xquantize/climate-anomaly-cells-50-locations-1990-2024

Climate Anomaly Cells (daily, 50 locations, 1990–2024)

Daily climate anomalies for 50 curated high-exposure urban/rural cells worldwide, 1990-01-01 to 2024-12-31, derived from NASA POWER (GMAO MERRA-2 reanalysis). Each cell has observed daily temperature, precipitation, wind and humidity; smoothed 1991–2020 day-of-year normals; additive anomalies; a hot-day flag; NWS heat index and wind chill; and an illustrative composite stress index. One UTF-8 CSV, 639,200 rows × 39 columns (50 cells × 12,784 days), complete with no gaps or duplicates.

Derived and analysis-ready good for demos, teaching, dashboards and coarse screening. Not ERA5, not station observations, not an official ETCCDI/WMO index product.

Load

from datasets import load_dataset
ds = load_dataset("YOUR_USERNAME/climate-anomaly-cells", split="train")

# or with pandas straight from the Hub
import pandas as pd
df = pd.read_csv("hf://datasets/YOUR_USERNAME/climate-anomaly-cells/climate_anomaly_cells.csv",
                 parse_dates=["time"])

The CSV sits at the repo root; if you move it, update data_files: path: in the YAML above.

Columns (39)

• Identity: location_id, city, country, latitude, longitude, zone_type, risk_category
• Time: time (ISO-8601, UTC midnight, a daily date label, see below), year, month, day, day_of_year
• Observed: temp_2m_{mean,max,min}_c, precip_sum_mm, wind_10m_max_kmh, apparent_temp_max_c, rh_2m_mean_pct
• Normals (*_normal) and anomalies (*_anom) for those seven variables, plus precip_pct_of_normal (null when the precip normal < 0.2 mm)
• Flags / indices: hot_day_threshold_c, hot_day (0/1), heat_index_c, wind_chill_c, climate_stress_index

Methodology (brief)

NASA POWER daily point API (community=AG), one request per city; WS10M_MAX converted m/s→km/h; apparent_temp_max_c via Steadman. Per cell, 1991–2020 day-of-year normals (centred 15-day circular smoothing); anomaly = value − normal. hot_day = max temp above the per-cell, per-day-of-year 90th percentile of the baseline (ETCCDI TX90p style, single-day, not official). heat_index_c is the NWS/Rothfusz formula, not capped to its valid range. climate_stress_index is an illustrative z-score blend, not official.

Time standard: POWER daily stats are computed over the local solar time (LST) day, stored at UTC midnight, treat time as a date label, not an exact UTC instant.

Quality and caveats

Automated verifier passes 19/20 (1 warning); an independent POWER re-fetch of Delhi Jan 2020 matched exactly. Against station normals the reanalysis shows the usual small warm/dry biases and damped extremes (coarse ~0.5°×0.625° grid). Two things to know before trusting tails:

• Precipitation (PRECTOTCORR) is MERRA-2, not gauges. Daily tails can be wildly high e.g. Jacobabad shows 400+ mm days and a 1,900 mm 2022 total, driving precip_pct_of_normal to ~30,000%. High-latitude precip (Verkhoyansk, Norilsk) is largely uncorrected model precip.
• Heat index reaches 85 °C in monsoon cities — Rothfusz extrapolation, not a physical apparent temperature. Use as a derived, out-of-range signal only.

Not for: legal/insurance/engineering use, official ETCCDI reporting, homogeneous trend detection, street-level UHI claims or using the stress index / uncapped heat index / percent-of-normal as decision-critical values.

License and citation

Released as CC-BY 4.0 (swap the YAML license line to cc0-1.0 if you prefer public domain). Underlying data is NASA POWER / MERRA-2 (US Government, free to use with acknowledgement) — please credit the NASA Langley POWER Project and cite MERRA-2: Gelaro et al. 2017 (doi:10.1175/JCLI-D-16-0758.1) and Reichle et al. 2017 for precipitation (doi:10.1175/JCLI-D-16-0570.1). NASA POWER: https://power.larc.nasa.gov/ (accessed 2026-08-17). Do not imply ECMWF/C3S produced this dataset (the ERA5/Open-Meteo path was not used).