The AI/ML/RS track will provide hands-on workshop sessions demonstrating machine learning methods using current CIROH modeling projects aimed at advancing AI/ML applications in operational hydrology. This track will include workshops on advanced ML methodologies such as hyperparameter and architecture selection, custom architecture development, training strategies, loss functions, end-to-end development, and AI/ML methods for remote sensing in hydrological modeling. Participants will leave with an improved understanding of data processing, machine learning models and applications, training and evaluation procedures, result visualization, and practical workflows they can apply to their own hydrological modeling objectives.
Leads: Jonathan M. Frame, University of Alabama
Workshop Listings
Foundations of ML: From Architecture to Optimization
Day 1 Session 1
Savalan Neisari (University of Alabama)
Leo Lonzarich (Penn State University)
This introductory workshop provides a practical foundation for building machine-learning workflows for hydrologic applications. Participants will explore major machine learning model classes and develop end-to-end modeling pipelines, including data preprocessing, model architecture design, and hyperparameter tuning.
Through guided, hands-on exercises, attendees will examine key concepts such as overfitting, validation strategies, and model generalization. The session emphasizes reproducible workflow design and practical decision-making for selecting models suited to forecasting, classification, and environmental prediction tasks.
Sensitivity Augmented Geo Exploration (SAGE) for Rapid Model Calibration
Day 1 Session 2
Jasper Vrugt (University of California, Irvine)
Jonathan Frame (University of Alabama)
This hands-on workshop introduces the Sensitivity Augmented Geo Exploration (SAGE) framework for accelerated calibration and uncertainty analysis of hydrologic models. Participants will run calibration experiments using large CAMELS basin datasets, explore parameter sensitivities and uncertainty structures, and learn how high-performance SAGE workflows enable model training and evaluation in minutes rather than days.
The workshop demonstrates how accelerated calibration and sensitivity-analysis methods support rapid experimentation, operational model tuning, and large-scale hydrologic forecasting research.
Hands-on LSTM and Transformer for Operational Streamflow Prediction
Day 2 Session 1
Jiangtao Liu (Penn State University)
This hands-on workshop presents a complete operational workflow for training, evaluating, and comparing deep-learning models for streamflow prediction. Participants will implement a standardized modeling pipeline using CAMELS data, train an LSTM baseline, and perform a controlled comparison with a lightweight Transformer architecture.
The session emphasizes practical considerations for operational forecasting, including model stability and performance optimization using techniques such as learning-rate scheduling, regularization, and gradient clipping.
Satellite Data Projection and Georeferencing with Satpy
Day 2 Session 2
Marouane Temimi (Stevens Institute of Technology)
Baya Cherif (Stevens Institute of Technology)
Jorge Bravo (Stevens Institute of Technology)
This workshop introduces practical methods for transforming raw satellite swath observations into georeferenced, analysis-ready raster products using the Satpy Python library. Participants will learn the fundamentals of satellite sensor geometry, coordinate systems, reprojection, and resampling through hands-on examples.
The session emphasizes building repeatable processing workflows that enable integration of satellite datasets into precipitation monitoring, flood mapping, and operational forecasting pipelines.
Incorporating Process-Based Models into NeuralHydrology
Day 3 Session 1
Daniel McKenzie (Colorado School of Mines)
Ziyu Li (Colorado School of Mines)
This workshop introduces differentiable modeling approaches that integrate process-based hydrologic models within the neuralhydrology deep-learning framework. Participants will explore implementation strategies using the CFE conceptual model and learn best practices for designing hybrid modeling workflows that combine physical interpretability with machine-learning predictive performance.
The session highlights how process-informed ML approaches can improve robustness, transparency, and operational forecasting capabilities.
GeoAI-Driven Flood Risk Modeling: Integrating Hazard, Exposure, and Vulnerability
Day 3 Session 2
Hemal Dey (University of Alabama)
In this hands-on workshop, participants will develop a GeoAI workflow that integrates hazard, exposure, and vulnerability data to produce predictive flood-risk maps. Attendees will work with historical flood-damage data, geospatial predictors, and machine-learning models such as Random Forests.
The session emphasizes explainable AI methods and reproducible modeling pipelines that support operational decision-making, risk prioritization, and emergency planning workflows.