Please join us on Thursday, June 27, 2019, from 2:00-4:15 PM, in room 214 Mudd for a SEAS Colloquium in Climate Science Mini-Workshop on Tropical Convection. The first talk will start at 2:00 PM and will be followed by at break from 3:00-3:15 PM. The second talk will start at 3:15 PM.
___________________________________________________
Talk #1: “Is atmospheric convection organized?: Information entropy analysis”
Speaker: Jun-ichi Yano, Meteo France, Toulouse
Abstract: In order to quantify the degree of organization of atmospheric convection, an analysis based on the information entropy, which is widely considered a measure of organization in information science, is performed. Here, the information entropy is defined in terms of the spectrum of the empirical orthogonal functions (EOFs). Satellite-based brightness temperature data from CLAUS (Cloud Archive User Service) is used over the domain covering the Indian Ocean and the Western Pacific with a spatial resolution of 2/3^degree from January 1985 to June 2009. The information entropy remains close to a mean value of 0.899 with a very small standard deviation of $2.7 imes 10^{-3}$, suggesting that the atmospheric convection is always disorganized under a measure of the information entropy, which is against our common understanding. To better interpret this result, some basic theoretical analyses are performed, and the values of the information entropy for different systems (English literature texts, turbulent flows) from previous studies are reviewed. The same analysis is further performed on the Ising model, which is characterized by a clustering tendency of spin distribution, akin to convective organization morphologically, at the critical temperature. The study suggests a need for a careful use of the term “organized”. Atmospheric convection represents a tendency for clustering up to the planetary scale in analogous manner as the critical--point behavior of the Ising model. However, neither is considered an “ordered” state under a measure of the information entropy.
(This talk will be followed by a 15-minute break)
___________________________________________________
Talk #2: "A probabilistic model to detect spatio-temporal patterns in the Indian summer monsoon rainfall”
Speaker: Vishal Vasan, International Centre for Theoretical Sciences, Bengaluru
Abstract: In this talk I'll introduce a probabilistic representation of the Indian summer monsoon rainfall data (Rajeevan et al. 2006) based on a Markov random field consisting of discrete states representing high and low rainfall at grid-scale. Using this model on data for 8 years, we obtain from clustering algorithms, 10 robust spatial patterns of daily rainfall over the Indian landmass. Unlike traditional linear models of data, each day is assigned one precisely one pattern (and not a linear combination) that approximates the daily rainfall. Such approximations are quite accurate for nearly 95% of the days, including days outside the original 8 year window. I will further introduce a notion of families of spatial patterns distinguished by their total rainfall amount and geographic spread. The patterns are characterized based on transition probabilities and we identify most commonly occurring sequences of patterns. Lastly, I will try to highlight the successes and limitations of this approach, how one may attempt to generalize these ideas and how they compare to more popular methods based on the singular value decomposition and kmeans clustering.
Hosted by Prof. Adam Sobel
___________________________________________________
This workshop is free and open to the public. Advanced registration is not required.
