Proactive Climate Risk Intelligence, Green Adaptation Financing, and Verification Command Center for Rural Microfinance in India
Umbrella enables Microfinance Institutions (MFIs) like Satin Creditcare to forecast, quantify, and preemptively manage agricultural climate risks across rural borrower clusters while strictly maintaining the fundamental decoupling axiom:
Umbrella operationalizes the Economics of Climate Adaptation (ECA) framework across a continuous closed-loop adaptation flywheel:
┌───────────────────────────────────────────────────────────────────────────────────────────────────┐
│ UMBRELLA COMPLETE ADAPTATION FLYWHEEL │
│ │
│ [Climate Risk Intelligence] (Open-Meteo, ERA5 reanalysis, CHIRPS 30-year climatology) │
│ │ │
│ ▼ │
│ [Portfolio Exposure Scoring] (Deterministic Synthetic MFI portfolio capital & borrowers) │
│ │ │
│ ▼ │
│ [Adaptation Recommendation Engine] (Rule-based suitability ranking for 6 resilience assets) │
│ │ │
│ ▼ │
│ [Green Finance Simulator] (Reducing-balance EMI, operational payback calculations) │
│ │ │
│ ▼ │
│ [Human Credit Appraisal] (MANDATORY human decision; automated lending prohibited) │
│ │ │
│ ▼ │
│ [Disbursement & Asset Spawning] (Creates tracked ResilienceAsset record) │
│ │ │
│ ▼ │
│ [Field Verification Subsystem] (Magic bytes, SHA-256 duplicate check, Haversine geofence) │
│ │ │
│ ▼ │
│ [Dual-Track Impact Engine] (1. Physical Loss Averted | 2. Activity Emissions Avoided) │
│ │ │
│ ▼ │
│ [Append-Only Audit Trail] (Complete cryptographic lifecycle provenance) │
└───────────────────────────────────────────────────────────────────────────────────────────────────┘
- Physical Environmental Hazard is computed purely from atmospheric and hydrological drivers. Physical flood hazard never increases simply because a lender has more loans in a village.
- Microfinance Portfolio Exposure tracks institutional capital volume (INR), active borrowers, and Joint Liability Groups (JLGs) in a geography.
-
Operational Priority Index transparently combines hazard (
$60%$ ) and exposure ($40%$ ) to guide lender attention and field operations. - Decision Support, NOT Automated Lending: Automated risk checks advise; certified human credit and operations officers decide. Umbrella never autonomously approves, rejects, restructures, or disburses loans.
-
Darbhanga District Pilot: Primary case study in flood-prone North Bihar (Bagmati, Kamla-Balan, and Adhwara basins) with official OpenStreetMap boundary (
Relation: 1568263) and 10 operational clusters. -
Flood Hazard Model v1.0: Deterministic composite scoring (
$0-100$ ) combining Forecast Accumulation ($35%$ ), Burst Intensity ($25%$ ), Soil Saturation ($15%$ ), Climatological Anomaly ($15%$ ), and Terrain Susceptibility ($10%$ ). -
Retrospective Disaster Replay: Forensic chronological replay (
$T-7$ to$T+3$ ) of the severe July 2020 North Bihar Flood with strict anti-leakage time bounds. -
Ground-Truth Observational Validation: Correlates model hazard with Central Water Commission (CWC) river gauges (Hayaghat
$+2.14\text{ m}$ above danger level) and Copernicus Sentinel-1 SAR acquisition passes (Relative Orbits 121 & 48). - Lender Command Center UI: Next.js 14 dashboard featuring interactive Leaflet maps, 5-component hazard explainability progress bars, sortable portfolio tables, and human authorization modals.
-
Green Adaptation Loan Simulator: Reducing-balance amortization calculator (
$EMI = P \frac{r(1+r)^n}{(1+r)^n - 1}$ ) with cost breakdowns, operating savings, and simple payback metrics. -
Tamper-Resistant Field Verification: Mobile-friendly inspection portal validating JPEG/PNG/WebP magic bytes, max 5MB cap, SHA-256 duplicate image detection across all assets, Haversine geofence deviation (
$\le 500\text{ m}$ PASS,$500-2000\text{ m}$ REVIEW,$> 2000\text{ m}$ FLAG), and physical checklists. -
End-to-End Asset Traceability: Full chronological audit trail (
/assets/[id]) connecting climate trigger$\rightarrow$ loan application$\rightarrow$ human approval$\rightarrow$ disbursement$\rightarrow$ installation$\rightarrow$ inspection sign-off$\rightarrow$ realized impact. -
Dual-Track Climate Impact Engine: Quantifies physical resilience (grain protected, flood loss averted, households safeguarded) alongside activity-based emissions avoided proxies via
UNFCCC AMS-I.AandFAO Post-Harvest (2021). - Illustrative Carbon Sensitivity Tool: Interactive scenario simulator ($5–$50/tCO2e) with non-negotiable uncertified proxy disclaimers for blended finance and donor concession modeling.
- Backend: Python 3.10+, FastAPI, Pydantic v2, Starlette, Uvicorn, AnyIO, Pytest
- Frontend: Next.js 14 (App Router), React 18, TypeScript, Tailwind CSS, Leaflet, Lucide Icons
- Data Integrations: Open-Meteo Weather Forecast API (ECMWF IFS & DWD ICON), ECMWF ERA5 & ERA5-Land Reanalysis, CHIRPS v2.0 Climatology, Copernicus Sentinel-1 SAR Metadata, CWC River Telemetry
- Standards & Methodologies: UNFCCC AMS-I.A, FAO Post-Harvest (2021), IPCC EFDB, CEA India CO2 Baseline Database v19
UMBRELLA/
├── src/umbrella/ # Python backend application
│ ├── adapters/ # Weather, climate, and portfolio data providers
│ ├── config/ # Geography and district configurations
│ ├── engine/ # Hazard, exposure, impact, recommendation, financing, verification, FSM
│ │ ├── catalog.py # Resilience interventions, green products, and methodologies
│ │ ├── adaptation_rec.py # Rule-based adaptation recommendation engine
│ │ ├── financing.py # Reducing-balance amortization & payback calculator
│ │ ├── state_machine.py # Deterministic FSM lifecycle validator
│ │ ├── verification.py # Magic bytes, SHA-256 duplicate check, Haversine geofence
│ │ ├── impact_engine.py # Dual-track impact & carbon scenario engine
│ │ ├── audit.py # Append-only audit trail logger
│ │ └── flywheel_store.py # In-memory lifecycle persistence & demo seeder
│ ├── schemas/ # Pydantic v2 domain schemas with strict decoupling
│ │ ├── hazard.py # Pure physical hazard models
│ │ ├── portfolio.py # Institutional portfolio exposure models
│ │ ├── impact.py # Operational priority & MFI recommendations
│ │ ├── resilience.py # Green finance, verification, and impact contracts
│ │ └── replay.py # Forensic disaster replay models
│ ├── validators/ # Sentinel-1 and observational validators
│ ├── api.py # FastAPI REST API with CORS middleware
│ └── pipeline.py # Central orchestration pipeline
├── frontend/ # Next.js 14 Command Center web interface
│ ├── app/ # App Router pages (/dashboard, /green-finance, /field-officer, /impact, etc.)
│ ├── components/ # Navbar, Sidebar, LeafletMap, HazardExplainability, Drawer, etc.
│ ├── lib/ # Typed API client, domain types, formatting utils
│ └── scripts/ # Automated frontend verification test scripts
├── data/ # Geospatial and historical event data
│ ├── events/ # July 2020 flood event registry
│ ├── geography/bihar/darbhanga/ # District boundary & operational clusters GeoJSON
│ └── weather/ # ERA5 retrospective reanalysis cache
├── docs/ # Comprehensive technical and methodology documentation
├── tests/ # Pytest test suite (91 tests, 100% pass rate)
├── .gitignore # Production exclusions (secrets, caches, build output)
├── .env.example # Environment configuration template
└── pyproject.toml # Python packaging and test configuration
- Python 3.10+ (tested on Python 3.14)
- Node.js 18+ (tested on Node v24.15)
# Start FastAPI backend on port 8000
python -m uvicorn umbrella.api:app --host 127.0.0.1 --port 8000API Documentation (Swagger UI): http://127.0.0.1:8000/docs
# Navigate to frontend directory
cd frontend
# Install dependencies (first time only)
npm install
# Start Next.js production server on port 3000
npm run start
# (or for development: npm run dev)Access the Command Center at: http://localhost:3000/dashboard
# Run backend pytest suite (91 passing tests, 100% pass rate)
python -m pytest tests -q
# Run frontend build and component structure verification (8/8 passing suites)
node frontend/scripts/test-frontend.mjs
# Run live E2E REST API integration test across all endpoints
node frontend/scripts/test-api-e2e.mjsEvery data point in Umbrella is marked with explicit provenance:
-
LIVE: Real-time weather forecasts fetched directly from Open-Meteo. -
REANALYSIS: Retrospective historical weather from ECMWF ERA5 / ERA5-Land. -
OBSERVATION: Satellite SAR metadata (Sentinel-1) or CWC river gauge records. -
DERIVED: Deterministic models (Flood Hazard Model v1.0, Priority Index, Amortization). -
LOCAL_DATASET: Static climatological normal tables (CHIRPS 30-year normal). -
SOURCED: Peer-reviewed scientific and regulatory benchmarks (IPCC EFDB, CEA India Grid Baseline, ICAR post-harvest loss). -
SYNTHETIC: Generated demonstration data (all borrower accounts, loans, and demo assets). -
DEMO_ASSUMPTION: Explicit sensitivity parameters (e.g., $15/tCO2e illustrative carbon price, ₹83/$1 FX).
- Decision Support, NOT Automated Lending: Umbrella is strictly a decision-support and risk-intelligence platform. It never autonomously approves, rejects, restructures, or disburses loans. All credit decisions and verification sign-offs require human officer appraisal.
- Synthetic Portfolio Disclaimer: All borrower names, counts, loan balances, and Joint Liability Groups (JLGs) are deterministic synthetic demo figures. They do not represent actual borrower records, proprietary MFI portfolios, or Satin Creditcare client data.
- Credit Default Disclaimer: Umbrella evaluates capital exposed to geographic climate hazards. It does not predict individual borrower creditworthiness, credit scores, or probability of default.
-
Historical Replay Disclaimer: Retrospective replay mode uses ECMWF ERA5 reanalysis and adheres strictly to anti-leakage causality: for any historical snapshot date
$T$ , zero meteorological data$> T$ is accessible. -
Carbon Metrics & Non-Credit Disclaimer: Avoided emissions are activity-based operational proxies (
ESTIMATED_EMISSIONS_AVOIDED) derived fromUNFCCC AMS-I.AandFAO Post-Harvest (2021). They do not constitute certified carbon credits, voluntary carbon units (VCUs), or guaranteed revenue.
- Final Audit & Hardening Report
- Claim Register & Scientific Integrity Matrix
- Historical Replay Verification & Anti-Leakage Audit
- Demo Dataset & Deterministic Seed Catalog
- Master 3-Minute Presentation Script & Flow
- Judge & Evaluator Q&A Playbook (25 Answers)
- Final System Architecture Specification
- Executive Feature Matrix
- Product Truth Table (Evaluator Guide)
- Upstream Source & Data Register
- Demo Resilience & Offline Playbook
- System Architecture
- Green Finance & Amortization Engine
- Physical Asset Verification Protocol
- Climate Impact & Emissions Avoidance Methodology
- Lifecycle State Machine
- Data Provenance & Modality Register
- Flood Hazard Model v1.0
- Pilot Selection (Darbhanga, Bihar)
- Historical Replay & Ground-Truth Sentinel-1 Validation
- Status: Working, Tested, Demo-Hardened Hackathon MVP (Closed-Loop Climate Adaptation Flywheel).
- Backend: Verified FastAPI REST service with 98 passing tests (100% pass rate) across 14 test suites with 86% statement coverage.
- Frontend: Next.js 14 App Router application with 11 production routes compiled, zero TypeScript errors, mobile-friendly field officer interface, and interactive impact dashboards.
- Resilience: Full deterministic offline mock fallback support with transparent provenance badging and global status bar monitoring.
- Pilot Geography: Darbhanga District, Bihar (10 operational clusters, July 2020 North Bihar Flood event).