Climate-vegetation classification: Linear Manifold Clustering as a means objectively to identify physical bifurcations, climate change, and GCM climate biases

Nancy Y. Kiang, NASA Goddard Institute for Space Studies, nancy.y.kiang@nasa.gov (Presenter)
Robert M. Haralick, City University of New York, haralick@aim.com
Art Diky, City University of New York, adiky@gc.cuny.edu
Xing Su, City University of New York, sunnyxing@gmail.com
Benjamin I. Cook, NASA Goddard Institute for Space Studies, benjamin.i.cook@nasa.gov

Climate classification has a millennia-long history of attempts to distinguish regions of the world that strongly differ in the terrestrial ecosystem types that they support, such as the well-known climate index schemes of Thornthwaite, Holdridge, and Köppen-Geiger. While these schemes can match a good diversity of vegetation types around the world, they remain broad, subjective, and imperfect indicators. Workers in the field have tried K-means clustering as an objective approach for classifying climate. This study examines Linear Manifold Clustering (LMC) as more appropriate. Whereas with K-means the cluster center is a point, LMC can better account for the temporal aspect of climate by clusters that may be around a point, a line, a plane, and so forth. We also examine the impact of spatial resolution of global observed (0.5° x 0.5°, 1° x 1°, and 2.5° x 2°) and simulated (2.5° x 2°) climate statistics on classification results. Results shown here are for the mid-20th c. (1951-1980), with observed monthly mean surface temperature and precipitation, and simulated climate from the NASA Goddard Institute for Space Studies general circulation model (GISS GCM), “GISS-E2-R NINT” configuration (Miller et al. 2014, Schmidt et al. 2014). The LMC clusters are compared to the Köppen-Geiger 34 climate classes. Physical characteristics for the clusters’ linear manifolds are interpreted relative to observed vegetation cover and plant physiological thresholds. This is part of a larger study that combines climate from the early 20th c. through present with plant physiological traits from the TRY database (Kattge et al. 2011), for which preliminary clustering results may be shown. We aim to define parsimonious sets of ecosystem and plant functional types (PFTs) for dynamic global vegetation models (DGVMs), delineate resolution limitations on capturing vegetation diversity, and identify biases in climates simulated with GCMs that may affect predictions of the carbon cycle.

Presentation Type:  Poster

Session:  General Contributions   (Tue 4:35 PM)

Associated Project(s): 

• Related Activity: Related Activity or Previously Funded CC&E Activity not listed ...details

Poster Location ID: 230