climateclaw skill

When to use

• User asks to fetch historical temperature, precipitation, or wind data for a location
• User wants a warming/cooling trend over a period
• User wants anomalies compared to a historical baseline
• User asks for mean climate statistics

When NOT to use

• Real-time or forecast weather: use weather / weathercli skill

Functions

fetch_historical(lat, lon, start, end, variable) — pull data from Open-Meteo

No API key required. Returns dates, values (numpy array), unit.

Supported variables:

• temperature_2m_mean / temperature_2m_max / temperature_2m_min
• precipitation_sum, rain_sum, snowfall_sum
• windspeed_10m_max
• et0_fao_evapotranspiration, shortwave_radiation_sum

compute_mean(data) — arithmetic mean

compute_anomaly(data, baseline) — departure from baseline mean

detect_trend(data) — linear trend via OLS -> slope, r², p-value

Quick examples

# Fetch + mean temperature for Hamburg, last 5 years
python3 -c "
import numpy as np
from climateclaw import fetch_historical, compute_mean
d = fetch_historical(53.5753, 10.0153, '2019-01-01', '2023-12-31')
vals = d['values'][~np.isnan(d['values'])]
print(f'{compute_mean(vals):.2f} {d[\"unit\"]}')
"

# Fetch + trend (annual means) for Paris, 20 years
python3 -c "
import numpy as np, json
from climateclaw import fetch_historical, detect_trend
d = fetch_historical(48.8566, 2.3522, '2004-01-01', '2023-12-31')
dates, vals = d['dates'], d['values']
years = sorted(set(x[:4] for x in dates))
annual = [np.nanmean([v for dt, v in zip(dates, vals) if dt.startswith(y)]) for y in years]
print(json.dumps(detect_trend(annual), indent=2))
"

# Anomaly vs 30-year WMO baseline for Berlin
python3 -c "
import numpy as np
from climateclaw import fetch_historical, compute_anomaly
base = fetch_historical(52.52, 13.405, '1991-01-01', '2020-12-31')
recent = fetch_historical(52.52, 13.405, '2021-01-01', '2023-12-31')
b = base['values'][~np.isnan(base['values'])]
r = recent['values'][~np.isnan(recent['values'])]
anom = compute_anomaly(r, b)
print(f'Mean anomaly: {anom.mean():+.2f} {recent[\"unit\"]}')
"

Example — is your city warming?

python3 << 'PYEOF'
import numpy as np, json
from climateclaw import fetch_historical, detect_trend, compute_mean

CITY   = "Hamburg"
LAT, LON = 53.5753, 10.0153

# Fetch 40 years of daily mean temperature
data = fetch_historical(LAT, LON, "1984-01-01", "2023-12-31")
dates, values = data["dates"], data["values"]
unit = data["unit"]

# Aggregate to annual means
years = sorted(set(d[:4] for d in dates))
annual_means = []
for y in years:
    year_vals = [v for d, v in zip(dates, values) if d.startswith(y) and not np.isnan(v)]
    annual_means.append(float(np.mean(year_vals)))

# Run trend
trend = detect_trend(annual_means)

# Report
coldest = years[int(np.argmin(annual_means))]
warmest = years[int(np.argmax(annual_means))]

print(f"\n  {CITY} — 40-year climate summary")
print(f"   Mean temperature : {compute_mean(annual_means):.2f} {unit}")
print(f"   Coldest year     : {coldest} ({min(annual_means):.2f} {unit})")
print(f"   Warmest year     : {warmest} ({max(annual_means):.2f} {unit})")
print(f"   Warming trend    : {trend['slope']:+.3f} {unit}/year")
print(f"   Over 40 years    : {trend['slope']*40:+.2f} {unit} total")
print(f"   R²               : {trend['r_squared']:.3f}")
print(f"   Significant?     : {'yes ' if trend['p_value'] < 0.05 else 'no ✗'} (p={trend['p_value']:.4f})")
PYEOF

Example output:

  Hamburg — 40-year climate summary
   Mean temperature : 10.31 °C
   Coldest year     : 1985 (7.94 °C)
   Warmest year     : 2020 (12.67 °C)
   Warming trend    : +0.051 °C/year
   Over 40 years    : +2.04 °C total
   R²               : 0.847
   Significant?     : yes  (p=0.0000)

Shell script usage

./scripts/climate.sh fetch 53.5753 10.0153 2019-01-01 2023-12-31 temperature_2m_mean
./scripts/climate.sh trend "7.94,8.12,8.45,9.1,9.8,10.2"
./scripts/climate.sh mean  "7.94,8.12,8.45,9.1,9.8,10.2"

Tips

• Aggregate daily → annual means before running detect_trend (daily noise kills signal).
• Use a 30-year WMO baseline (e.g. 1991–2020) for compute_anomaly.
• Strip NaN before any analysis: vals = values[~np.isnan(values)].