Presented by Tim Osborn to the Royal Meteorological Society's meeting on 'Measuring Climate Change' in October 2021

1. Extending the
modern
record back
in time using
proxy data
Tim Osborn
Climatic Research Unit, University of East Anglia
Royal Met Soc
Measuring climate change
Oct 2021

2. Focus / Outline
I will focus on how we use instrumental and proxy data together
Outline
• What is a climate proxy record? (Tree rings as an example)
• Desirable characteristics
• When does instrumental data get used?
• Networks of proxy records for reconstructing climate at large spatial scales

3. Holocene temperature changes at the global scale
Adapted from Box TS.1, Fig. 1 from Arias et al. (2021, in press) IPCC AR6 WGI Technical Summary

4. Low
emissions
High
emissions
Projected
change
by 2100
Adapted from Box TS.1, Fig. 1 from Arias et al. (2021, in press) IPCC AR6 WGI Technical Summary

5. What is a proxy record for climate?
An archive that builds up
+
Quantity that we can measure & that is influenced by local
weather/climate
=
CLIMATE PROXY

6. Example climate proxy: tree rings
Jones & Briffa coring an oak tree in Sotterley Park in 1982, with spectators

7. Subfossil tree section from the Austrian Alps, preserved in a lake

8. Photo: Tom Melvin, UEA

9. Photo: Tom Melvin, UEA

10. Photo: Tom Melvin, UEA
Early wood -------> Late wood
<-----------Ring width---------->

11. Photo: Tom Melvin, UEA
Early wood -------> Late wood
<-----------Ring width---------->

12. Adapted from Fig.1 of Björklund et al. (2017) New Phytologist 216, 728
x-ray photo of
tree sample
wood density
from x-ray photo
Tree ring width TRW
Tree rings: what can we measure? Width, density, colour, isotopes, cells

13. Pros and cons of tree-rings
• Biological organism, multiple influences
– Temperature, water, sunlight, etc.
– Nutrients, pests, competition for light, etc.
• Resolution, dating, replication, coverage
– Annual values, precisely dated, from many trees, from many different
environments
• How to minimise the cons by building on the pros?
– Replication with multiple samples – reduces uncorrelated noise/errors
– Comparison of multiple datasets – evaluate systematic errors
– Sample near the margins of the ecological range

14. Sampling near the margins of the ecological range: moisture-limited sites
Qilian juniper, NE Tibetan Plateau

15. Qilian Juniper, NE Tibetan Plateau
Prof Bao Yang & colleagues
Chinese Academy of Sciences, Lanzhou
Yang, Qin, Wang, He, Melvin, Osborn, Briffa (2014) A 3,500-year tree-ring record of annual precipitation
on the northeastern Tibetan Plateau. PNAS 111, 2903 (doi:10.1073/pnas.1319238111)

16. So where does instrumental data come into it?
Ideally, we understand the mechanisms that link the local
environment (including weather/climate) to the measured proxy
Nevertheless we almost never* have a precise model (physical,
biological, chemical) that links climate to our proxy
Fundamentally, reconstructing climate is an empirical activity
We use instrumental and proxy data to build an empirical model
*Borehole temperature profiles may be one exception

17. Ideally we don’t use instrumental data when developing or selecting the proxy record
Understand the mechanisms
Select & develop the record
Empirical calibration
(with instrumental data)
Reconstruction
(including error estimates)
Hypothesize about the mechanisms
Use instrumental data to test
hypothesis, select some records,
reject others
Empirical calibration
(with instrumental data)
Reconstruction
(including error estimates)

18. Example moisture-limited site: NE Tibetan Plateau
Yang, Qin, Wang, He, Melvin, Osborn, Briffa (2014) A 3,500-year tree-ring record of annual precipitation on the NE Tibetan Plateau. PNAS
1203 trees, ~800k rings
Average series length ~600 years
Dead & living span 3,500 years
All rings precisely dated

19. Example moisture-limited site: NE Tibetan Plateau
Yang, Qin, Wang, He, Melvin, Osborn, Briffa (2014) A 3,500-year tree-ring record of annual precipitation on the NE Tibetan Plateau. PNAS
3,500-year chronology of normalized tree growth
Not calibrated (not climate)
<0 = narrower rings than normal
>0 = wider rings than normal

20. Example moisture-limited site: NE Tibetan Plateau
Calibrated against annual precipitation
(1957-2011) via linear regression
r = 0.84

21. 3,500 years of annual precipitation for NE Tibetan Plateau
Yang, Qin, Wang, He, Melvin, Osborn, Briffa (2014) A 3,500-year tree-ring record of annual precipitation on the NE Tibetan Plateau. PNAS
Correlation with observed annual precipitation = 0.84

22. Reconstruction uncertainties
A comprehensive model of reconstruction errors should contain:
• Proxy uncertainties
• May grow further back in time
• Random or systematic?
• Calibration errors (which use instrumental data)
• Are they correlated? Uncorrelated errors reduce more
quickly with averaging over time or space
• Structural errors (methodological choices)
• Sensitivity to proxy selection criteria
• Sensitivity to statistical calibration method

23. Sampling near the margins of the ecological range: temperature-limited sites
Siberian larch, Yamal Peninsula, Arctic Russia

24. Fig. 18 from Mazepa et al. (2011) Climate-Driven Change of the Stand Age Structure in the Polar Ural Mountains
Fig. 18 (Mazepa et al. 2011)
The bottom of eastern slope of
Malaya Chernaya Mountains
(66°50.751’N, 65°32.770’E,
286 m above sea level)
Yamal Peninsula &
Northern Polar Ural
Mountains: a region of
rapid temperature &
vegetation change

25. 2,000 years of summer temperature for Yamal (N Siberia)
Fig. 11, Briffa et al. (2013) QSR https://doi.org/10.1016/j.quascirev.2013.04.008
Last 1,000 yr
15-yr smoothing
Last 2,000 yr
100-yr smoothing
Correlation with observed June-July temperature = 0.70
Red:
observed summer T
Black:
reconstructed summer T with
95% confidence interval shaded
(pink: calibration uncertainty;
blue: chronology uncertainty)

26. Large spatial scales & networks of proxies
Many approaches to combine spatial and temporal data
• Spatial aggregation (e.g. composite plus scale)
• Spatially average proxies & instrumental data separately
• Then calibrate the spatial averages
• Utilise spatial patterns in proxies
• Identify dominant spatial patterns in proxies
• Use these patterns to reconstruct individual grid cells (e.g. point-by-point regression)
• Or use these patterns to reconstruct patterns in the instrumental data (e.g. principal
components regression)
• Use spatial patterns from climate models (e.g. last millennium reanalysis)
• Use instrumental data to calibrate “proxy system models”, i.e. proxy = f(climate)
• For each year, select many possible fields from model simulations, predict the implied
networks of proxy values, select/weight model fields to minimise differences between
predicted and observed proxy values in that year

27. 1,250 years of summer
temperature for NH land
(north of 40oN)
Wilson et al. (2016) N-TREND. Quaternary Science Reviews
Here are the
Yamal and Polar Urals records
54 tree-ring series
(maximum latewood density,
latewood blue intensity,
ring width)

28. 1,250 years of summer temperature for NH land (north of 40oN)
Wilson et al. (2016) N-TREND. Quaternary Science Reviews

29. 1,250 years of summer temperature for NH land (north of 40oN)
Wilson et al. (2016) N-TREND. Quaternary Science Reviews
A composite over 11 eruptions
demonstrates clear summer
cooling following explosive
volcanic eruptions

30. Global temperature reconstructions for last 2,000 years: PAGES2k
PAGES2k Consortium (2017) A global multiproxy database for temperature reconstructions… Sci. Data https://doi.org/10.1038/sdata.2017.88

31. Global temperature reconstructions for last 2,000 years: PAGES2k
PAGES2k Consortium (2019) Consistent multidecadal variability in global temperature recons… Nat. Geosci. doi:10.1038/s41561-019-0400-0
PAGES2k (2019) 257 records after regional temperature screening

32. Global temperature reconstructions for last 2,000 years: PAGES2k
PAGES2k Consortium (2019) Consistent multidecadal variability in global temperature recons… Nat. Geosci. doi:10.1038/s41561-019-0400-0

33. Recap & Caveats
• Huge global effort to develop so
many temperature-sensitive proxies
• Instrumental data used for
screening proxies, weighting & calibrating, quantifying errors
– Using instrumental data in both selecting and calibrating has to be taken into account
– Number of apparently temperature-sensitive records removed in screening is large
– Screening, calibration & testing all more robust if longer overlap with instrumental
record (a longer & more reliable early instrumental record will help)
– Trends can get false sense of confidence: capturing detrended variability is a more
powerful test
– Becomes more problematic with low resolution records (e.g. sediments), though
PAGES2k have gone to some lengths address this – comparison with high resolution
proxies

34. Uncertainties & assessment
• Remember the reconstruction
error terms I mentioned earlier?
Which have been addressed in the
PAGES2k (and hence IPCC assessment)?
– Proxy record uncertainties – partially
– Calibration uncertainties – yes
– Structural uncertainties in statistical calibration methods – yes
– Structural uncertainties in proxy selection – partially
• IPCC AR6 assessment takes these limitations into account
– Last decade global T more likely than not higher than any multi-century average during
the Holocene. Rate of global warming during last 50 years unprecedented in at least
the last 2,000 years (medium confidence)