Publications

To address the limitations of cloud cover and misclassification in traditional flood mapping, this study introduces a novel hybrid framework that combines metaheuristic optimization algorithms with deep learning semantic segmentation models. By applying this methodology to Sentinel-1 SAR imagery of the 2019 Iran flood, the integration of a Swarm-based Simulated Annealing feature selection technique with a CPNet architecture achieved highly precise, high-resolution flood extent mapping.

‍

This study introduces a novel Process-Guided Graph Transformer framework that combines hydrological modeling with Graph Neural Networks and Transformers to improve streamflow prediction in ungauged and data-sparse river basins.

The proposed framework demonstrated improved prediction accuracy, extreme-event detection, and computational efficiency across river basins in India, Australia, and the United States.

SM2RAIN rainfall estimation is limited by infiltration physics and soil moisture depth. This study introduces a Green-Ampt based approach and optimizes parameters using Bayesian methods, improving rainfall retrieval accuracy, though performance decreases with depth. Results highlight land cover dependent behavior and provide more reliable rainfall estimates with quantified uncertainty.

‍

This study presents a refined optimal estimation approach to simultaneously retrieve soil moisture and soil organic matter from microwave dielectric measurements. Unlike conventional models that focus only on soil moisture, the proposed method explicitly incorporates organic matter into a continuous dielectric mixing framework, reducing ambiguity in dual-parameter estimation. Validation using SMAPVEX12 field data demonstrates strong agreement for both soil moisture and organic matter estimates. The approach improves retrieval accuracy and enables the use of microwave sensors for assessing soil carbon content, with direct applications in irrigation management, soil health monitoring, and carbon accounting.

This study introduces a self-calibration framework for bottom-up rainfall estimation that eliminates the need for rain gauge data by using clustering-based parameter identification. Applied to SM2RAIN-NWF, the approach demonstrates robust performance across validation strategies, with K-means and Growing Neural Gas showing strong reliability and accuracy, making it well-suited for large-scale and data-sparse regions.

‍

This paper demonstrates that simultaneously assimilating radar and radiometer soil moisture retrievals produces measurable improvements in numerical weather prediction, beyond what is achieved using a single sensor.

By jointly using ASCAT radar soil moisture and SMAP radiometer soil moisture within the Korean Integrated Model coupled with the NASA Land Information System, the study shows how complementary microwave observations strengthen land atmosphere coupling in forecasting systems.

‍