🌤️ Weather-Tajikistan-1940-2026

Comprehensive Hourly Weather Dataset for 11 Regions of Tajikistan (1940–2026)

This dataset provides 8,337,912 hourly weather records from 11 cities/regions of Tajikistan, spanning from 1940-01-01 to 2026-06-20 . It is designed for climate research, time-series analysis, and machine learning applications.

✨ Key Features

• 📊 11 regions across Tajikistan
• 🕐 Hourly resolution from 1940 to 2026 (86 years)
• 🌡️ 20+ meteorological variables:
  • Temperature (2m, apparent, dew point)
  • Humidity (relative, vapour pressure deficit)
  • Pressure (MSL, surface)
  • Precipitation (rain, snow, total)
  • Cloud cover (total, low, mid, high)
  • Wind speed & direction
  • Evapotranspiration
  • WMO weather codes
• 💾 Efficient Parquet format with Snappy compression
• 🔬 Clean and preprocessed (missing precipitation filled with 0)
• 🌍 Real data from Open-Meteo and Meteostat

📊 Dataset Statistics

Metric	Value
Total Records	8,337,912
Regions	11
Time Range	1940-01-01 — 2026-06-20
Resolution	Hourly
Columns	19
File Size	149.16 MB (Parquet)

🏙️ Regions Covered

• Ayni (757,992 records)
• Bokhtar (757,992 records)
• Danghara (757,992 records)
• Dushanbe (757,992 records)
• Khorug (757,992 records)
• Khujand (757,992 records)
• Kulob (757,992 records)
• Norak (757,992 records)
• Panj (757,992 records)
• Panjakent (757,992 records)
• Rasht (757,992 records)

📊 Key Variables

Variable	Description	Unit
temperature_2m	Air temperature at 2m height	°C
relative_humidity_2m	Relative humidity at 2m height	%
pressure_msl	Atmospheric pressure at mean sea level	hPa
precipitation	Total precipitation (rain + snow)	mm
cloud_cover	Total cloud cover	%
wind_speed_10m	Wind speed at 10m height	km/h
snow_depth	Snow depth	m
weather_code	WMO weather code	integer

📝 Data Sources

• Primary source: Open-Meteo — ERA5 reanalysis data
• Secondary: Meteostat — historical weather data
• Period: 1940-01-01 to 2026-06-20

🚀 Usage

Loading with 🤗 Datasets

from datasets import load_dataset
import pandas as pd

# Load entire dataset
dataset = load_dataset("arabovs-ai-lab/Weather-Tajikistan-1940-2026", split="train")
print(f"Total records: {len(dataset)}")

# Convert to pandas for analysis
df = dataset.to_pandas()
print(df.head())

# Filter by city
dushanbe = df[df['city'] == 'Dushanbe']
print(f"Dushanbe records: {len(dushanbe)}")

Loading with Pandas (direct Parquet)

import pandas as pd

# Load Parquet file directly
df = pd.read_parquet("weather_tajikistan_1940_2026.parquet")
print(df.info())

Time-Series Analysis Example

import pandas as pd
import matplotlib.pyplot as plt

# Load and prepare
df = pd.read_parquet("weather_tajikistan_1940_2026.parquet")
df['time'] = pd.to_datetime(df['time'])

# Filter Dushanbe, 2020-2024
dushanbe = df[(df['city'] == 'Dushanbe') & (df['time'] >= '2020-01-01')]

# Plot temperature
plt.figure(figsize=(12, 6))
plt.plot(dushanbe['time'], dushanbe['temperature_2m (°C)'], linewidth=0.5)
plt.title('Temperature in Dushanbe (2020-2024)')
plt.xlabel('Date')
plt.ylabel('Temperature (°C)')
plt.grid(True, alpha=0.3)
plt.show()

📈 Data Quality

• Precipitation columns (precipitation, rain, snow) have missing values filled with 0.0
• Evapotranspiration missing values filled with 0.0
• All float columns optimized to float32 for efficiency
• Weather code may contain missing values (historical WMO data not always available)

📄 License

This dataset is released under the MIT License.

📚 Citation

If you use this dataset, please cite:

@misc{weathertajikistan2026,
  author = {Arabov, Mullosharaf K.},
  title = {Weather-Tajikistan-1940-2026: A Comprehensive Hourly Weather Dataset for 11 Regions of Tajikistan},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/arabovs-ai-lab/Weather-Tajikistan-1940-2026}
}

🔗 Related Work

• WeatherReason-2026: arabovs-ai-lab/WeatherReason-2026
• Open-Meteo API: https://open-meteo.com/
• Meteostat: https://meteostat.net/

---

Built for climate research and the Central Asian NLP community. 🌍