bsrn 0.2.1

A Python package for quality control (QC) checks on BSRN station-to-archive files.

bsrn

This GitHub repository is dazhiyang/bsrn: the source code and development tooling for the bsrn Python package.

bsrn is a community-developed toolbox that provides a set of robust functions and classes for processing and analyzing solar radiation data. The core mission of bsrn is to provide an open, reliable, interoperable, and benchmark-standard set of tools tailored specifically for the Baseline Surface Radiation Network (BSRN).

It features automated quality control (QC), high-precision solar geometry, clear-sky modeling, clear-sky detection (CSD), cloud enhancement event (CEE) detection, irradiance separation, and comprehensive data retrieval and visualization capabilities.

📖 Documentation (Read the Docs)

🚀 Getting Started

Installation

The core bsrn package is designed to be lightweight and fast. You can install it using pip:

From PyPI (stable release):

pip install bsrn

From GitHub (latest development version):

pip install git+https://github.com/dazhiyang/bsrn.git

Optional Visualization Tools

If you want to use the built-in plotting features (like data availability charts or clear-sky calendars), you will need to install the optional visualization dependencies (plotnine, matplotlib, and scipy):

pip install bsrn[viz]

Usage

For standard quality control and clear-sky modeling, simply import the base package:

import bsrn

# Access core modules like bsrn.qc, bsrn.modeling, bsrn.io, bsrn.archive

If you installed the [viz] extra and want to generate plots, you must explicitly import the visualization submodule:

import bsrn.visualization

# Access plotting tools like bsrn.visualization.calendar.plot_calendar()

Quick Example — BSRNDataset (recommended)

import bsrn

# One station-month from a BSRN LR0100 archive (.dat.gz)
ds = bsrn.BSRNDataset.from_file("data/QIQ/qiq0125.dat.gz")

# Typical pipeline (each step mutates the cached frame and returns it)
ds.solpos()                             # solar geometry + extraterrestrial
ds.clear_sky(model="rest2")             # ghi_clear / bni_clear / … (REST2; MERRA-2 via Hugging Face)
ds.qc_test()                            # flag columns: 0 = pass, 1 = fail
ds.qc_mask()                            # NaN failed irradiance; drop flag columns

df = ds.data()                          # minute-resolution table for analysis or export

# Visualize directly from the dataset (requires bsrn[viz])
ds.plot.daily("2025-01-15")             # UTC date inside the loaded month
ds.plot.table()                         # QC summary table

Quick Example — Functional API

The same steps are available as standalone functions, useful for non-BSRN data or custom DataFrames:

from bsrn.io.retrieval import download_bsrn_stn, get_bsrn_file_inventory
from bsrn.physics.geometry import add_solpos_columns
from bsrn.modeling.clear_sky import add_clearsky_columns
from bsrn.qc.wrapper import run_qc

# 1. See what data is available
inventory = get_bsrn_file_inventory(["QIQ"], username="your_user", password="your_pass")

# 2. Download data for a station
download_bsrn_stn("QIQ", "data/QIQ", username="your_user", password="your_pass")

# 3. Read via BSRNDataset and get the DataFrame
ds = bsrn.BSRNDataset.from_file("data/QIQ/qiq0125.dat.gz")
df = ds.data()

# 4. Add solar position (recommended before time-averaging or clear-sky)
df = add_solpos_columns(df, "QIQ")

# 5. Add clear-sky reference columns (defaults to Ineichen)
df = add_clearsky_columns(df, "QIQ")

# 6. Run Quality Control (QC)
df = run_qc(df, "QIQ")

# 7. Add satellite-derived CAMS CRS all-sky columns
from bsrn.io.crs import add_crs_columns
df = add_crs_columns(df, "QIQ")

# 8. Visualize with plotnine
from bsrn.visualization.clearsky_models import plot_clearsky_models
plot_clearsky_models(df, "QIQ", date="2024-06-20", save_path="clearsky_qiq.pdf")

🛠 Features

The QC features, of which the implementation is primarily based on the BSRN Operations Manual (2018) and Forstinger et al. (2021). See code for other references.

• Level 1 (Physically Possible): Absolute physical bounds for $G_h, B_n, D_h$, and $L_d$.
• Level 2 (Extremely Rare): Climatological limits for specific regimes.
• Level 3 (Comparison): Consistency checks ($G_h$ vs $B_n \cos Z + D_h$) with zenith-dependent thresholds.
• Level 4 (Diffuse Ratio): Diffuse-fraction and $k$–$k_t$ checks combining $G_h$, $D_h$, and extraterrestrial irradiance.
• Level 5 (K-Indices): Advanced clearness-index and $k_b$/$k_t$ index tests using clear-sky benchmarks and site elevation.
• Level 6 (Tracker-Off Detection): Identify tracking errors by comparing measured values with clear-sky and extraterrestrial irradiance.

Other important features include:

• Solar Geometry: Native NREL SPA implementation for high-precision solar position calculations.
• Clear-Sky Models: Ineichen (monthly Linke turbidity), McClear (CAMS SoDa API, from 2004 onward), and REST2 (MERRA-2 from Hugging Face).
• Satellite Data: Load CAMS solar radiation service (CRS) and National Solar Radiation Database (NSRDB) all-sky irradiance directly from Hugging Face into memory.
• Clear-Sky Detection (CSD): Reno, Ineichen, Lefevre, and BrightSun methods to identify clear-sky periods from irradiance time series.
• Cloud Enhancement Event (CEE) Detection: Killinger, Yang, and Gueymard methods to detect events when measured GHI significantly exceeds references.
• Irradiance Separation: Erbs, BRL, Engerer2, and Yang4 models to estimate diffuse fraction and DHI/BNI from GHI.
• Robust Retrieval: High-level API for FTP downloads from BSRN-AWI with exponential backoff retries (analysis functions assume one station-to-archive file at a time).
• Station-to-archive formatting: The bsrn.archive subpackage provides LR_SPECS, Fortran-style validators in validation.py (names referenced by each field’s validate_func), and ASCII output via get_bsrn_format. Scalar/header fields on the Pydantic LR* models use a single lr_spec(lr_code, field_name, type, …) annotation so metadata and post-parse checks stay in one place; LR0100/LR4000 minute columns use field_validator with yearMonth for vector length checks. Concrete types (LR0001–LR4000CONST) live in records_models and are re-exported from bsrn.archive; get_azimuth_elevation is in archive_lr_formats (also re-exported).
• Visualization: Data availability heatmaps and k vs kt separation plots via the very pretty plotnine (which reminds me of the good old R days).

📂 File Structure

[!NOTE] Not all files are uploaded with Git. Data files and intermediate outputs are excluded via .gitignore.

bsrn-qc/
├── pyproject.toml
├── LICENSE
├── README.md
├── .gitignore
├── .readthedocs.yaml              # Read the Docs build config
├── src/
│   └── bsrn/
│       ├── __init__.py
│       ├── dataset.py                 # BSRNDataset: central monthly data object + pipeline methods
│       ├── constants.py               # Station database, Linke turbidity & physical constants
│       ├── archive/                   # Station-to-archive logical records (WRMC-style LR layouts)
│       │   ├── __init__.py            # Re-exports LR* models, LR_SPECS, get_azimuth_elevation, …
│       │   ├── specs.py               # LR_SPECS + station directory & A3–A7 code tables
│       │   ├── archive_lr_formats.py  # get_bsrn_format + get_azimuth_elevation (LR0004)
│       │   ├── records_base.py        # ArchiveRecordBase, make_archive_after_validator
│       │   ├── records_models.py      # lr_spec / lr_spec_field; LR0001–LR4000CONST Pydantic models
│       │   ├── formatting.py          # Fortran-style field formatting mixin
│       │   └── validation.py          # BSRN archive field validators (LR_SPECS validate_func)
│       ├── io/
│       │   ├── reader.py              # Read xxxmmyy.dat.gz station-to-archive files
│       │   ├── retrieval.py           # FTP downloads with retries
│       │   ├── merra2.py              # MERRA-2 parquet fetch (Hugging Face → RAM)
│       │   ├── mcclear.py             # SoDa McClear client helpers
│       │   ├── crs.py                 # SoDa CAMS solar radiation service (CRS) client helpers
│       │   ├── nsrdb.py               # NREL NSRDB all-sky data client helpers
│       │   └── writers.py             # Export results
│       ├── physics/
│       │   ├── spa.py                 # Native NREL SPA (solar position algorithm)
│       │   └── geometry.py            # Solar position and extraterrestrial irradiance
│       ├── qc/
│       │   ├── ppl.py                 # Physically possible limits (Level 1)
│       │   ├── erl.py                 # Extremely rare limits (Level 2)
│       │   ├── closure.py             # Internal consistency checks (Level 3)
│       │   ├── diff_ratio.py          # Diffuse ratio checks (Level 4)
│       │   ├── k_index.py             # Radiometric index tests (Level 5)
│       │   ├── tracker.py             # Solar tracker off detection (Level 6)
│       │   └── wrapper.py             # High-level QC pipeline
│       ├── visualization/
│       │   ├── availability.py        # File coverage heatmaps (plotnine)
│       │   ├── qc_table.py            # QC result tables
│       │   ├── separation.py          # Decomposition visualization
│       │   └── timeseries.py          # Time series plots
│       ├── utils/
│       │   ├── calculations.py        # Supporting math
│       │   ├── quality.py             # Quality utilities
│       │   ├── clear_sky_detection.py # Clear-sky detection (Reno, Ineichen, Lefevre, BrightSun)
│       │   └── cee_detection.py       # Cloud enhancement detection (Killinger, Yang, Gueymard)
│       └── modeling/
│           ├── clear_sky.py           # Ineichen clear-sky model
│           └── separation.py          # Irradiance separation (Erbs, BRL, Engerer2, Yang4)
├── docs/
│   ├── conf.py                        # Sphinx config; source dir = docs/ (tutorials + sphinx/ RST)
│   ├── index.rst                      # Site homepage (root index.html for Read the Docs)
│   ├── requirements.txt               # Sphinx / Read the Docs dependencies
│   ├── examples/                      # Examples landing page (index.rst) + optional scripts
│   │   └── index.rst
│   ├── tutorials/                     # Jupyter tutorials + index.rst (nbsphinx)
│   │   ├── 1.data_downloading.ipynb
│   │   ├── 2.quality_control.ipynb
│   │   ├── 3.time_averaging.ipynb
│   │   ├── 4.clear_sky_detection.ipynb
│   │   └── 5.cloud_enhancement_event.ipynb
│   └── sphinx/                        # RST (user_guide, api, _static); not the doc homepage
│       ├── api/                       # API reference (io, qc, physics, …)
│       └── user_guide/                # installation, getting_started, package_overview, …

📖 Examples

Solar Position

import pandas as pd
from bsrn.physics.geometry import get_solar_position, get_bni_extra

times = pd.date_range("2024-07-01", periods=1440, freq="1min", tz="UTC")
solpos = get_solar_position(times, lat=47.80, lon=124.49, elev=170)

print(solpos[["zenith", "apparent_zenith", "azimuth"]].head())

Extraterrestrial Irradiance

from bsrn.physics.geometry import get_bni_extra

bni_extra = get_bni_extra(times)  # Spencer (1971) method

Clear-Sky GHI (Ineichen)

from bsrn.modeling.clear_sky import add_clearsky_columns

# Automatically computes solar geometry if missing, but it is highly
# recommended to call `add_solpos_columns(df)` first for 1-minute data!
df = add_clearsky_columns(df, "QIQ")
# Adds columns: ghi_clear, bni_clear, dhi_clear

Clear-Sky GHI from McClear (CAMS)

from bsrn.modeling.clear_sky import add_clearsky_columns

# McClear data are available from 2004-01-01 onward.
# McClear 数据自 2004-01-01 起可用。
df = add_clearsky_columns(
    df,
    station_code="QIQ",
    model="mcclear",
    mcclear_email="[e-mail]",  # SoDa / CAMS account email
)
# Adds columns: ghi_clear, bni_clear, dhi_clear based on CAMS McClear

Clear-Sky GHI from REST2 (MERRA-2 via Hugging Face)

REST2 uses MERRA-2 atmospheric inputs only from the Hugging Face dataset dazhiyang/bsrn-merra2 (hourly Parquet files per station, station_code/*.parquet). The bsrn package fetches them into RAM (no disk cache) when model="rest2" is used.

from bsrn.modeling.clear_sky import add_clearsky_columns

# MERRA-2 is fetched from Hugging Face into RAM automatically.
df = add_clearsky_columns(df, station_code="QIQ", model="rest2")
# Adds columns: ghi_clear, bni_clear, dhi_clear based on REST2 + MERRA-2

The dataset README for Hugging Face is maintained in this repo at data/bsrn_static_assets/README.md (published to the Hub separately from PyPI).

All-Sky GHI from NSRDB (NREL via Hugging Face)

Similar to REST2, NSRDB all-sky data is fetched directly from the Hugging Face dataset dazhiyang/bsrn-nsrdb-conus (and other variants).

from bsrn.io.nsrdb import add_nsrdb_columns

# Fetch NSRDB all-sky GHI/DNI/DHI from Hugging Face
df = add_nsrdb_columns(df, station_code="QIQ", variant="conus")
# Adds columns: ghi_nsrdb, bni_nsrdb, dhi_nsrdb

Clear-Sky Detection

from bsrn.utils import detect_clearsky

# Requires GHI and clear-sky GHI (e.g. from add_clearsky_columns)
out = detect_clearsky("reno", ghi=df["ghi"], ghi_clear=df["ghi_clear"], times=df.index)
# out["is_clearsky"] is True/False/NA; out["cloud_flag"] is 0/1/NaN
# Other methods: "ineichen", "lefevre", "brightsun" (different inputs)

Cloud Enhancement Event (CEE) Detection

from bsrn.utils.cee_detection import detect_cee

# Killinger CEE detection: requires 1‑min GHI, clear-sky GHI, zenith, and a 1‑min index
out_cee_k = detect_cee(
    "killinger",
    ghi=df["ghi"],
    ghi_clear=df["ghi_clear"],
    zenith=df["zenith"],
    times=df.index,
)
# out_cee_*["is_enhancement"] is True/False/NA; out_cee_*["cee_flag"] is 0/1/NaN

Data Availability Heatmap

from bsrn.visualization.availability import plot_bsrn_availability

fig = plot_bsrn_availability(inventory_df, station_code="QIQ")
fig.save("availability.png", dpi=300)

Station-to-archive logical records (bsrn.archive)

Logical records are Pydantic v2 models (LR0001, …, LR0100, LR4000, LR4000CONST, …) defined in records_models and re-exported from bsrn.archive. The legacy umbrella type BSRNRecord is removed—use a concrete LR* model and call get_bsrn_format on the instance.

• LR_SPECS holds per-field format, missing tokens, defaults, and validate_func names.
• Scalars: validation runs through Pydantic AfterValidator, which calls the matching function in bsrn.archive.validation.
• LR0100 / LR4000 minute vectors: validators need yearMonth; those columns use a model-level field_validator instead.

from bsrn.archive import LR0001, LR_SPECS

# Required keys for LR0001 are listed in LR_SPECS["LR0001"]
out = LR0001(stationNumber=94, month=1, year=2024, version=1).get_bsrn_format()

For minute blocks, pass yearMonth="YYYY-MM" and pandas.Series or numpy.ndarray per column (see LR_SPECS["LR0100"] / ["LR4000"]), then LR0100(...).get_bsrn_format(changed=True) (and similarly for LR4000).

Regression check (repository checkout): from the repo root, generate a monthly .dat and compare to the checked-in reference (should match byte-for-byte):

PYTHONPATH=src python tests/2025-01/Code/2.station_to_archive.py \
  -o tests/2025-01/Output/qiq0125_run.dat --no-gzip
cmp tests/2025-01/Output/qiq0125_run.dat tests/2025-01/Output/qiq0125_ref.dat

Edit the CONFIG block at the top of 2.station_to_archive.py for station-specific paths and metadata; the script expects the minute table at tests/2025-01/Output/qiq0125.txt for the default QIQ January 2025 example.

📜 License

MIT License. See LICENSE for details.