38
Uncertainty & confidence in tree‐ring records at mul5‐decadal to centennial 5mescales Tim Osborn Clima5c Research Unit, School of Environmental Sciences, UEA with contribu5ons from Keith Briffa and Tom Melvin TwiGer: @TimOsbornClim 3 September 2015

Uncertainty & confidence in tree-ring records at centennial timescales

Embed Size (px)

Citation preview

Page 1: Uncertainty & confidence in tree-ring records at centennial timescales

Uncertainty & confidence in tree‐ring records at 

mul5‐decadal to centennial 5mescales 

Tim Osborn 

Clima5c Research Unit, School of Environmental Sciences, UEA   

with contribu5ons from 

Keith Briffa and Tom Melvin 

TwiGer: @TimOsbornClim 3 September 2015 

Page 2: Uncertainty & confidence in tree-ring records at centennial timescales

Tree‐ring measurements as climate proxies  

Problems include: •  Biological proxy – Mul5ple controlling processes (T, light, moisture, nutrients etc.) –  Age effects 

Advantages include: •  Precise, absolute da5ng –  Possible because mul5ple samples with annual resolu5on & strong common signal allows cross‐da5ng 

•  Mul5ple samples –  Stronger climate signal and allows uncertainty es5mates 

Page 3: Uncertainty & confidence in tree-ring records at centennial timescales

Realis5c error es5mates for proxy data & climate reconstruc5ons  

Interes6ng challenge, combining process understanding with sta6s6cal analysis –  Great poten5al for sta5s5cal analysis for tree‐rings because we have large samples with precise, absolute da5ng 

Two components (this talk focusses on first one): •  Confidence in the proxy record itself •  Confidence in the clima5c interpreta5on & reconstruc5on 

Page 4: Uncertainty & confidence in tree-ring records at centennial timescales

Es5ma5ng confidence in tree‐ring records ..1  

Centennial–millennial ambi6on but approaches oDen only measure annual–decadal behaviour 

–  (1) Expressed Popula5on Signal (EPS) –  where n = sample size (number of trees sampled) –  and rbar = mean inter‐series correla5on –  But rbar determined mostly by annual–decadal common signal 

–  (2) Calibra5on and verifica5on against climate observa5ons –  But result determined mostly by annual–decadal correspondence because limited length of instrumental record 

Esper et al. (2004) Climate reconstruc5ons: low‐frequency ambi5on and high‐frequency ra5fica5on. EOS 85, 113‐120 Wigley et al. (1984) On the average value of correlated 5me series… J Clim Appl Meteor 23, 201‐213 

EPS ! nr1+ n"1( )r

Page 5: Uncertainty & confidence in tree-ring records at centennial timescales

Es5ma5ng confidence in tree‐ring records ..2  

Errors tend to diminish from annual to decadal 6mescales, but may then increase again out to longer 6mescales –  Different mean growth rates, differences between trees’ life cycles, uncertainty in the expected growth–age curve 

 Centennial–millennial confidence can only be assessed by considering agreement between independent long records 

Briffa et al. (2013) Reassessing the evidence for tree‐growth and inferred temperature change… QSR 72, 83‐107 

Page 6: Uncertainty & confidence in tree-ring records at centennial timescales

Yamalia case study: Yamal & Polar Urals Living & sub‐fossil Larix sibirica trees 

Polar Urals: TRW & MXD 

Yamal: TRW, but bigger sample 

Fig. 1, Briffa et al. (2013) QSR Yamal: 473 sub‐fossil, 160 living 

Polar Urals: 61 sub‐fossil (+18 for MXD), 62 living 

Page 7: Uncertainty & confidence in tree-ring records at centennial timescales

Yamal chronology: Full sample mean & 

standard error  

 

Page 8: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 9, Briffa et al. (2013) QSR Red/Black: two versions of the overall mean chronology (thin lines where < 6 trees sampled) 

Cyan: +/‐2 SE of the mean for 50‐year smoothed data. Grey: sample counts 

                        1000                                            1400                                            1800                 2000 

                400 BC                                     0                                       400 CE                                800 

Page 9: Uncertainty & confidence in tree-ring records at centennial timescales

Yamal chronology: Calibrated JJ temperature reconstruc5on with error 

components 

Page 10: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 11, Briffa et al. (2013) QSR Red/Black: two versions of the overall mean chronology (thin lines where < 6 trees sampled) 

Cyan: +/‐2 SE of the mean for 50‐year smoothed data. Grey: sample counts 

                   1000                                                  1400                                                  1800                   2000 

Page 11: Uncertainty & confidence in tree-ring records at centennial timescales

Yamal chronology: Compare different sites 

Only for recent centuries where mul6ple sites with living trees are available 

 

Page 12: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 2, Briffa et al. (2013) QSR Mean chronology for each site and SUB for sub‐fossil (thin lines where < 4 trees sampled) 

1600  1800  2000 

1600  1800  2000 

Page 13: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 2, Briffa et al. (2013) QSR Mean chronology for each site and SUB for sub‐fossil (thin lines where < 4 trees sampled) 

BoGom panel smoothed with a 20‐yr spline 

1600  1800  2000 

1600  1800  2000 

Page 14: Uncertainty & confidence in tree-ring records at centennial timescales

Yamal chronology: Compare different growth 

rate samples Two independent chronologies by alloca6ng half samples to each group based on their mean growth rate aDer removing age effects & common (climate) signals 

 

Page 15: Uncertainty & confidence in tree-ring records at centennial timescales

Ring age (yr)   50              100                              200                               300                              400 

        0                                  500                               1000                               1500                              2000 Fig. 3, Briffa et al. (2013) QSR      ‐      Mean chronology for each growth‐rate class (thin lines where < 8 trees sampled) 

Page 16: Uncertainty & confidence in tree-ring records at centennial timescales

Yamal chronology: Compare with Polar Urals 

Two Polar Urals records (TRW & MXD) but both have much smaller sample sizes & 

span only 1 millennium MXD sensi6ve to a longer summer season 

Page 17: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 3, Briffa et al. (2013) QSR Mean chronology for each growth‐rate class (thin lines where < 8 trees sampled) 

Page 18: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. 3, Briffa et al. (2013) QSR Mean chronology for each growth‐rate class (thin lines where < 8 trees sampled) 

 r( Yamal , Polar Urals) = 0.82 (< 15 yr)   0.78 (15–100 yr)   0.69 (> 100 yr) 

 r( Yamal , Polar MXD) = 0.55 (< 15 yr)   0.72 (15–100 yr)   0.52 (> 100 yr) 

Page 19: Uncertainty & confidence in tree-ring records at centennial timescales

Closing remarks 

Confidence in tree‐ring records on century‐to‐millennium 6mescales: 

–  Don’t assume they are reliable.  Evaluate them –  Don’t assume errors are random.  Try to iden5fy systema5c biases –  Larger samples are needed to establish confidence at the longest 5mescales 

–  Clear advantage of tree‐rings: mul5ple, precisely dated samples 

Scope to extend this by collabora6on with sta6s6cians –  using hierarchical sta5s5cal models of the data genera5ng process –  extension to spa5o‐temporal models (this example is one site, but a widespread network exists across parts of the NH) 

Page 20: Uncertainty & confidence in tree-ring records at centennial timescales

Spare slides relevant for some ques5ons 

 

 

Page 21: Uncertainty & confidence in tree-ring records at centennial timescales

Qilian Mountains case study: NE Tibetan Plateau

Fig. 1, Yang et al. (2014) PNAS 1203  sub‐fossil + living 

Qilian Juniper, very long lived, slow growing, TRW only 

Page 22: Uncertainty & confidence in tree-ring records at centennial timescales

Qilian chronology: Compare different 

sites  Living and archaeological samples 

 

Page 23: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. SB2, Yang et al. (2014) PNAS 

Page 24: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. SB2, Yang et al. (2014) PNAS 

Page 25: Uncertainty & confidence in tree-ring records at centennial timescales

Qilian chronology: Compare different growth 

rate samples Four independent chronologies by alloca6ng quarter of samples to each group based on their mean growth rate aDer removing age 

effects & common (climate) signals 

 

Page 26: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. SB9, Yang et al. (2014) PNAS 

Page 27: Uncertainty & confidence in tree-ring records at centennial timescales

Fig. SB10, Yang et al. (2014) PNAS 

Page 28: Uncertainty & confidence in tree-ring records at centennial timescales

 Sampling sites near the edge of a 

species’ ecological range to obtain a single dominant 

growth control Near the la6tudinal or eleva6onal tree line 

(can also reduce influence of stand dynamics where this coincides with a more open canopy forest) 

 

Page 29: Uncertainty & confidence in tree-ring records at centennial timescales

Magnified view 

Pine tree 

Processed wood core 

Page 30: Uncertainty & confidence in tree-ring records at centennial timescales

Natural archives of past climate?

Page 31: Uncertainty & confidence in tree-ring records at centennial timescales

Trees cannot grow above/beyond their tree-line  In very cold regions, tree growth is limited by temperature 

Page 32: Uncertainty & confidence in tree-ring records at centennial timescales

 Polar Urals root samples Difference in mean ring width between bole (trunk) & root collar samples can 

result in biased tree‐ring record 

  

Page 33: Uncertainty & confidence in tree-ring records at centennial timescales

 Example of sampling bole (trunk) at “breast height” Tom Melvin, Northern Norway, 1 June 1999 

Page 34: Uncertainty & confidence in tree-ring records at centennial timescales

 Example of root collar with large diameter than the bole Remnant Scots Pine, near Torneträsk, Sweden 

Page 35: Uncertainty & confidence in tree-ring records at centennial timescales

Polar Urals TRW record: original used bole samples, “update” included root collar samples

Page 36: Uncertainty & confidence in tree-ring records at centennial timescales

 Iden5fying previous periods with elevated tree mortality 

End‐aligned spans of tree samples, but note that sub‐fossil segments may not be complete 

(outer rings not preserved, pith not sampled)   

Page 37: Uncertainty & confidence in tree-ring records at centennial timescales

Khadytla River One of the living-tree

sites in the Yamal region

Page 38: Uncertainty & confidence in tree-ring records at centennial timescales

-6000 -4000 -2000 0 2000Year BC/AD

0

100

200

300

400

500

600

700

800

900

Num

ber o

f sta

cked

sam

ples

Example for Tornetrask, northern Sweden