BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.2.3.1//EN TZID:Asia/Kolkata X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT UID:143@cds.iisc.ac.in DTSTART;TZID=Asia/Kolkata:20250908T100000 DTEND;TZID=Asia/Kolkata:20250908T110000 DTSTAMP:20250828T133556Z URL:https://cds.iisc.ac.in/events/ph-d-thesis-defense-102-cds-seminar-hall -08-september-2025-investigation-of-the-indian-summer-monsoon-rainfall-usi ng-statistical-and-machine-learning-techniques/ SUMMARY:Ph.D. Thesis Defense: #102: CDS Seminar Hall: 08\, September 2025 " Investigation of the Indian Summer Monsoon Rainfall Using Statistical and Machine Learning Techniques" DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Def ense\n\n\n\nSpeaker : Ms. AKANKSHA MANOJ RAJAK\nS.R. Number : 06-18-01-10- 12-18-2-16460\nTitle : "Investigation of the Indian Summer Monsoon Rainfal l Using Statistical and Machine Learning Techniques"\nResearch Supervisor : Dr. Deepak Subramani\nThesis examiner : Prof. Srinivasa Ramanujam Kannan \, Indian Institute of Technology\, Bhubaneswar\nDate &\; Time : Septem ber 08\, 2025 (Monday)\, 10:00 AM\nVenue : # 102 CDS Seminar Hall\n\n\n\nA BSTRACT\n\nThe Indian Summer Monsoon is an important atmospheric phenomeno n\, marked by a characteristic seasonal wind reversal pattern\, delivering 70 to 90% of the annual rainfall to the Indian subcontinent. Monsoon rain profoundly impacts agricultural productivity\, water resource management\ , and thereby the Indian economy. The first significant event of the monso on season is its start\, called Monsoon Onset over Kerala (MOK)\, which oc curs around the first week of June. After that\, the monsoon rain makes it s way to the central India region in the next two to three weeks and cover s the entire country by mid-July. The season has active and break spells\, traditionally defined on the basis of the rain anomaly being more than or less than one standard deviation. The focus of the first part of this the sis is to understand the relationship of change points in rain\, outgoing longwave radiation\, winds\, and cloud cover to the MOK and rain spells. T he second part of this thesis focuses on forecasting the MOK one season ah ead using deep learning techniques\, reanalysis and remotely-sensed data.\ n\nIn the first part of the thesis\, statistical change point detection me thods are applied to identify significant transitions in monsoon patterns across different regions in India subcontinent.\n\nChange point detection identifies the time at which a statistically significant change in the da ta generation properties occurs in a time series\, making it inherently su itable for detecting monsoon transitions in an unsupervised manner. We foc us on two critical regions: Kerala\, the traditional entry point of the so uthwest monsoon into mainland India\, and Central India\, a core monsoon z one. For these regions\, we implement both Pruned Exact Linear Time (PELT) offline algorithm and Bayesian Online Change Point Detection (BOCD) metho ds. These techniques are applied to univariate rainfall time series as wel l as multivariate datasets incorporating outgoing longwave radiation\, win d patterns\, and cloud cover from April 1 onward each year. In the Kerala region\, our change point analysis reveals significant shifts in monsoon d ynamics over the 1975-2024 period. The mean and standard deviation of rain fall have increased for the month of May while decreasing for June and Jul y during the last quarter of the 20th century. We observe a 3% increase in overall rainfall variability despite a 3% decrease in the southwest monso on contribution to annual rainfall. These findings highlight the increasin g pre-monsoon activity and evolving monsoon characteristics that challenge conventional fixed threshold-based definitions. We propose using the firs t significant rainfall change point as an objective proxy for Monsoon Onse t over Kerala (MOK)\, demonstrating several advantages over the IMD’s co nventional threshold-based criteria\, particularly in its ability to adapt to changing baseline conditions. When extending this methodology to Centr al India\, we discover a consistent propagation lag of 21-24 days between corresponding change points in Kerala and Central India\, providing a stat istical framework for understanding monsoon progression across regions. Ou r analysis further reveals fundamental constraints in monsoon behavior: ap proximately 57.8% of all transitions either originate from or remain in no rmal conditions\, with a complete absence of direct transitions between ex treme states (break-to-active or active-to-break). These findings establis h a robust approach for characterizing interannual variability patterns\, the frequency of high and low intensity rainfall periods\, and the spatial propagation of monsoon signals\, with significant implications for intras easonal forecasting.\n\nIn the second part of the thesis\, we transition f rom detecting monsoon onset to predicting it through the development of a dvanced deep learning approaches. Our research progresses from classificat ion to precise date prediction\, and from static to temporal models\, each with increasing sophistication and performance. First\, we establish a b aseline approach by developing a CNN-based classification model for catego rizing Monsoon Onset over Kerala (MOK) timing into Early\, Normal\, and De layed classes. This model uses monthly mean sea surface temperature (SST) data from March\, providing approximately a 60-day lead time for predictio ns. The network was initially trained from scratch on ERA5 reanalysis data \, achieving an accuracy of 60.0% on the test dataset. We then implemented an efficient transfer learning approach\, fine-tuning the pre-trained ERA 5 model on MODIS satellite data\, which resulted in an accuracy of 53.3%. Notably\, this represents a substantial improvement over the 40.0% accurac y obtained when applying the ERA5-trained model directly to MODIS data wit hout fine-tuning\, where the model failed entirely to identify Early and N ormal onset classes (0% recall for both classes). The fine-tuning effectiv ely addressed the systematic differences between datasets observed in the Arabian Sea\, Bay of Bengal\, and Indian Ocean regions\, improving the mac ro-average F1-score from 0.190 to 0.564. This adaptation demonstrates the model’s flexibility across different data sources and its potential for satellite-based operational forecasting. Gradient-weighted Class Activatio n Mapping (GradCAM) visualizations revealed valuable insights into the mod el’s decision-making process\, highlighting how it focuses on specific r egions in the Arabian Sea and Bay of Bengal when making predictions for di fferent onset categories.\n\nWhile effective\, this initial model relies o nly on monthly mean data without incorporating temporal evolution of meteo rological conditions. To address this limitation\, we developed more sophi sticated spatio-temporal neural network architectures that capture the tem poral dynamics of pre-monsoon conditions. Two distinct models were created : a March model using data available through March 31 (providing a 60-day lead time) and an April model incorporating data through April 30 (providi ng a 30-day lead time). These models were trained to process multiple mete orological variables from ERA5 reanalysis\, including sea surface temperat ure (SST)\, total cloud cover (TCC)\, net top thermal radiation for clear sky (TTRC/OLR)\, winds at 850 hPa and 200 hPa pressure levels\, and mean s ea level pressure (MSLP). Using data from 1940 to 2009 for training and va lidation\, with 2010 to 2024 reserved for testing\, we systematically eval uated the predictive capacity of individual variables and their combinatio ns. Our most effective architecture integrates all meteorological variable s into a unified spatio-temporal framework\, achieving remarkable performa nce with an RMSE of approximately 2 days\, a correlation coefficient of 0. 89\, and a 30-day lead time. The implementation of early stopping and L2 r egularization techniques effectively prevented overfitting\, creating a mo del that balances computational efficiency with prediction accuracy. The m odel demonstrates consistent performance across different test periods\, s uggesting robust real-world applicability. The enhanced prediction capabil ities of these developed models provide valuable lead time for agricultura l planning and water resource management\, potentially improving decision- making processes for marginal farmers\, crop insurance providers\, and gov ernment irrigation departments across India. By establishing a progression from simple classification to precise onset date prediction\, this resear ch contributes practical tools that can be implemented in operational fore casting settings to support communities and economies dependent on accurat e monsoon predictions.\n\n\n\nALL ARE WELCOME CATEGORIES:Events,Thesis Defense END:VEVENT BEGIN:VTIMEZONE TZID:Asia/Kolkata X-LIC-LOCATION:Asia/Kolkata BEGIN:STANDARD DTSTART:20240908T100000 TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE END:VCALENDAR