Technical Report | Science Communication Draft

Built-Up Context and T anomaly (1926-1935 baseline)

Using a fixed 100-year station set from GHCN-M v4 QCU unadjusted and local 5 km built-up fractions from GHSL, the most built-up station group has a 2.02 C mean anomaly in 2015-2025 against the 1926-1935 baseline, versus 1.41 C in the least built-up group. The recent high-minus-low spread is 0.61 C, while linear fits from 1926-2025 run from 0.140 C/decade in Q1 to 0.206 C/decade in Q9.

Author: Codex 5.4 Generated: 2026-03-19 UTC Series: GHCN-M v4 QCU Window: 1926-2025

471Strict-continuity stations

100Common analysis years

9Equal-count BU bins

5 kmLocal built-up radius

Main takeaway. On this fixed QCU station set, the high-BU end stays substantially warmer than the low-BU end in the T anomaly (1926-1935 baseline) metric. The highest-BU group is 0.61 C warmer than the lowest-BU group in the 2015-2025 mean, 0.66 C warmer in 2025, and the high-minus-low linear-fit gap is +0.066 C/decade.

High-Level Input Brief

This analysis was refactored into explicit Python workflow stages so the same code can support exports and an inline notebook tutorial.

Use GHCN-M monthly temperature data and find the best long stations.
Apply an upfront continuity screen: target 100 years and require roughly 9 valid months per year.
Measure local urban context with GHSL built-up data in a 5 km radius, validating the calculation against Munich and Tokyo.
Bin stations by built-up level and compare anomaly curves in a science-communication layout with a written report.

Quality Criteria Up Front

Consistent station setAll bins use the same strict station universe through the whole 100-year window.

Continuity firstAt least 9 valid months in every year, before any BU analysis.

No latitude screenStation selection is based on record continuity rather than a latitude cutoff.

Modern urban context metricGHSL gives fractional built-up intensity, avoiding the blunt urban versus non-urban logic of older binary masks.

Figure 1. T anomaly (1926-1935 baseline) by Equal-Count Built-Up Bin

T anomaly (1926-1935 baseline) curves by built-up bin — Each station contributes annual anomalies relative to the 1926-1935 monthly climatology, then values are averaged within its BU bin. The bins contain nearly equal numbers of stations and remain fixed through time. Solid curves show 11-year rolling means and dashed curves show linear fits from 1926-2025.

What the Curves Show

The least built-up group reaches a 1.41 C mean anomaly in 2015-2025 against the 1926-1935 baseline. The most built-up group reaches 2.02 C.

Q1 spans 0.00-0.60% built-up within 5 km.
Q9 spans 18.63-38.57% built-up within 5 km.
The recent Q9 minus Q1 spread is 0.61 C.
The single-year Q9 minus Q1 spread in 2025 is 0.66 C.
Linear fits from 1926-2025 run from 0.140 C/decade in Q1 to 0.206 C/decade in Q9.

Why This Design

The workflow is split into screening, built-up measurement, time-series analysis, static storytelling, and notebook display.
The study is not trying to estimate a global mean trend.
The goal is to compare a stable long-record station set across local urban-context bins.
Using a fixed 1926-1935 anomaly baseline keeps all stations on the same reference period.

Figure 2. Where the Equal-Count Built-Up Bins Sit

Maps of each equal-count built-up bin — Each panel highlights one BU bin against the rest of the selected station set in grey. The point is not perfect spatial balance; the point is a transparent, fixed, high-quality set that lets the BU contrast be seen without station churn over time.

Figure 3. Continuous 5 km BU Map

Station map colored by built-up fraction — Continuous station coloring from low built-up blue to high built-up red. The notebook version also renders this view as an interactive Folium map.

Figure 4. Munich and Tokyo Validation

Built-up validation examples for Munich and Tokyo — Munich comes out at 24.8% and Tokyo at 40.6% within 5 km, which supports the GHSL scaling used here.

Interactive Maps

Folium versions of the map views were also exported for browser-based exploration.

Interactive station BU map | Interactive equal-bin map

Methods in Brief

Temperature series: NOAA GHCN-M v4 QCU monthly mean temperature.
Common window: 1926-2025, the latest complete 100 calendar years in the current release.
Continuity rule: at least 9 valid months in every year.
Latitude screen: none.
Built-up metric: overlap-area-weighted 5 km mean from GHSL, using raw cell value divided by 1,000,000.
Anomaly baseline: station annual anomalies are computed from the 1926-1935 monthly climatology before BU-bin averaging.
Interactive maps: Folium versions are emitted as HTML, while the static report keeps deterministic PNG figures.

Typical Objections, Answered

Q. Why not grid the stations?

Because the long-record sample is still modest, and the main design goal is a consistent fixed station set. Gridding would add interpolation and area-weighting choices without improving the core comparison.

Q. Why not use more stations by relaxing the filters?

Because station churn can create moving-sample artifacts. The quality criteria were fixed up front so the comparison stays transparent across the full century.

Q. Why use GHSL instead of an older binary mask?

Because GHSL gives continuous built-up intensity around each station, which is better aligned with a 5 km local-context metric than a binary urban flag.

Q. Does the 1926-1935 baseline change the group separation?

It sets the common reference period for the anomalies. The high-minus-low contrast remains a contrast between the same station groups under that shared baseline.

Tables

Equal-count BU bins

BU bin	Stations	2015-2025 mean T anomaly vs 1926-1935 (C)	2025 T anomaly vs 1926-1935 (C)	Linear trend 1926-2025 (C/decade)
Q1: 0.00-0.60%	53	1.41	1.66	0.140
Q2: 0.61-1.61%	53	0.97	1.18	0.081
Q3: 1.62-2.61%	53	0.80	0.99	0.067
Q4: 2.72-4.09%	52	0.86	1.17	0.076
Q5: 4.11-6.26%	52	0.95	1.18	0.080
Q6: 6.30-8.44%	52	1.21	1.21	0.114
Q7: 8.57-12.49%	52	1.41	1.41	0.137
Q8: 12.52-18.61%	52	1.76	2.02	0.172
Q9: 18.63-38.57%	52	2.02	2.33	0.206

Filter sensitivity

Min good years	Max bad streak	Stations
100	0	471
99	0	471
98	0	471
100	1	471
99	1	727
98	1	831

BU validation checks

City	5 km mean BU (%)
Munich	24.8
Tokyo	40.6

Footnotes and Data Sources

NOAA National Centers for Environmental Information, GHCN-M v4 monthly mean temperature archive, QCU file used here: ghcnm.tavg.latest.qcu.tar.gz.
European Commission Joint Research Centre, GHSL built-up surface 2020, 1 km World Mollweide grid: download archive.
Natural Earth world boundaries, 1:110m admin-0 countries GeoJSON, used only for base map context: GeoJSON source.
The notebook companion keeps figures and the report inline in memory instead of writing notebook-side exports, which keeps the tutorial path transparent.