Enter Note Done
Go to previous page in this tab
Session
  • Presentation | H13M: Advancing Measurements and Modeling of Invisible Components of the Water Budget: Challenges, Opportunities, and Applications II Poster
  • Poster
  • Share Icon
    Bookmark Icon
  • [ONLINE] H13M-VR8829: Reconstructing Monthly Groundwater Depth Dataset in India (2002–2021) using Cluster-Based Machine Learning and Temporal Disaggregation Approach
    • Schedule
    Notes
  • Online
    Online
    Set Timezone

Generic 'disconnected' Message
Author(s):
Tapis Samaiya, Indian Institute of Technology BHU (Varanasi) (First Author, Presenting Author)
Himanshu Singh, Indian Institute of Technology Gandhinagar
M Naik, Indian Institute of Science Education and Research Bhopal
Rajesh Singh, Indian Institute of Technology Gandhinagar
Hiren Solanki, Indian Institute of Technology Gandhinagar
Vimal Mishra, Indian Institute of Technology Gandhinagar


Groundwater is a vital freshwater source in India, supporting nearly 60% of irrigation needs and serving as the primary water supply for millions. However, long-term assessment of groundwater trends is hindered by data inconsistencies and substantial missing values. This study presents a cluster-based machine learning and temporal disaggregation approach to construct a consistent, high-resolution, and monthly groundwater dataset from 2002 to 2021 across India. We used quarterly groundwater depth observations from over 4,300 wells across India, which contain less than 30% missing values. To address spatial variability, we grouped wells into clusters using a k-means clustering algorithm based on hydro-climatic and geological characteristics. An Extra Trees regression model was trained separately for each cluster using relevant hydro-climatic and geological parameters to fill missing values and reconstruct continuous quarterly groundwater depth data. The gap-filled quarterly dataset was then evaluated using standard performance metric, which shows good agreement. Further, we applied the Chow-Lin temporal disaggregation method to convert the quarterly data into monthly time series while preserving seasonal patterns. The final dataset improves temporal resolution and effectively captures both seasonal and long-term groundwater variations, providing a valuable resource for groundwater monitoring, sustainable water management, and policy-making, especially in water-stressed regions in India.



Scientific Discipline
Neighborhood
Type
Where to Watch
Main Session
Discussion