dazhiyang/bsrn-merra2

BSRN MERRA-2 Atmospheric Inputs

Point-extracted MERRA-2 reanalysis data for Baseline Surface Radiation Network (BSRN) stations. These parquet files provide atmospheric and aerosol inputs for the REST2 clear-sky radiation model [2].

Dataset Description

Each file contains hourly MERRA-2 variables [1] at a single BSRN station location. Extraction is performed via Google Earth Engine (GEE) from NASA's 0.5° × 0.625° global grid. Data are aligned to the MERRA-2 grid cell nearest the station coordinates.

File Structure

Station folders use the lowercase BSRN three-letter code (e.g. qiq, ber, bil, bon).

{station}/
  {station}{MM}{YY}_merra2.parquet   # One file per month

Examples:

• qiq/qiq0124_merra2.parquet — QIQ, January 2024
• ber/ber0325_merra2.parquet — BER, March 2025

Variables

Column	Description	MERRA-2 Source	Units (raw)
AOD55	Aerosol optical depth at 550 nm	TOTEXTTAU	dimensionless
ALPHA	Ångström exponent	TOTANGSTR	dimensionless
ALBEDO	Surface albedo	ALBEDO	[0–1]
TQV	Total column precipitable water vapor	TQV	kg/m²
TO3	Total column ozone	TO3	Dobson
PS	Surface pressure	PS	Pa

• Index: UTC DatetimeIndex (hourly, MERRA-2 native resolution).
• Time coverage: MERRA-2 spans 1980–present; files are generated only for months with BSRN station-to-archive data on the FTP.

Use with REST2

These parquet files are designed for the REST2 clear-sky model [2]. The bsrn Python package fetches MERRA-2 from this dataset into RAM (no disk cache) and provides:

• fetch_rest2(index, station_code) — fetch from HF into RAM, reindex to 1-min target, interpolate, derive BETA, and convert units for REST2
• Raw parquet: use pandas.read_parquet on a path from huggingface_hub (see below)

REST2 expects: PS (hPa), ALBEDO, ALPHA, BETA (derived from AOD55 and ALPHA), TO3 (atm·cm), TQV (atm·cm).

Conversion tips (raw → REST2):

Variable	Raw unit	REST2 unit	Conversion
PS	Pa	hPa	÷ 100
ALBEDO	[0–1]	[0–1]	no conversion
ALPHA	dimensionless	dimensionless	no conversion
BETA	—	—	AOD55 × 0.55^ALPHA (use 0.001 if AOD55=0)
TO3	Dobson	atm·cm	÷ 1000
TQV	kg/m²	atm·cm	÷ 10

Usage

Load from Hugging Face

from huggingface_hub import hf_hub_download
import pandas as pd

# Download a single file; path is {station}/{station}{MM}{YY}_merra2.parquet
path = hf_hub_download(
    repo_id="dazhiyang/bsrn-merra2",
    filename="qiq/qiq0124_merra2.parquet",
    repo_type="dataset",
)
df = pd.read_parquet(path)
# df has DatetimeIndex (UTC) and columns: AOD55, ALPHA, ALBEDO, TQV, TO3, PS

Use with bsrn package

The bsrn package fetches MERRA-2 from Hugging Face into RAM (no disk cache). You will see Fetching MERRA-2 from Hugging Face: {filename} when it runs.

from bsrn.modeling.clear_sky import add_clearsky_columns

# Option 1: Load raw parquet (from path, e.g. after hf_hub_download)
df = load_merra2_parquet(path)

# Option 2: REST2-ready inputs (fetches from HF into RAM, interpolated to 1-min, units converted)
# target_index = your BSRN 1-min DatetimeIndex
rest2_inputs = fetch_rest2(target_index, station_code="QIQ")

# Option 3: Add clear-sky columns to BSRN data (fetches MERRA-2 from HF into RAM automatically)
df = add_clearsky_columns(df, station_code="QIQ", model="rest2")

Data Sources

• MERRA-2: NASA GMAO, GES DISC
• Extraction: Google Earth Engine (GEE) — this dataset uses GEE; NCSS is an alternative source.
• Station inventory: BSRN FTP (months with .dat.gz files only)

GEE extraction and validation

The data in this dataset is extracted via Google Earth Engine (GEE). GEE's MERRA-2 pixel boundaries are offset by half a cell in latitude relative to raw NetCDF. To correct this, a −0.25° latitude shift is applied when querying GEE so that the returned pixel aligns with the MERRA-2 grid cell used by raw MERRA-2 and NASA GESDISC NCSS.

The dataset creator has confirmed the correctness of the GEE extraction by comparing it against raw MERRA-2 data from NASA and point extractions from NASA GESDISC THREDDS NCSS.

References

1. Gelaro, R., McCarty, W., Suárez, M. J., Todling, R., Molod, A., Takacs, L., ... & Zhao, B. (2017). The modern-era retrospective analysis for research and applications, version 2 (MERRA-2). Journal of Climate, 30(14), 5419–5454.

2. Gueymard, C. A. (2008). REST2: High-performance solar radiation model for cloudless-sky irradiance, illuminance, and photosynthetically active radiation—Validation with a benchmark dataset. Solar Energy, 82(3), 272–285.

3. Sun, X., Bright, J. M., Gueymard, C. A., Acord, B., Wang, P., & Engerer, N. A. (2019). Worldwide performance assessment of 75 global clear-sky irradiance models using principal component analysis. Renewable and Sustainable Energy Reviews, 111, 550–570.

Citation

If you use this dataset, please cite the references above and the bsrn package: dazhiyang/bsrn.

License

This dataset mirrors publicly available MERRA-2 reanalysis data. MERRA-2 is produced by NASA and is freely available. See NASA's data use policy for terms of use.