Ph.D. Candidate | Hydrology & Atmospheric Sciences

University of Arizona | Integrating Physics, Machine Learning & Data Science

I am a Ph.D. candidate in Hydrology and Atmospheric Sciences at the University of Arizona, with research interests spanning physical hydrology, geospatial data science, and machine learning.

My experience includes large-scale physical and hydrologic modeling with Noah-MP and RAPID, along with geospatial analysis and scientific computing using xarray, GDAL, GIS, and Python. I work extensively with large environmental datasets, hydrologic workflows, and reproducible model evaluation pipelines.

I also develop deep learning, hybrid physics–ML, and distributed GPU-based workflows for hydrologic prediction. My current focus is on building data pipelines and modeling frameworks in PyTorch, including translating Fortran-based models such as Noah-MP into GPU-enabled, deep-learning-ready implementations that can be coupled with neural networks.

I am particularly interested in building scalable and interpretable systems that unify physical models, geospatial data, and modern machine learning for water-resources applications.

• Noah-MP & RAPID models for infiltration, runoff, baseflow, and soil moisture dynamics
• Process-based model development, calibration, and uncertainty quantification
• High-performance computing (HPC) implementations for large-scale simulations
• Snow-hydrology interactions and subsurface water dynamics

• PyTorch implementations for GPU-accelerated parameter optimization
• Hybrid physics-guided + machine learning systems
• Differentiable modeling enabling automatic differentiation through hydrology
• Neural networks with physical constraints and conservation laws

• Deep learning approaches: LSTMs, Transformers, attention mechanisms
• Physics-informed neural networks (PINNs) for process representation
• Transfer learning for ungauged basins
• Multi-step ahead forecasting with uncertainty quantification

• Python, xarray, geopandas for scalable geospatial analysis
• SQL databases and cloud computing for big hydrology
• HPC workflows and containerization (Docker)
• Open science practices and reproducible pipelines

• Linking precipitation intensity to groundwater recharge and terrestrial water storage (TWS)
• Extremes analysis (droughts, floods) in water-limited environments
• Urban hydrology and stormwater management (SWMM, HEC-RAS)
• Climate adaptation and water security in drylands

Status: In Development (Private) | Release: Q3 2026

A fully differentiable, GPU-accelerated implementation of Noah-MP coupled with Graph Neural Network routing for scalable hydrologic modeling.

Key Features:

• ⚡ 50x+ speedup over traditional Noah-MP via GPU vectorization
• 🔄 Fully differentiable – automatic differentiation through entire model
• 🧠 Graph Neural Networks for learnable streamflow routing
• 🔗 Hybrid physics-ML – seamless integration with machine learning
• 📈 Scalable across 1000s of catchments simultaneously

Technical Stack: PyTorch, CUDA, XLA, Physics-Guided ML

Early Access: Available for research collaborations

Status: Active & Available

A modular deep learning framework for streamflow prediction using sequence models, developed for hydrologic forecasting experiments and comparative model analysis.

Key Capabilities:

• 🧠 Models: LSTM and Transformer architectures for basin-scale streamflow prediction
• 📦 Modular framework: organized pipeline for data access, preprocessing, training, evaluation, visualization, and experiment management
• ⚙️ Experiment workflows: configurable YAML-based runs, checkpointing, early stopping, sweep analysis, and model comparison
• 📊 Evaluation tools: basin-wise metrics, hydrographs, parity plots, flow-regime diagnostics, seasonal skill, and report generation
• 🌍 Dataset support: MiniCAMELS-based workflows with split-aware analysis and reproducible preprocessing

Tech Stack: PyTorch, YAML Configs, Sequence Modeling, Hydrologic Evaluation, Scientific Python**

Status: Active · Private repository

Research code accompanying the study “Process-Aware AI for Rainfall–Runoff Modeling: A Mass-Conserving Neural Framework with Hydrological Process Constraints”.

Key Capabilities:

• 🌧️ Physics-aware rainfall–runoff modeling: mass-conserving neural framework with explicit hydrological process constraints
• 🧠 Hybrid AI design: integrates machine learning flexibility with physically interpretable storage–flux relationships
• 🔬 Research focus: process representation, interpretability, and improved hydrologic prediction across catchments
• 📄 Associated preprint: arXiv:2603.25093

Tech Stack: PyTorch, Physics-Guided ML, Hydrologic Process Modeling, Scientific Machine Learning

Repository is currently private.

Tech Stack: Python, Rasterio, Xarray, NetCDF, YAML, Google Earth Engine, NASA Earthdata API**

Full portfolio, publications, talks, and CV: