Bias Correction: Localize WRF to Weather Stations

Use historical weather station observations to correct WRF model bias locally.

Basic Localization

# ⚠️  Currently WRF + hourly temperature only
data = (cd
    .activity_id("WRF")
    .institution_id("UCLA")      # Specify WRF producer
    .variable("t2")              # Hourly 2m temperature
    .table_id("1hr")             # Must be hourly
    .processes({
        "bias_adjust_model_to_station": {
            "stations": ["KSAC", "KSFO", "KLAX"]
        }
    })
    .get())

# Data now bias-corrected to observations

Available Weather Stations

# List all available weather stations
cd.show_station_options()  # Returns station codes (ICAO format)

# Use with clip to find nearby station
data = (cd
    .processes({
        "bias_adjust_model_to_station": {
            "stations": ["KSFO"]  # San Francisco airport
        }
    })
    .get())

How Bias Correction Works

• Training: Uses historical station observations (1981-2010 baseline)
• Method: Quantile delta mapping (preserves model trends while matching observations)
• Result: WRF temperature distribution matches local observations
• Benefit: Reduces systematic bias for climate projections

Limitations

Currently available for: - ✅ WRF data only (not LOCA2 statistical downscaling) - ✅ Hourly temperature (t2) only
- ✅ HadISD weather stations (~600 globally, ~200 in western US)

Why these limitations?

Bias correction requires: - High-frequency observations (hourly) to capture temperature variability that drives quantile mapping - WRF hourly data because WRF's fast-varying dynamics need point-wise calibration - LOCA2 is already bias-corrected by design using quantile mapping to observations during downscaling (no bias correction needed) - Weather station coverage — only HadISD provides consistent historical hourly data

For other scenarios: - Use direct model output (LOCA2 is already bias-corrected) - Implement alternative bias correction method for daily/monthly aggregates - Contact support for custom approaches

---