×

Developing a deep learning model for sub-seasonal to seasonal agriculture drought prediction system in India

Professor: Prof. Karthikeyan Lanka

Description:

The demand for water has increased globally due to the growth in population, which led to a substantial crop expansion to maintain food security. Droughts further exacerbate the water demand. Agricultural droughts – which occur due to deficit in soil water content – threaten the food security due to their impact on crop yields. There is a need to model droughts to assist the stakeholders (government and farmers) for an effective crop management to sustain the crop productivity. Deep learning gained importance recently due to the ability to learn from the abundant data that has been accumulated from reanalysis and satellite data sources. This project aims to setup a timeseries prediction model using state-of-the-art deep learning techniques to forecast agricultural droughts at sub-seasonal to seasonal (S2S) scales over India. Predictors include past information of rainfall, temperature, vapor pressure deficit, soil moisture, winds, evapotranspiration, vegetation, and atmospheric teleconnections. Predictand include soil moisture state in current timestep. All datasets are available for more than four decades. They are converted to anomalies and are processed to 25 km at monthly resolution. The student will be provided the processed dataset.

Objectives:

1. Test various regression-based state-of-the-art time series prediction models on the dataset that would be provided. Various skill metrics will be used to check the quality of results (correlation, rmse, false alarm ratio etc.). There should be an option to examine the feature importance.


2. Deliver a well-commented code (preferably in Python) of the best performing model.

Pre-requisites:

• Good understanding of deep learning models; Programming skills.

×

Lightning variability over India

Professor: Prof. S. Ravichandran

Description:

Lightning is an awesome phenomenon that accompanies severe storms. Enormous amounts of positive and negative charge build up in different regions of the cloud, generating humongous electric fields and leading to dielectric breakdown in air. Lightning is also a major hazard, accounting for about a third of all disaster-related fatalities. We will study, using radar and satellite data, the frequency and intensity of lightning over the Indian subcontinent, in order to delineate the roles of atmospheric parameters such as humidity, temperature and aerosol optical depth. Existing models do not capture the intra-annual variability of lightning, and do not convincingly account for the effects of aerosols. We will develop low-order models for lightning variability, which may be used to understand and predict lightning occurrence.

Pre-requisites:

• Experience with Python, NetCDF, etc. will be useful.


• The eagerness and willingness to learn about cloud (micro)physics and lightning.

×

Charges and collisions in clouds: the road to lightning

Professor: Prof. S. Ravichandran

Description:

Lightning is an awesome phenomenon that accompanies severe storms. Enormous amounts of positive and negative charge build up in different regions of the cloud, generating humongous electric fields and leading to dielectric breakdown in air. Lightning is also a major hazard, accounting for about a third of all disaster-related fatalities. The build-up of charge in lightning occurs through the collision of ice particles of different sizes. We will build a model of such collisions based on the study of Davila and Hunt (2001). Existing models for charging in numerical weather prediction solvers are either simple ad-hoc models, or large and unwieldy with many fitting parameters. A simple physically justified model, applied judiciously, will be able to explain the charging of clouds and, eventually, the occurrence of lightning.

Reference:

Davila and Hunt, Settling of small particles near vortices and in turbulence. Journal of Fluid Mechanics. 2001; 440:117-145. doi:10.1017/S0022112001004694

Pre-requisites:

• Experience with numerical solution of differential equations.


• Knowledge of fluid dynamics.


• The eagerness and willingness to learn about cloud (micro)physics and lightning.

×

Identification of Weather regimes to inform Renewable Energy Forecasts

Professor: Prof. Vishal Dixit

Description:

Problem Statement:

As wind and solar power become more significant in India’s electricity supply, managing their variability is essential. Over several days, this variability is influenced by persistent weather patterns, known as weather regimes. In the context of monsoon rains, weather regimes are identified with active and break phases of monsoons while similar understanding for renewable energy availability is currently lacking. During the active monsoon phases, stronger winds and more cloud cover might boost wind energy while reducing solar energy output. Conversely, during break monsoon phases, there may be less wind but more sunshine, enhancing solar energy production. Understanding these cycles can help optimize renewable energy deployment, ensuring that wind farms and solar panels are placed in areas where their output is complementary during these cycles. This can lead to more stable energy supply and better long-term planning for renewable energy forecasts in monsoon-affected regions. This project will employ traditional as well as machine learning methods to identify renewable energy regimes to ultimately push skilful renewable energy forecast horizon to 10 days.

Key references:

1. Balancing Europe’s wind-power output through spatial deployment informed by weather regimes | Nature Climate Change


2. The influence of weather regimes on European renewable energy production and demand - IOPscience

Objectives:

1. Calculation of Solar and Wind Energy availability over India for last 10 years using ERA5 data


2. Identification of weather regimes in Solar and Wind Energy availability using traditional EoF and data driven K-Means Clustering algorithms


3. Comparing the regimes with the well-known active-break cycles of monsoons

Deliverables:

1. A python code for identification of Weather regimes


2. Plots showing spatial and temporal scale information of regimes for Wind energy, Solar Energy and their co-occurrences during regimes and contrast with non-regime conditions


3. Engagement with the renewable energy company to find usefulness of this information.

Pre-requisites:

• Python, Basics of K-Means / EoF algorithm, familiarity with NetCDF files.

×

A Journey into the Hysteresis of Temperature and Precipitation Response to Anthropogenic Forcing

Professor: Prof. Angshuman Modak

Description:

This project aims to investigate the complex hysteresis of temperature and precipitation (hydrological cycle) responses to anthropogenic forcing, primarily due to changes in greenhouse gases (GHGs). Hysteresis refers to the asymmetric response of the climate system to increasing and decreasing GHG levels, resulting in differing effects on temperature and precipitation patterns. This indicates potential irreversibility that the Earth's climate system may not fully revert to its original state, either permanently or within a specific timeframe, even if atmospheric GHG levels are reduced. Understanding these responses is crucial for predicting future climate dynamics and informing effectiveness of the mitigation strategies such as the Carbon Dioxide Removal method. Insights gained from this study will also enhance our understanding of changes in regional climate patterns, which is vital for assessing impacts on water availability, food security, and resource management in a changing climate.

Objectives:

1. Examining hysteresis: To use climate models to analyze the hysteresis effects on temperature and precipitation response focussing on the climate feedbacks and energy budget of the climate system.


2. Assessment of Regional climate: To investigate how hysteresis in temperature and precipitation patterns is associated with regional climate change.

Deliverables:

1. A comprehensive report that outlines the analysis of hysteresis effects on temperature and precipitation responses, with a particular focus on the energy budget of the climate system. It will summarize the key findings from climate model simulations, illustrating the non-linear and asymmetric responses.


2. An evaluation of regional precipitation and temperature changes to GHG forcing.

Pre-requisites:

• Python while familiarity with NetCDF files is desirable.


• High motivation to engage in scientific analysis with a drive to produce novel research.


• Candidates from Climate Science, Atmospheric Science, Meteorology, Environmental Science, Engineering Physics, Chemical Engineering, Civil Engineering, and Mechanical Engineering are encouraged to apply.