BDAirFusion: District-level meteorology and satellite atmospheric observations for Dhaka Division, 1 January–30 October 2026

Published: 4 November 2026| Version 1 | DOI: 10.17632/s892k4krr2.1
Contributor:
Md Abdur Rahman

Description

BDAirFusion provides an integrated district-level dataset of meteorological conditions and satellite atmospheric retrievals for Dhaka Division, Bangladesh, covering 1 January–30 October 2026. It combines ERA5-Land estimates with MODIS MAIAC and Sentinel-5P products to support research on atmospheric variability, meteorological influences and differences between districts. The dataset covers Dhaka, Faridpur, Gazipur, Gopalganj, Kishoreganj, Madaripur, Manikganj, Munshiganj, Narayanganj, Narsingdi, Rajbari, Shariatpur and Tangail. Fixed ADM2 polygons define the spatial units, and daily records follow Asia/Dhaka local dates. The principal daily table contains 3,939 district-day records and 33 columns, including 30 environmental variables. An hourly meteorological table contains 94,536 records and 20 columns. A separate quality-control table contains 118,170 district-day-variable records. CSV and Parquet provide alternative representations of the same products. Meteorological variables include air temperature, dewpoint, relative humidity, wind components, speed and direction, surface pressure, precipitation, radiation-related energy amounts, skin and soil temperature, soil moisture and net evaporation. Satellite variables include aerosol optical depth at 470 and 550 nm, selected atmospheric gas columns, methane column mixing ratio, cloud properties and absorbing aerosol index. These represent model estimates and satellite retrievals rather than district monitoring-station measurements. Ground and CAMS particulate-matter fields are excluded. Meteorological fields are aggregated using polygon–grid intersection-area weights in EPSG:6933. Satellite processing preserves source-image timestamps, product masks and native-projection pixel-area statistics. Physical conversions, temporal aggregation and variable-specific quality rules are documented. Source availability, validity and completeness are distinguished. Missing values remain null without temporal interpolation or zero filling; satellite gaps and incomplete final-day support are retained. External-reference evaluation uses seven NOAA stations in six districts, providing 19,899 station-variable pairs at 5,417 unique station-times, including 325 matched 24-hour precipitation totals. Strict-QA reference coverage is mainly January–March. Station-specific error metrics, monthly summaries and sampling intervals report both agreement and weak wind/rainfall results. Spatial-support differences and possible assimilation overlap limit independence. AERONET Level 1.5 AOD550 comparisons are exploratory. The release includes data dictionaries, source attribution, licence information, raw-source evidence, processing code, validation reports, checksums and reproducibility instructions. It supports coverage-aware environmental analysis and methodological research, with documented limitations on observational completeness and independent validation.

Files

Steps to reproduce

1. Download and extract the dataset ZIP. Read README.md, documentation/Methodology.md and documentation/Validation.md before analysis. 2. The main daily, hourly and variable-specific QC tables are in data/. Consult metadata/Data_Dictionary.csv and metadata/QC_Data_Dictionary.csv for units, keys, aggregation rules and missing-value meanings. CSV and Parquet are alternative formats of the same tables. 3. Use Python 3.12. From the extracted package directory, install the pinned dependencies: python -m pip install -r reproducibility/requirements-reproduce.lock.txt 4. Verify the reproducibility workspace: python reproducibility/code/reproduce.py --package reproducibility --verify-only 5. Regenerate hourly meteorology from the bundled provider NetCDF files and published polygon weights, followed by daily model and satellite aggregation: python reproducibility/code/raw_model_reproduce.py --package reproducibility --output ../BDAirFusion_model_replay 6. Reproduce strict weather-reference selection, temporal pairing and descriptive validation metrics: python reproducibility/code/raw_reference_reproduce.py --package reproducibility --output ../BDAirFusion_reference_replay Output directories must be new and outside the extracted package. These portable replay steps require no API credentials or additional downloads. 7. Compare the generated results with the supplied tables and verification reports. Model-table replay checks exact missing-value masks and numerical agreement within rtol=1e-10 and atol=1e-7. The replay uses published spatial weights and preserved satellite reduction statistics. It does not recreate upstream reanalysis assimilation, independently construct GIS intersection weights or rerun Earth Engine pixel reductions. Processing reproducibility does not establish independent environmental accuracy. Additional instructions and limitations are provided in REPRODUCIBILITY.md.

Institutions

Categories

Meteorology, Atmospheric Science, Environmental Science, Remote Sensing

Licence