Usage Notes

Seven rules to read before integration — all verified on the sample data.

Time Zones

File contents (date_time) are Beijing time; the file-name issue time is UTC — an 8-hour difference. See Time Conventions.

Temperature Units

t2m is Kelvin; using it directly as Celsius introduces a 273.15 bias:

df["t2m_c"] = df["t2m"] - 273.15

Pressure Units

sp is in pascals (Pa); business commonly uses hectopascals (hPa):

df["sp_hpa"] = df["sp"] / 100

Solar Radiation Accumulation

You must difference the accumulated values to obtain per-interval irradiance — never use them as instantaneous values:

W/m² = (accumulated end − accumulated start) ÷ interval seconds    # ÷ 900 for 15-minute steps

• Nighttime values plateau and can indicate sunrise / sunset periods;
• Switch between both conventions visually on the Live Demo page.

rr Is Not Rainfall

rr is reflected radiation (surface-reflected shortwave), not precipitation — do not ingest it as a rain field. In the sample it satisfies rr ≈ 0.216 × tr, usable as a data check.

Cloud Cover Out of Bounds

Cloud fractions are 0-1 (not percentages); grid interpolation may cause slight out-of-range values (sample observed −0.18 to 1.13). Recommended:

df[["lcc", "mcc", "hcc", "tcc"]] = df[["lcc", "mcc", "hcc", "tcc"]].clip(0, 1)

Full Replacement on New Files

When a new run's file arrives it should fully replace the previous run's data to avoid mixing issue times. Partition or version by farm_id + issue for idempotent ingestion.

Appendix: Self-check List

Check	Expected
Row count	1441 (15 days × 96 points/day, both endpoints included)
Column count	19, order per the File Format page
Time intervals	Strictly regular 15 minutes, no gaps or duplicates
Missing	Empty string denotes missing (sample verified: none)
Radiation monotonicity	dr / sr / rr / tr non-decreasing over time
Radiation consistency	rr ≈ 0.216 × tr; dr + sr ≈ 1.21 × tr