The influence of anthropogenic surface processes and inhomogeneities on gridded global climate data

1
The influence of anthropogenic surface processes
and inhomogeneities on gridded global climate data

• Ross McKitrick
• Department of Economics
• University of Guelph
• Guelph ON Canada
• Presentation to the
• American Chemical Society
• Denver CO via Webinar
• August 28 2011

2
Surface Climate Data

• The global temperature

3
Summary

• Climate data is the output of a model
• Raw data daily T-Min and T-Max readings from
  inhabited places
• This isnt what the climate analyst is interested
  in it must be converted into climate data
  using a statistical adjustment model.
• How do we know the adjustment model works?
• Many papers merely describe the adjustment steps
  in enthusiastic detail
• I have focused on devising statistical tests of
  the results

4
Conclusions

• Based on analysis of multiple data sets, and
  after addressing a long list of statistical
  rebuttals, I find the evidence convincing that
• The adjustment models are inadequate
• The resulting climate record over land is
  contaminated with patterns of socioeconomic
  development
• This adds a net warming bias to the global trend
  and may lead to misattribution of spatial
  patterns to greenhouse gases
• A valid empirical model of the spatial pattern of
  observed warming must include anthropogenic
  surface processes

5
Papers

• McKitrick, Ross and Patrick J. Michaels (2004).
  A Test of Corrections for Extraneous Signals in
  Gridded Surface Temperature Data Climate
  Research 26 pp. 159-173.
• McKitrick, Ross R. and Patrick J. Michaels.
  (2007) Quantifying the influence of
  anthropogenic surface processes and
  inhomogeneities on gridded surface climate data.
  Journal of Geophysical Research-Atmospheres 112,
  D24S09, doi10.1029/2007JD008465.
• McKitrick, Ross R. and Nicolas Nierenberg (2010)
  Socioeconomic Patterns in Climate Data. Journal
  of Economic and Social Measurement, 35(3,4) pp.
  149-175. DOI 10.3233/JEM-2010-0336.
• McKitrick, Ross R. (2010) Atmospheric
  Oscillations do not Explain the
  Temperature-Industrialization Correlation. Statis
  tics, Politics and Policy, Vol 1 No. 1, July
  2010.
• rossmckitrick.weebly.com

6
Core Methodology

• There is a spatial pattern of warming and
• cooling trends since 1980
• Climate models predict the pattern as a response
  to
• GHGs, solar changes, etc.
• The predicted pattern is uncorrelated with
• spatial pattern of socioeconomic development
• But raw weather data is known to be correlated
  with socioeconomic development
• The adjustment models are supposed to remove
  these effects.
• Therefore If the adjustments are adequate, the
  climate data should be uncorrelated with
  socioeconomic patterns

7
Core Methodology

• There is a spatial pattern of warming and
• cooling trends since 1980
• Climate models predict the pattern as a response
  to
• GHGs, solar changes, etc.
• The predicted pattern is uncorrelated with
• spatial pattern of socioeconomic development
• But raw weather data is known to be correlated
  with socioeconomic development
• The adjustment models are supposed to remove
  these effects.
• Therefore If the adjustments are adequate, the
  climate data should be uncorrelated with
  socioeconomic patterns

8
Core Methodology

• There is a spatial pattern of warming and
• cooling trends since 1980
• Climate models predict the pattern as a response
  to
• GHGs, solar changes, etc.
• The predicted pattern is uncorrelated with
• spatial pattern of socioeconomic development
• But raw weather data is known to be correlated
  with socioeconomic development
• The adjustment models are supposed to remove
  these effects.
• Therefore If the adjustments are adequate, the
  climate data should be uncorrelated with
  socioeconomic patterns

9
Core Methodology

• There is a spatial pattern of warming and
• cooling trends since 1980
• Climate models predict the pattern as a response
  to
• GHGs, solar changes, etc.
• The predicted pattern is uncorrelated with
• spatial pattern of socioeconomic development
• But raw weather data is known to be correlated
  with socioeconomic development
• The adjustment models are supposed to remove
  these effects.
• Therefore If the adjustments are adequate, the
  climate data should be uncorrelated with
  socioeconomic patterns

10
Core Methodology

• Hypothesis
• spatial pattern of trends in surface climate
  data
• is uncorrelated with
• spatial pattern of socioeconomic development
• In a series of papers I have shown that this
  hypothesis is strongly rejected

11
Sources of climate data

• CRU, NOAA, NASA all produce global climate data
  products
• All rely on same underlying archive
• Global Historical Climatology Network (run by
  NOAA)
• The 3 data products are very similar since they
  all use the same input data and similar, though
  not identical, averaging methods

12
Sources of observational error

• Changing sample size
• Changing sample locations
• Build up of surrounding landscape
• Equipment changes
• Poor quality control
• Local air pollution
• Waste heat from buildings and traffic, etc.

13
GHCN sample 1885

• Locations of weather stations

14
GHCN sample 1925

• Locations of weather stations

15
GHCN sample 1945
16
GHCN sample 1965

• Locations of weather stations

17
GHCN sample 1985

• Locations of weather stations

18
GHCN sample 2005

• Locations of weather stations

19
GHCNsamplesize overtime
20
GHCN fraction of sample fromurban airports
21
Climate data the record as if the land surface
was never modified and equipment never varied

• Temp data from cities adjustment
  algorithm True record

22
Structure of data set

• Cross-sectional
• Observational unit is a 5ox5o grid cell
• Dependent variable is 1979-2002 trend

23
Measurement Model

• Where
• qi observed climatic trend oC/decade
• Ti true trend
• f (Si) surface processes like urbanization and
  agriculture
• g (Ii) data inhomogeneities

24
For gridcell i

• Ti (ideal temperature trend) represented by
• TROPi trend in troposphere over same gridcell
  as measured by satellites

25
For gridcell i

• Surface processes f (Si) measured by
• pi growth in population density
• mi growth in real average income
• yi growth in real national GDP
• ci growth in national coal consumption

26
For gridcell i

• Inhomogeneities g (Ii) measured by
• gi GDP density (GDP per square km)
• ei availability of educated workers (sum of
  literacy postsecondary education)
• xi rate of missing observations ( missing
  months in cell)

27
Regression equation

• Surface proc. Inhom.
• GLS with clustering-robust std error matrix

28
First pair of studies

• McKitrick and Michaels (2004)
• Tested 218 raw series and corresponding CRU
  gridded data
• Both exhibited significant imprint of
  socioeconomic data with v. similar coefficients
• Adjustment hypothesis rejected at high
  confidence level
• McKitrick and Michaels (2007)
• Complete sample of (available) surface grid cells
• Independence hypothesis again rejected at high
  confidence level
• Both studies nonclimatic signals likely add up
  to a net warming bias in global average

29
2007 Results

• Probability that effects are zero
• Joint P 0.0000 (7x10-14)

30
Specification tests

• Bootstrap resampling
• Remove outliers, re-estimate
• RESET test
• Cross-validation tests
• Hausman endogeneity test (P 0.9962)

31
Generating clean trends

• Set GDP density and education to US levels
• Set all other surface and inhomogeneity effects
  to 0
• Use model coeffs to generate adjusted predicted
  values
• Observed average surface trend 0.30 oC/decade
• MSU average 0.23
• Adjusted average surface trend 0.17

32
IPCC Report

• How did the IPCC deal with this?
• IPCC AR4 page 244
• McKitrick and Michaels (2004) and De Laat and
  Maurellis (2006) attempted to demonstrate that
  geographical patterns of warming trends over land
  are strongly correlated with geographical
  patterns of industrial and socioeconomic
  development, implying that urbanisation and
  related land surface changes have caused much of
  the observed warming. However, the locations of
  greatest socioeconomic development are also those
  that have been most warmed by atmospheric
  circulation changes (Sections 3.2.2.7 and 3.6.4),
  which exhibit large-scale coherence. Hence, the
  correlation of warming with industrial and
  socioeconomic development ceases to be
  statistically significant.
• No supporting citation given

33
IPCC Report

• I obtained correlation fields between gridded
  temperatures and AO, ENSO and PDO

34
IPCC Report

• I augmented data sets for MM 2004 and MM 2007
  with circulation terms
• 2004 Model
• Circulation index effects are insignificant
• Including them anyway does not remove the
  significance of the conclusions
• 2007 Model
• Circulation index effects are jointly barely
  significant
• Including them increases size and significance of
  socioecononomic terms
• Conclusion IPCC claim is false.
• (McKitrick 2010, Statistics Politics and Policy
  July 2010)

35
IPCC Report

• I augmented data sets for MM 2004 and MM 2007
  with circulation terms
• 2004 Model
• Circulation index effects are insignificant
• Including them anyway does not remove the
  significance of the conclusions
• 2007 Model
• Circulation index effects are jointly barely
  significant
• Including them increases size and significance of
  socioecononomic terms
• Conclusion IPCC claim is false.
• (McKitrick 2010, Statistics Politics and Policy
  July 2010)

36
IPCC Report

• I augmented data sets for MM 2004 and MM 2007
  with circulation terms
• 2004 Model
• Circulation index effects are insignificant
• Including them anyway does not remove the
  significance of the conclusions
• 2007 Model
• Circulation index effects are jointly barely
  significant
• Including them increases size and significance of
  socioecononomic terms
• Conclusion IPCC claim is false.
• (McKitrick 2010, Statistics Politics and Policy
  July 2010)

37
IPCC Report

• I augmented data sets for MM 2004 and MM 2007
  with circulation terms
• 2004 Model
• Circulation index effects are insignificant
• Including them anyway does not remove the
  significance of the conclusions
• 2007 Model
• Circulation index effects are jointly barely
  significant
• Including them increases size and significance of
  socioecononomic terms
• Conclusion IPCC claim is false.
• (McKitrick 2010, Statistics Politics and Policy
  July 2010)

38
Schmidt (2009) Spurious correlation between
recent warming and indices of local economic
activity. International Journal of Climatology
10.1002/joc.1831

• 3 arguments against our findings
• surface temperature field exhibits spatial
  autocorrelation (SAC) so results are
  insignificant
• Use of RSS satellite series rather than UAH
  series removes significance of results
• Data generated by climate model yields apparent
  correlations with socioeconomic data, yet is
  uncontaminated by construction, so effects must
  be a fluke

39
Schmidt (2009) Spurious correlation between
recent warming and indices of local economic
activity. International Journal of Climatology
10.1002/joc.1831

• 3 arguments against our findings
• surface temperature field exhibits spatial
  autocorrelation (SAC) so results are
  insignificant
• Use of RSS satellite series rather than UAH
  series removes significance of results
• Data generated by climate model yields apparent
  correlations with socioeconomic data, yet is
  uncontaminated by construction, so effects must
  be a fluke

40
Schmidt (2009) Spurious correlation between
recent warming and indices of local economic
activity. International Journal of Climatology
10.1002/joc.1831

• 3 arguments against our findings
• surface temperature field exhibits spatial
  autocorrelation (SAC) so results are
  insignificant
• Use of RSS satellite series rather than UAH
  series removes significance of results
• Data generated by climate model looks correlated
  with socioeconomic data, yet is uncontaminated by
  construction, so effects must be a fluke

41
McKitrick NierenbergSocioeconomic patterns in
climate data J Econ Soc Measurement 2010

• Responses
• Schmidt did not actually test SAC. We do, and
  show that while depvar is ACd, regression
  residuals are not, as long as socioecon variables
  are included in model.
• Use of RSS data diminishes individual
  significance but effect due to a small number of
  outliers. Once these removed, RSS yields
  strongest results of all data sets
• Model-based data cannot replicate observed
  patterns predicts opposite signs

42
McKitrick NierenbergSocioeconomic patterns in
climate data J Econ Soc Measurement 2010

• Responses
• Schmidt did not actually test SAC. We do, and
  show that while depvar is ACd, regression
  residuals are not, as long as socioecon variables
  are included in model.
• Use of RSS data diminishes individual
  significance but effect due to a small number of
  outliers. Once these removed, RSS yields
  strongest results of all data sets
• Model-based data cannot replicate observed
  patterns predicts opposite signs

43
McKitrick NierenbergSocioeconomic patterns in
climate data J Econ Soc Measurement 2010

• Responses
• Schmidt did not actually test SAC. We do, and
  show that while depvar is ACd, regression
  residuals are not, as long as socioecon variables
  are included in model.
• Use of RSS data diminishes individual
  significance but effect due to a small number of
  outliers. Once these removed, RSS yields
  strongest results of all data sets
• Model-based data cannot replicate observed
  patterns predicts opposite signs

44
Data variations

• Surface
• Observed CRU, CRU2v, CRU3v
• Modeled GISS-E GCM average
• Troposphere
• Observed UAH, RSS
• Modeled GISS-E GCM average

45
Spatial Autocorrelation Tests
OBSERVED SAC DISAPPEARS
MODELS SAC REMAINS
46
Estimation with SAC model
47
Estimation with SAC model
MODELS INSIGNIFICANT
OBSERVATIONS SIGNIFICANT
48
GCM Counterfactual

• Schmidt 2009, p.2
• There is a relatively easy way to assess whether
  there is any true significance to these
  correlations. We can take fully consistent model
  simulations for the same period and calculate the
  distribution of the analogous correlations. Those
  simulations contain no unaccounted-for processes
  (by definition!) but plenty of internal
  variability, locally important forcings and
  spatial correlation. If the distribution
  encompasses the observed correlations, then the
  null hypothesis (that there is no contamination)
  cannot be rejected.

49
Results

50
Results
1 climate model reproduces observed effect,
0 failure to do so

51
Filtering results on surface data

• Set GDP density and education to US levels
• Set all other surface and inhomogeneity effects
  to 0
• Use model coeffs to generate adjusted predicted
  values

52
Filtering results on surface data

• Set GDP density and education to US levels
• Set all other surface and inhomogeneity effects
  to 0
• Use model coeffs to generate adjusted predicted
  values

53
Filtering results on surface data

• Set GDP density and education to US levels
• Set all other surface and inhomogeneity effects
  to 0
• Use model coeffs to generate adjusted predicted
  values
• This method should not reduce mean trend in GISS
  data

54
Conclusions

• In general, I reject the null hypothesis that
  adjustment models yield climate data
• socioeconomic patterns are highly significant
  across wide variety of specifications and data
  combinations
• socioeconomic data are necessary for
  well-specified error term
• This suggests a causal interpretation of the
  regression results

55
Responses to critiques

• IPCC claim that the results were statistically
  insignificant due to natural circulation
  patterns was a fabrication
• The claim was both unsubstantiated and untrue
• Various critiques have not held up
• SAC is not a source of bias
• Results hold up across numerous data sets
• Climate models cannot reproduce results

56
Thank you

• ross.mckitrick_at_uoguelph.ca
• rossmckitrick.weebly.com