Nath et al. (2026) Hybrid AI modelling for imputation and modelling of remotely sensed surface water in climate-sensitive wetland

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2026
• Date: 2026-01-01
• Authors: Roshan Nath, Somil Swarnkar, Vikas Poonia, Vinod K Kurmi
• DOI: 10.1016/j.rsase.2026.101955

Research Groups

• Department of Earth and Environmental Sciences, IISER Bhopal, Bhopal, Madhya Pradesh, India
• Department of Civil Engineering, MANIT Bhopal, Bhopal, Madhya Pradesh, India
• Department of Data Sciences and Engineering, IISER Bhopal, Bhopal, Madhya Pradesh, India

Short Summary

This study developed a hybrid AI framework to reconstruct and predict monthly water surface area (WSA) for Bhojtal Lake (1990–2022), revealing increased drought susceptibility and low-WSA months under a +2 °C warming scenario.

Objective

• To develop a transferable, data-efficient hybrid AI framework for long-term monitoring and climate-impact assessment of urban wetland surface water area, specifically addressing data gaps from remote sensing and predicting future dynamics under warming scenarios.

Study Configuration

• Spatial Scale: Bhojtal Lake, a Ramsar-designated urban wetland in Central India.
• Temporal Scale: Historical reconstruction and modeling from 1990 to 2022 (33 years) at a monthly resolution, with future projections under a stabilized +2 °C warming level.

Methodology and Data

• Models used:
  • Mode-based kernel correction with an objective kernel-selection protocol (for spatial data gaps).
  • Bayesian-optimized Feedforward Neural Network (FNN) (for temporal data gaps).
  • Recursive Long Short-Term Memory (LSTM) ensemble (for future WSA simulation).
• Data sources:
  • Landsat-derived JRC Global Surface Water data (for Water Surface Area).
  • Lagged hydroclimatic variables: precipitation, temperature, evapotranspiration, and soil moisture (as drivers for FNN and LSTM).
  • Climate inputs representing a stabilized +2 °C warming level (for future LSTM forcing).

Main Results

• The developed hybrid AI framework demonstrated strong skill in Water Surface Area (WSA) reconstruction and prediction, achieving an R² of approximately 0.9 and a Root Mean Square Error (RMSE) of approximately 2–3 square kilometres.
• Ensemble simulations under a +2 °C warming scenario revealed pronounced seasonal asymmetry in WSA dynamics and an increased frequency of months with low WSA, indicating heightened drought susceptibility for the wetland.

Contributions

• Presents a novel, transferable, and data-efficient hybrid AI framework that integrates remote sensing, climate forcing, and artificial intelligence for long-term monitoring and climate-impact assessment of urban wetlands.
• Addresses critical challenges in surface water monitoring, such as cloud contamination, sensor discontinuities, and limited in situ observations, through robust spatio-temporal imputation techniques.
• Provides a valuable tool for climate-resilient water management and Sustainable Development Goal (SDG)-aligned planning, particularly in data-limited regions.

Funding

Not specified in the provided text.

Citation

@article{Nath2026Hybrid,
  author = {Nath, Roshan and Swarnkar, Somil and Poonia, Vikas and Kurmi, Vinod K},
  title = {Hybrid AI modelling for imputation and modelling of remotely sensed surface water in climate-sensitive wetland},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2026},
  doi = {10.1016/j.rsase.2026.101955},
  url = {https://doi.org/10.1016/j.rsase.2026.101955}
}