LOAF (Local Observations and Atmospheric Forecasting)

Open source hyperlocal weather forecasting you can deploy yourself.

Why LOAF?

Standard weather forecasts operate on 3km grids. That resolution can't capture the wind patterns at your specific site—whether it's a backyard wind turbine, a fire-prone hillside, or a remote research station.

Recent ML research shows that fusing gridded forecasts with local sensor data via transformers can reduce prediction error by up to 80%. Commercial services like Tomorrow.io offer this, but they're proprietary and subscription-based. The academic code exists, but there's no practical way to go from "I have a Raspberry Pi" to "I have hyperlocal forecasts."

LOAF bridges that gap: open source hardware, open source models, no vendor lock-in. Build a sensor, train a model for your region, run inference locally.

Built on research from MIT Earth Intelligence Lab:

• Earth-Intelligence-Lab/LocalizedWeather
• Yang, Q., et al. (2024). Local Off-Grid Weather Forecasting with Multi-Modal Earth Observation Data

Architecture

flowchart TB
    subgraph Data Sources
        HRRR[NOAA HRRR<br/>3km gridded forecasts]
        MADIS[MADIS Stations<br/>Regional observations]
        LOCAL[Local Sensor<br/>Ultrasonic anemometer]
    end

    subgraph Data Pipeline
        FETCH[Data Fetcher<br/>Hourly HRRR download]
        ALIGN[Alignment<br/>Spatial/temporal sync]
    end

    subgraph Model
        SPATIAL[Spatial Encoder<br/>Grid transformer]
        STATION[Station Encoder<br/>Observation tokens]
        FUSION[Cross-Attention<br/>Fusion module]
        DECODER[Decoder<br/>Local prediction]
    end

    subgraph Output
        FORECAST[6-48hr Forecast<br/>Wind speed & direction]
        HA[Home Assistant<br/>Widget display]
    end

    HRRR --> FETCH
    FETCH --> ALIGN
    MADIS --> ALIGN
    LOCAL --> ALIGN

    ALIGN --> SPATIAL
    ALIGN --> STATION

    SPATIAL --> FUSION
    STATION --> FUSION
    FUSION --> DECODER

    DECODER --> FORECAST
    FORECAST --> HA

Data Flow

1. Gridded Forecasts: NOAA HRRR provides 3km resolution wind forecasts, downloaded hourly
2. Station Observations: MADIS network stations provide ground truth from the surrounding region
3. Local Sensor: DIY ultrasonic anemometer captures hyperlocal conditions
4. Fusion: Multi-modal transformer combines coarse forecasts with sparse observations
5. Prediction: Model outputs corrected forecasts for the exact sensor location

Hardware Stack

Component	Description
Sensor	DIY ultrasonic anemometer (40kHz transducers)
Logger	Raspberry Pi 4 with RS-485 interface
Power	20W solar panel + 12V battery
Enclosure	3D printed weatherproof housing

Model Architecture

The transformer architecture processes two input streams:

• Spatial encoder: Treats HRRR grid points as tokens, applies self-attention to capture regional patterns
• Station encoder: Each weather station becomes a token with positional encoding based on lat/lon
• Cross-attention fusion: Target location queries both encoders to aggregate relevant information
• Decoder: Predicts wind speed/direction at forecast horizons (6-48 hours)

Training uses historical HRRR forecasts paired with MADIS observations (2020-2024), validated against held-out local sensor data.

Deployment Architecture

┌─────────────────┐      ┌─────────────────────────────────────┐      ┌─────────────────┐
│  DIY Sensors    │ ──── │          Raspberry Pi               │ ──── │ Home Assistant  │
│  (Anemometer,   │      │  • Data storage                     │      │                 │
│   Temperature,  │      │  • ML forecast processing           │      │  Weather entity │
│   etc.)         │      │  • Local prediction generation      │      │  integrations   │
└─────────────────┘      └─────────────────────────────────────┘      └─────────────────┘

Physical Data Flow:

1. Sensors → Raspberry Pi: Local sensors connect to the Pi via a common sensors library, logging observations to local storage
2. Forecast Processing: The Pi runs the trained ML model, combining local sensor data with regional forecasts (HRRR/GFS) to generate hyperlocal predictions
3. Home Assistant Integration: Predictions are exposed as Home Assistant entities, enabling integration with automations, dashboards, and alerts