Ph.D. Thesis Colloquium of
Mr. Vineeth V
[Research Supervisor: Prof. Parthasarathy Ramachandran]
Date: 5TH March 2026 [Thursday]
Time: 10:30 AM
Venue: Seminar Hall [Management Studies]
Title:” Urban Groundwater Management: A case study of Bengaluru, India.”
Urban groundwater systems in hard-rock aquifers are highly heterogeneous and strongly influenced by human activities, making conventional physically based modelling difficult, particularly under limited data availability. Bengaluru exemplifies this condition, where fragmented hydro-meteorological information restricts traditional analysis. To address this gap, the study adopts a data-driven approach using historical groundwater-level observations to identify recharge mechanisms and forecast groundwater dynamics, thereby supporting sustainable management in rapidly growing urban environments.
The first part of the thesis identifies the major components of urban groundwater recharge using Empirical Orthogonal Function (EOF) analysis of piezometric head data from 153 monitoring wells across Bengaluru. The results reveal that rainfall-induced recharge and water supply pipeline leakage are the major components of urban groundwater recharge in the study area. Spatial interpolation using kriging indicates that rainfall recharge is relatively low in the highly urbanised central region and increases towards the peripheral zones. Groundwater levels remain comparatively higher in the city centre despite lower rainfall recharge.
The second part of the thesis evaluates the feasibility of urban groundwater level forecasting based solely on historical groundwater observations, without incorporating hydro-meteorological inputs. Three neural network architectures, a Simple Feed Forward Neural Network (SFFN), a Long Short-Term Memory (LSTM) network, and a One-Dimensional Convolutional Neural Network (1D-CNN), were developed and evaluated. Convolutional neural network–based models demonstrated superior predictive performance.
The third part of the study developed a spatiotemporal framework for forecasting urban groundwater levels. Time-series piezometric head data were transformed into grid-based grayscale images representing spatial groundwater distribution. These image sequences were used to train a two-dimensional Convolutional Long Short-Term Memory (2D-ConvLSTM) network for groundwater level forecasting. The model simultaneously captures spatial and temporal groundwater dynamics, enabling city-scale groundwater forecasting.
The deep learning models significantly improved forecasting accuracy, with ConvLSTM providing robust spatiotemporal predictions. The results confirm that reliable urban groundwater forecasting is feasible even under hydro-meteorological data constraints.
Overall, the study contributes a scalable methodology for urban groundwater assessment where hydro-meteorological data are limited. By integrating statistical decomposition and deep learning-based forecasting, the framework supports groundwater management, planning, and sustainability assessment in rapidly urbanising regions.
ALL ARE WELCOME