Centre for the Study of Regional Development,
School of Social Sciences
Jawaharlal Nehru University, New Delhi
Invites you to a student Workshop on
Statistical learning with climate data using R
by
Dr. Prasenjit Acharya
Department of Geography, Vidyasagar University, West Bengal
Date : March 13, 2023 (Monday)
Time : Lecture - 10:00 am -12:00 pm &
Venue: Carto Lab, CSRD, SSS III (Ist Floor)
Tutorial /Hands-on Training - 2:30 pm - 4:30 pm
Abstract: Over the last two decades, progresses in the computing system have transformed the landscape of climate data analysis. From classical linear, non-linear modelling and auto-regression models to data-driven machine learning models, the scale of analysis in climate science, with the arrival of big data (both in space and time context), has increased multifold. With the availability of open-source packages in programming interfaces in conjunction with parallel computing systems to process these big data sets, climate projection at various space-time scales using statistical learning is one of the most sought-after areas in climate science. In this tutorial, a brief but comprehensive demonstration of regression analysis would be covered using the state-of-the-art open-source computing platform of R-Studio. The tutorial includes a fundamental idea of machine learning, the data requirements for it, and the difference between the classical regression approach and machine learning approach, including a practical demonstration of random forest regression, one of the machine learning approaches, using climate data. The practical problem which we will deal with in that session deals with aerosol optical depth (AOD), a measure which manifests the amount of pollution in the atmosphere. AOD is a column integrated measure of the extinction of light energy (in the visible channel) through absorption and scattering. As AOD depends on many environmental parameters such as temperature, wind speed, NO2, SO2 concentration, relative humidity, planetary boundary layer height etc., making them as co-variates, we can project the AOD magnitude over space-time complex. The tutorial also includes a hands-on demonstration of the deep neural network (DNN) using the same data set. DNN is a subset of data-driven machine learning methods featured with a network of nodes and connections, inspired by the human brain, which is capable of regressing a weighted sum method in combination with back-propagation system. The pros and cons of these two methods will be discussed towards addressing the best-fit model.
