Identifying the extreme precipitation circulation patterns over India in the future world using Convolutional Neural Networks.
Abstract of the project
Latest studies show that India has been experiencing a three-fold rise in overall extreme rainfall events over India since 1950 (Roxy, al., 2017). While this is predicted to happen on a warming planet, the understanding of physical mechanisms of how climate change alters the local and regional precipitation extremes over India remains incomplete. Moreover, the role of changes in atmospheric circulation needs to be discussed. The convolutional neural network (CNN) is an advanced machine learning tool that can be used to study the characteristics of large-scale atmospheric circulation patterns associated with extreme precipitation (Davenport and Diffenbaugh ., 2021). The CNN analysis can be carried out by using daily mean sea-level pressure, 500-hPa geopotential height, and vertical motion of air (ω) at 500 hPa anomalies from European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) and it can identify the precipitation extremes days over India with high statistical confidence. The India Meteorological Department (IMD) observed daily precipitation data (at 0.25o X 0.25o spatial resolution) can be used to train the CNN to identify the extreme precipitation circulation patterns (EPCP). We can understand the changes in the moisture flux, precipitation intensity, and frequency of EPCP and non-EPCP days over the early and late decades using the CNN analysis. After estimating the model weight of CNN using the observed precipitation dataset, CNN analysis (using the estimated model weights) can be used to investigate the future projections of extreme weather events over India. The Weather Research and Forecasting (WRF) Model simulated downscaled data over the Indian region with lateral boundary conditions from SSP (Shared Socioeconomic Pathways; SSP245 and SSP585) scenarios of Coupled Model Intercomparison Project Phase 6 (CMIP6) can provide more robust results for the future projection. We use a publicly available bias-corrected global gridded dataset based on 18 models from the CMIP6 and the ERA5 as the lateral boundary conditions for WRF simulation. The CNN analysis on the WRF downscaled dataset can provide an insight into the future projection of frequency and intensity of EPCP and non-EPCP days. Understanding the future projection of atmospheric circulation patterns responsible for extreme precipitation events is crucial as extreme weather phenomena have the most considerable impact on civil safety and the economy.
Project Sanction Order details: SERB-SRG/2022/001261
The project got the financial support from SCIENCE & ENGINEERING RESEARCH BOARD (SERB under Startup Research Grant (SRG) scheme.
Duration of the Project: 24 months.
Total budget of the project: Rs 2772980
Amal P Jose
Junior Research Fellow, SERB Project
Graduated MSc in atmospheric science from CUSAT, Kochi with distinction. Did thesis project at IITM, Pune on the identification of biases associated with CFSv2 coupled model. Awarded UGC NET in Environmental Sciences and CSIR NET in Earth Sciences. Worked at CSIR-4PI, Bangalore as project associate on “Anthropogenic Aerosol Emissions: Source Apportionment and Climate Effects”. Research interests are on the application of AI and ML to weather and climate related problems, Numerical weather modeling, Detection of extreme weather events.