Page 47 - Informatics_July_2023

International eGov Update

Cutting-Edge AI/ML System for Accurate Monsoon Rainfall

Predictions in India

state-of-the-art AI/ML model has been developed collaboratively
by the Department of Science and Technology Centre of Excellence
in
A Climate Modeling at the Indian Institute of Technology in Delhi
(IIT-Delhi), the Indraprastha Institute of Information Technology (IIIT-Delhi),
and universities in the United States and Japan. This cutting-edge model
accurately predicts monsoon rainfall, providing valuable information for
sectors such as agriculture and water resource management.
The AI/ML model predicts an All India Summer Monsoon Rainfall
(AISMR) of approximately 790 mm for the upcoming monsoon season, indi-
cating a normal monsoon this year. To make this prediction, the model
utilized historical AISMR data, the Niño3.4 index, and categorical Indian
Ocean Dipole (IOD) data from 1901 to 2001. Its performance surpasses that
of existing physical models used for monsoon predictions in the coun-
try, achieving an impressive forecast success rate of 61.9% during the test
period from 2002 to 2022.
The model can make forecasts several months in advance, contingent 68.4% of eroded soil in India. Globally, rainfall-induced soil erosion poses
upon the availability of Niño3.4 index and IOD forecasts. These inputs can a significant environmental challenge. However, traditional assessments
be continuously updated to reflect evolving conditions. The data-driven in India are often limited to specific catchments or regions, hindering a
models offer flexibility and better capture the nonlinear relationships comprehensive evaluation in a geographically diverse country like India.
among monsoon drivers, while being computationally efficient. To address this, a study at IIT-Delhi conducted the first-ever pan-India
Accurate monsoon predictions have significant implications for criti- assessment of rainfall erosivity. By utilizing multiple national and global
cal decision-making across multiple socio-economic sectors, including gridded precipitation datasets, the researchers created a high-resolution
agriculture planning, energy resource management, water resource utili- map identifying erosion-prone areas in India. This step contributes to
sation, disaster management, and addressing health concerns. The tech- building a national-scale soil erosion model, enabling watershed manag-
niques developed in this study will also be extended to provide state-wise ers to identify and prioritize locations for essential watershed develop-
monsoon rainfall predictions, enhancing their regional applications. ment activities to mitigate soil erosion.
Rainfall erosivity, a key factor in soil degradation, affects approximately Source- https://opengovasia.com/

Machine Learning to Resolve Aircraft Stability and Evasion

Challenge

IT researchers have developed an innovative method inspired
by the movie “Top Gun: Maverick” to address complex challenges
M related to aircraft stability and evasion. Using a machine learning
approach, their technique surpasses current safety standards and enhances
stability by a factor of ten. The method successfully guided a virtual fighter
jet through a narrow passage, impressing even experts in high-dimensional
dynamics.
Traditional approaches simplify complex stabilise-avoid problems using
mathematical techniques, while reinforcement learning trains an agent
through trial and error. However, balancing stability and obstacle avoidance
in these problems is challenging. The MIT researchers tackled the problem
in two steps: redefining it as a constrained optimization problem and trans-
forming it into the epigraph form for deep reinforcement learning.
To handle the epigraph form, the researchers derived new mathematical
expressions specific to their system. They combined these with established
engineering techniques, creating a controller that outperformed baselines
by preventing crashes and achieving stable alignment with the desired goal.
This technique has potential applications in designing controllers for the optimization process and testing the algorithm on physical hardware.
dynamic robots and assisting in stabilising autonomous vehicles. It excels in The researchers aim to bridge the gap between model dynamics and real-
extreme scenarios, providing reinforcement learning with safety and stabili- world dynamics for practical implementation.
ty guarantees for mission-critical systems.
Future enhancements will focus on better handling uncertainty during Source- https://news.mit.edu/

July 2023 informatics.nic.in 47