Station Network

Overview

The Real-Time Anomaly Detection system monitors 14 meteorological stations operated by the National Observatory of Athens (NOA). Understanding the spatial relationships between these stations is crucial for the dual-verification strategy.

Interactive Map

Open in new window

Map Legend

• Blue Markers: Weather station locations
• Red Lines: Neighbor connections (stations within 100km)
• Hover/Click: View station details

Station Details

Station List

Neighbor Selection Criteria

Distance Threshold: 100km

The system uses a 100km radius to define neighbors based on:

1. Weather Correlation: Meteorological phenomena typically affect regions within 50-150km
2. Station Density: Balances coverage vs. computational cost
3. False Positive Reduction: Sufficient neighbors for reliable correlation (typically 2-5 per station)

Distance Calculation

Haversine formula for geographic distance:

from math import radians, sin, cos, sqrt, atan2

def haversine_distance(lat1, lon1, lat2, lon2):
    """
    Calculate distance between two points on Earth.
    Returns distance in kilometers.
    """
    R = 6371  # Earth's radius in km

    dlat = radians(lat2 - lat1)
    dlon = radians(lon2 - lon1)

    a = sin(dlat/2)**2 + cos(radians(lat1)) * cos(radians(lat2)) * sin(dlon/2)**2
    c = 2 * atan2(sqrt(a), sqrt(1-a))

    return R * c

Neighbor Relationships

Example: uth_volos Station

Location: 39.3636°N, 22.9530°E

Neighbors (within 100km):

Use Case:

When uth_volos shows an anomalous temperature drop:

1. Check if volos (3km away) shows the same drop → If yes, likely weather
2. Check if zagora and larissa also show drops → Confirms regional weather event
3. If only uth_volos is anomalous → Likely device failure

Isolated Stations

Some stations have no neighbors within 100km:

• metsovo: High altitude, mountainous region
• preveza: Coastal, isolated location

For these stations:

• Spatial verification is skipped
• Classification relies solely on temporal detection
• Higher false positive rate expected

Network Topology

Connectivity Graph

graph LR
    A[uth_volos] --- B[volos]
    A --- C[zagora]
    A --- D[larissa]

    B --- C
    B --- D
    B --- E[anavra]

    C --- F[pelion]

    D --- G[karditsa]
    D --- H[domokos]
    D --- I[trikala]

    I --- J[pyli]
    I --- K[metsovo]

    K --- L[ioannina]

    L --- M[agrinio]

    M --- N[preveza]

Network Statistics

Spatial Correlation Patterns

Typical Weather Event

When a weather front passes through:

Station A: Temp 15°C → 10°C (↓5°C)
Station B: Temp 16°C → 11°C (↓5°C)
Station C: Temp 14°C → 9°C (↓5°C)

→ Correlation: 0.95 (all drop together)
→ Classification: Weather Event

Typical Device Failure

When one sensor malfunctions:

Station A: Temp 15°C → 99°C (sensor error)
Station B: Temp 16°C → 16°C (stable)
Station C: Temp 14°C → 14°C (stable)

→ Correlation: -0.05 (no pattern)
→ Classification: Device Failure

Edge Case: Local Phenomenon

Microclimates can cause legitimate but isolated anomalies:

Station A (coastal): Temp 20°C → 15°C (sea breeze)
Station B (inland): Temp 22°C → 22°C (stable)
Station C (inland): Temp 21°C → 21°C (stable)

→ Correlation: 0.35 (weak)
→ Classification: Suspected (requires manual review)

Dynamic Neighbor Selection

Current Implementation

Static: All neighbor relationships are pre-computed based on GPS coordinates.

Advantage: Fast lookup, no runtime overhead.

Future Enhancement

Dynamic: Adjust neighbor weights based on:

1. Historical Correlation: Stations that historically correlate better get higher weight
2. Elevation Similarity: Stations at similar altitude may correlate better
3. Wind Direction: Use upstream stations when wind data is available
4. Seasonal Adjustment: Different neighbor sets for summer vs. winter

Pseudocode:

def get_dynamic_neighbors(station, current_conditions):
    candidates = get_all_neighbors_within_radius(station, 100)

    # Weight by historical correlation
    for n in candidates:
        n.weight = historical_correlation(station, n)

    # Adjust for elevation
    for n in candidates:
        elev_diff = abs(station.elevation - n.elevation)
        if elev_diff > 500:  # 500m difference
            n.weight *= 0.5  # Reduce weight

    # Adjust for wind direction (if available)
    if current_conditions.wind_direction:
        for n in candidates:
            if is_upstream(n, station, current_conditions.wind_direction):
                n.weight *= 1.5  # Increase weight

    # Return top N weighted neighbors
    return sorted(candidates, key=lambda x: x.weight, reverse=True)[:5]

Map Generation

The interactive map is generated using the generate_map.py script:

# Generate or update the station network map
python generate_map.py

Output: spatial_network_map.html

Technology: Folium - Python wrapper for Leaflet.js

Data Source: Station GPS coordinates from NOA API

Customization

Edit the map generation parameters in generate_map.py:

# Configuration
NEIGHBOR_RADIUS_KM = 100
MAP_CENTER = [39.0, 22.0]  # Central Greece
MAP_ZOOM = 8
LINE_COLOR = 'red'
LINE_WEIGHT = 2
MARKER_COLOR = 'blue'