![Page 1: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/1.jpg)
Reconstructing the past land cover using fossil pollen
estimates
Behnaz Pirzamanbin Lund university
Sweden
![Page 2: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/2.jpg)
Collaborators:
2 Behnaz Pirzamanbin
Marie-Jose Gaillard Department of Biology and
Environmental Sciences, Linnaeus University, Sweden
Shinya Sugita Institute of Ecology, Tallinn
University, Estonia
Palaeoecology
Anneli Poska Department of Physical Geography
and Ecosystems Analysis, Lund University, Sweden
Benjamin Smith Department of Physical
Geography and Ecosystems Analysis, Lund University,
Sweden
Climate modeling
Johan Lindström Centre for Mathematical Sciences,
Lund University, Sweden
Mathematical Statistics
![Page 3: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/3.jpg)
![Page 4: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/4.jpg)
Vegetation (Land-cover) is a inherent part of climate system.
![Page 5: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/5.jpg)
Fossil pollen has been extensively used to estimate past vegetation in regional and global scales.
→ Pine Birch
→ →
![Page 6: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/6.jpg)
REVEALS model (Regional Estimates of VEgetation Abundance from Large Sites)
(Sugita S. 2007*)
REVEALS model
Pollen count
REVEALS Vegetation proportions
Dispersal & deposition
Relative pollen
productivity
*) Sugita S. 2007 .Theory of quantitative reconstruction of vegetation I: pollen from large sites REVEALS regional vegetation composition. The Holocene 17 (2), 229-241.
![Page 7: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/7.jpg)
PFT Description LCT TBE1 TBE2 IBE TSE
Shade-tolerant-boreal (Picea) Shade-tolerant-temperate (Abies) Shade-intolerant-boreal (Pinus) Tall shrub evergreen trees(Juniperus)
Coniferous forest
IBS TBS TSD
Shade-intolerant–boreal (Alnus) Shade-tolerant-temperate (Carpinus) Tall shrub summergreen trees(Salix)
Broadleaved forest
LSE GL AL
Low evergreen shrub (Calluna) Grassland – all herbs (Filipendula) Agricultural land- cereals (Cereals)
Unforested land
Pollen data: • 25 taxa • 10 plant functional types (PFT) • 3 groups, land-cover types (LCT)
![Page 8: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/8.jpg)
Goals
• To construct a model that explains Reveals data for present day time window, Present day (1850- 1950),
• To reconstruct land cover at locations with no Reveals estimates for past time windows, 200 BP (1600-1850 AD), and 600 BP 3000 BP 6000 BP (4250-3750 BC),
Behnaz Pirzamanbin 8
![Page 9: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/9.jpg)
Data
• REVEALS land cover proportions, Pollen based, 24 taxa, 10 plant functional types (PFTs), and 3 land
cover types (LCTs), Present day, 200 years before present (BP), 6000 BP.
• LPJ-GUESS potential land cover proportions, Vegetation model force by bio-climate variables
• KK10 estimates of human land use proportions.
9 Behnaz Pirzamanbin
![Page 10: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/10.jpg)
Data:
10 Behnaz Pirzamanbin
![Page 11: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/11.jpg)
LPJ-GUESS adjusted with KK10
11
P(𝐶𝐶𝑎𝑎𝑎𝑎𝑎𝑎)=P(C)∗(1−P(HLU))
P(𝐵𝐵𝑎𝑎𝑎𝑎𝑎𝑎)=P(B)∗(1−P(HLU))
P(𝑈𝑈𝑎𝑎𝑎𝑎𝑎𝑎)=P(U)∗(1−P(HLU))+P(HLU)
Behnaz Pirzamanbin
![Page 12: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/12.jpg)
Compositional data: For Land cover data set, 𝑦𝑦𝑖𝑖 , 𝑖𝑖 = 1, … ,𝐷𝐷 = 3 with
�𝑦𝑦𝑖𝑖 𝑠𝑠 = 1, 𝑦𝑦𝑖𝑖 ∈ 0,1 . 𝑖𝑖
If we model 𝑦𝑦𝑖𝑖 using linear regression
𝑦𝑦𝑖𝑖 = 𝑋𝑋𝛽𝛽𝑖𝑖 + 𝜀𝜀𝑖𝑖 , 𝑦𝑦�𝑖𝑖 = 𝑋𝑋�̂�𝛽𝑖𝑖 but
𝑦𝑦�𝑖𝑖 ∉ 0,1 . We use the log transform for compositional data, j = 1, 2
𝑢𝑢𝑎𝑎 𝑠𝑠 = 𝑙𝑙𝑙𝑙𝑙𝑙𝑦𝑦𝑎𝑎(𝑠𝑠)𝑦𝑦𝐷𝐷(𝑠𝑠) , 𝑢𝑢𝑎𝑎 ∈ −∞,+∞ .
Back transformation of the data,
𝑦𝑦𝑎𝑎 𝑠𝑠 =exp (𝑢𝑢𝑎𝑎(𝑠𝑠))
1 + ∑ exp (𝑢𝑢𝑎𝑎(𝑠𝑠))𝑎𝑎, 𝑦𝑦𝐷𝐷 𝑠𝑠 =
11 + ∑ exp (𝑢𝑢𝑎𝑎(𝑠𝑠))𝑎𝑎
,
Behnaz Pirzamanbin 12
![Page 13: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/13.jpg)
Transfromation of compositional data:
13 Behnaz Pirzamanbin
![Page 14: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/14.jpg)
Model Land cover types 𝑦𝑦𝑖𝑖 , 𝑖𝑖 = 1, … ,𝐷𝐷 = 3 , 𝑢𝑢𝑎𝑎 𝑠𝑠 = 𝑙𝑙𝑙𝑙𝑙𝑙 𝑦𝑦𝑖𝑖(𝑠𝑠)
𝑦𝑦𝐷𝐷(𝑠𝑠), 𝑗𝑗 = 1, 2,
𝑢𝑢1𝑢𝑢2 = A
z1z2 +
µ1µ2 + 𝜀𝜀𝑖𝑖 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝜀𝜀𝑖𝑖 ∈ 𝑁𝑁 (0, 𝐼𝐼𝜎𝜎2)
• µ is mean field and modeled as a linear regression
𝜇𝜇𝑖𝑖 = 1.𝛽𝛽0,𝑖𝑖 +�𝐵𝐵𝑝𝑝𝛽𝛽𝑝𝑝,𝑖𝑖𝑝𝑝
• B is covariate matrix
B=Coniferous
BroadleavedElevation
T,
• Z is spatial dependency field and modeled using SPDE/GMRF
z1z2 ∈ 𝑁𝑁 0, 1 𝜌𝜌
𝜌𝜌 1 ⊗ 𝑄𝑄−1 .
Behnaz Pirzamanbin 14
![Page 15: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/15.jpg)
Validation of model:
To check the model performance • We evaluate the model with present-day
European Forest Institue Forest Map (EFI-FM) data 2006 vs. 1950
• We use 6-block cross-validation
16 Behnaz Pirzamanbin
![Page 16: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/16.jpg)
Present-day
17 Behnaz Pirzamanbin
![Page 17: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/17.jpg)
EFI-FM & REVEALS vs. IGMRF & RM
18 Behnaz Pirzamanbin
![Page 18: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/18.jpg)
Comparison
20 Behnaz Pirzamanbin
![Page 19: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/19.jpg)
200 BP
21 Behnaz Pirzamanbin
![Page 20: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/20.jpg)
6000 BP
22 Behnaz Pirzamanbin
![Page 21: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/21.jpg)
Results:
• In general the reconstructed maps from both models are close to EFI-FM,
• IGMRF reconstructed maps capture the local variability of REVEALS data,
• RM reconstructed maps smooth REVEALS data.
23 Behnaz Pirzamanbin
![Page 22: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/22.jpg)
New study’s goal
• Estimating the human land use, • Investigating what LCTs is most likely to be
used by humans.
24 Behnaz Pirzamanbin
![Page 23: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/23.jpg)
Hierarchical model:
25 Behnaz Pirzamanbin
• Data model: P(𝑦𝑦|𝛼𝛼, 𝑧𝑧(X))~Dir(𝛼𝛼. 𝑧𝑧(𝑋𝑋))
• Latent field: P X 𝜃𝜃 ~𝑁𝑁(0,𝜌𝜌⊗𝑄𝑄−1 𝑎𝑎 )
• Unknown parameters: 𝛼𝛼,𝜌𝜌,𝑎𝑎
![Page 24: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/24.jpg)
Hierarchical model: • Data model:
P(𝑦𝑦|𝛼𝛼, 𝑧𝑧(X))~Dir(𝛼𝛼. 𝑧𝑧(𝑋𝑋)) • Latent field:
P X 𝜃𝜃 ~𝑁𝑁(0,𝜌𝜌⊗𝑄𝑄−1 𝑎𝑎 ) • Unknown parameters: 𝛼𝛼,𝜌𝜌,𝑎𝑎
𝐶𝐶𝐵𝐵𝑈𝑈
~ 𝑎𝑎𝑙𝑙𝑤𝑤 X1, X2
𝐻𝐻𝐻𝐻𝑈𝑈 ~ 𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑙𝑙(X3)
𝐶𝐶𝑎𝑎𝑎𝑎𝑎𝑎𝐵𝐵𝑎𝑎𝑎𝑎𝑎𝑎𝑈𝑈𝑎𝑎𝑎𝑎𝑎𝑎
~C∗(1−HLU)B∗(1−HLU)
U∗(1−HLU)+HLU
26 Behnaz Pirzamanbin
![Page 25: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/25.jpg)
Latent field: P X 𝜃𝜃 ~𝑁𝑁(0,𝜌𝜌 ⊗ 𝑄𝑄−1 𝑎𝑎 ) Before, we considered the transformed data directly in the models, Now, we want to model the components on the right side of the above equations, 27 Behnaz Pirzamanbin
P(𝐶𝐶𝑎𝑎𝑎𝑎𝑎𝑎) = P(C)∗(1−P(HLU)) P(𝐵𝐵𝑎𝑎𝑎𝑎𝑎𝑎) = P(B)∗(1−P(HLU)) P(𝑈𝑈𝑎𝑎𝑎𝑎𝑎𝑎) = P(U)∗(1−P(HLU))+P(HLU)
![Page 26: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/26.jpg)
MCMC
To fit the model and estimate the parameters we use
1. MH: P 𝛼𝛼 𝑦𝑦, 𝑧𝑧 𝐗𝐗 2. MH: P 𝑎𝑎,𝜌𝜌 𝐗𝐗 3. MH, using “Taylor expansion” or “Riemann
manifold Langevin” ∶ P(𝐗𝐗|𝛼𝛼,𝑎𝑎,𝜌𝜌,𝑦𝑦)
Behnaz Pirzamanbin 28
![Page 27: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/27.jpg)
Future plan • To reconstruct the LCTs for specific regions
with higher temporal resolution, for example Baltic area, and consider the temporal dependency between the time windows.
• To reconstruct the 10 PFTs as specified in
climate models, since there exist locations where some of these PFTs are nonexistence (zero proportion). This is an interesting open statistical problem!
29 Behnaz Pirzamanbin
![Page 28: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/28.jpg)
Thanks
![Page 29: Reconstructing the past land cover using fossil pollen estimates...2014/06/24 · • LPJ-GUESS potential land cover proportions, Vegetation model force by bio -climate variables](https://reader035.vdocuments.us/reader035/viewer/2022071510/612eb14a1ecc51586942f978/html5/thumbnails/29.jpg)
Did you know that Badminton is • the world fastest racket sport & • the second largest participation sport in the world (after football) 31