, Planet, SAR (Sentinel-1)
  • Raster ingestion pipelines (COG/STAC compliant)
  • Index computation: NDVI, EVI, SAVI, NDWI
  • Temporal stacking & versioning
  • Cloud masking & gap handling (future: fusion)

Weather Intelligence

  • Open-Meteo integration (global baseline)
  • Historical + forecast weather
  • Derived metrics: GDD (Growing Degree Days), ET₀, rainfall accumulation
  • Spatial interpolation at field level

Soil Intelligence

  • Global + regional datasets: SoilGrids (global), POLARIS (US), ESDAC (EU)
  • Depth-aware soil profiles (0–5, 5–15, 15–30, 30–60, 60–100, 100–200 cm)
  • Core properties: texture (sand/silt/clay), SOC, pH, CEC, bulk density, nitrogen proxies
  • Uncertainty retained as first-class signal

Field & Spatial Context

  • Field boundary detection (FTW deep learning model)
  • Manual & imported boundaries (GeoJSON/KML)
  • Terrain layers: elevation, slope, aspect
  • Spatial indexing (PostGIS)

Extensible Inputs (Future)

  • IoT sensors (soil moisture, EC, pH)
  • Farm operations data
  • External APIs (market, irrigation)

Output of Observation Layer

A standardized, queryable field data model:

Field = {
  geometry,
  time_series: { satellite, weather },
  soil_profile: { ... },
  terrain: { ... },
  metadata
}

Layer B — Intelligence Engine

"What does it mean?"

The heart of OpenFarm — where raw signals are converted into explainable, agronomically meaningful insights. This is the strategic differentiator.

Field Understanding Engine

  • Temporal analytics (trend, deviation, anomaly)
  • Field segmentation (zones based on variability)
  • Change detection (intra-season + inter-season)

Crop Intelligence

  • Crop type detection (future ML models)
  • Phenology tracking (stage detection from NDVI/EVI temporal curves)
  • Growth progression modeling

Stress & Risk Intelligence

  • Water stress signals (satellite + soil + weather)
  • Disease/pest risk models: weather + crop stage + anomaly signals
  • Nutrient deficiency indicators (proxy-based initially)

Soil-Derived Intelligence

  • Water holding capacity (derived via Rosetta PTF)
  • Drainage / runoff / leaching risk
  • Nutrient buffering capacity
  • Salinity / compaction susceptibility proxies
  • Soil × weather × crop interaction modeling

Predictive Intelligence (Progressive Build)

  • Yield estimation (multi-season learning)
  • Irrigation advisory
  • Fertilizer response zones (not prescriptions initially)
  • Carbon baseline & sequestration opportunity

Explainability & Trust Layer

Every insight carries:

  • Confidence scores per insight
  • Input signal breakdown: "This alert is based on NDVI drop + rainfall deficit + sandy soil"
  • Historical comparison: "This field behaved similarly in 2022"

Output of Intelligence Layer

Insight = {
  type: "stress" | "growth" | "risk" | "zone",
  confidence,
  drivers: [satellite, weather, soil],
  spatial_extent,
  temporal_context,
  explanation
}

Layer C — Delivery Surfaces

"How is it consumed?"

Ensures OpenFarm is not just a tool — but a platform others can build on.

Visual Interface (UI)

  • Interactive field maps (MapLibre + PMTiles)
  • Time-series charts (NDVI, rainfall, etc.) via ECharts
  • Layer toggles (satellite, soil, weather)
  • Zone visualization

Reports & Outputs

  • Field intelligence reports (shareable links)
  • Seasonal summaries
  • Export: GeoJSON, CSV, raster tiles, PDF

API-first Platform

  • REST API (/v1) with JWT + RBAC
  • Webhooks for alerts
  • Versioned public API with API keys (future)

Agent / LLM Interface (Future)

  • Natural language field queries
  • Model Context Protocol (MCP) server for AI agents
  • AI plugin ecosystem

Lightweight Field Tools

  • Scouting: geotagged observations with photo upload
  • Ground truth collection

Deployment Philosophy

  • Self-hostable (core identity)
  • Optional managed cloud (future)
  • Multi-tenant + enterprise-ready architecture

What OpenFarm IS

An open, modular field intelligence platform that fuses satellite, weather, soil, and time to explain what is happening in a field — and why.

What OpenFarm is NOT

  • ❌ Farm ERP
  • ❌ Task/operations manager
  • ❌ Generic ag marketplace