Anders Ahlström
Senior lecturer
Using PCA and Global Smoothing to Explore Differences between Global Vegetation Models
Author
Summary, in English
A common method for comparing the result of different global circulation models (GCMs) under different emission scenarios is to study global climate response variables, such as mean temperature. An interesting alternative measure of climate sensitivity is to study the biosphere’s response to the different climate scenarios. The Lund-Postdam-Jena (LPJ) global vegetation model and its extension LPJ-GUESS is a dynamic global vegetation model that can be coupled to GCMs and used to explore the effect of varying climates on vegetation and carbon uptake.
Using the output from different GCMs under different emission scenarios LPJ-GUESS can be used to generate global vegetation and carbon uptake patterns that are specific to each forcing climate scenario. We investigate if important regional and global differences exist between the vegetation patterns from different GCMs and emission scenarios. An important question is if potential differences are primarily due to the different emission scenarios or to the different GCMs.
In order for us to carry out the above analysis we need to both reduce the noise in the LPJ-GUESS predictions and reduce the vast amount of data. To accomplish both these goals we compute smooth principal components. A problem when computing the PCA and the smoothing is that LPJ-GUESS output is generated on a regular longitude-latitude grid, implying that both the size and distance between grid cells vary. To handle this irregular data on a sphere we use a Gaussian Markov random field (GMRF) approximation of Thin Plate Splines (TPS) that generalises the TPS to general manifolds (such as a sphere). The well known computational advantages of GMRFs greatly aids the analysis, given the large amount of data obtained from LPJ-GUESS.
Using the output from different GCMs under different emission scenarios LPJ-GUESS can be used to generate global vegetation and carbon uptake patterns that are specific to each forcing climate scenario. We investigate if important regional and global differences exist between the vegetation patterns from different GCMs and emission scenarios. An important question is if potential differences are primarily due to the different emission scenarios or to the different GCMs.
In order for us to carry out the above analysis we need to both reduce the noise in the LPJ-GUESS predictions and reduce the vast amount of data. To accomplish both these goals we compute smooth principal components. A problem when computing the PCA and the smoothing is that LPJ-GUESS output is generated on a regular longitude-latitude grid, implying that both the size and distance between grid cells vary. To handle this irregular data on a sphere we use a Gaussian Markov random field (GMRF) approximation of Thin Plate Splines (TPS) that generalises the TPS to general manifolds (such as a sphere). The well known computational advantages of GMRFs greatly aids the analysis, given the large amount of data obtained from LPJ-GUESS.
Department/s
- Mathematical Statistics
- Dept of Physical Geography and Ecosystem Science
- MERGE: ModElling the Regional and Global Earth system
Publishing year
2011
Language
English
Pages
3946-3952
Publication/Series
Proceedings of the 58th World Statistics Congress of the International Statistical Institute (ISI 2011)
Full text
- Available as PDF - 203 kB
- Download statistics
Document type
Conference paper
Publisher
International Statistical Institute
Topic
- Probability Theory and Statistics
- Physical Geography
Conference name
58th World Statistics Congress of the International Statistical Institute (ISI 2011)
Conference date
2011-08-21 - 2011-08-26
Conference place
Dublin, Ireland
Status
Published
ISBN/ISSN/Other
- ISBN: 978-90-73592-33-9