The ML track will provide an overview of the theory behind various ML methods currently used within select CIROH projects and apply the methods to different modeling objectives. The workshop lineup will begin with an overview of the ML methods and hands on sessions for developing a ML pipeline. The hands-on workshop sessions will exemplify ML methods using current CIROH modeling projects seeking to advance the application of ML in operational hydrology, such as streamflow modeling, bias correction, snow modeling, and forcing data preparation. CIROH will provide the CIROH Cloud workspace to ensure environmental stability, workshop leads will provide the material on GitHub, and tentatively cover LSTM, Physics Informed ML, CNN, XGBoost, and MLP modeling algorithms. Track attendees can expect to leave with greater knowledge of data processing, ML models and their respective applications, training and evaluation procedures, result visualization, and a stronger foundation to apply the workflows to their unique hydrological modeling objectives.
Hydrological Applications of Machine Learning Workshops
Working with National Water Model data in Amazon SageMaker and demo of Amazon BedRock (AI tool)
(Cross-listed with the Cross-Cutting Track)
Day 1 Session 2
Scott Hendrickson
Savalan Naser Neisary
Eric Christensen
Arpita Patel
This workshop will explore foundational aspects of Amazon SageMaker as well as some advanced features. Participants will be provided AWS accounts where they will be hands-on deploying, training and running inference on models using NWM data.
The workshop will also include a demonstration of Bedrock AI.
CNNs for Predicting daily orographic precipitation gradients for atmospheric downscaling
Day 2 Session 1
Savanna Wolvin
This workshop offers participants an introduction to convolutional neural networks (CNNs) for their application in hydro-meteorology. Core concepts to be covered include the layers within a CNN, the learning process of the CNN, and techniques related to Explainable AI (XAI). The hands-on portion of the workshop focuses on customizing a CNN through hyperparameter exploration, adjusting CNN layers, and manipulating inputs. Participants will gain an understanding of CNN architecture, practical skills in customizing a CNN, and the apply the models in Northern Utah for downscaling ERA5 data for the quantity of liquid precipitation. Participants will use GitHub to fork the repository and clone to their machine using the CIROH Cloud Computing environment.
Day 2 Session 2
Kel Markert
This workshop will introduce participants to Google’s Flood Forecasting system and demonstrate how to train and predict streamflow using the open-source methods with Google Cloud services like Earth Engine and Vertex AI. Through hands-on exercises, attendees will gain practical experience in building and applying AI models for streamflow prediction.
Participants will explore the model using the Snow Water Equivalent Machine Learning (SWEML) model in a data-driven machine learning (ML) platform with a modular structure to account for the heterogeneity of climate and topographical influences on SWE across the western United States. The SWEML pipeline assimilates nearly 700 snow telemetry (SNOTEL) and California Data Exchange Center (CDEC) sites and combines with processed lidar-derived terrain features for the prediction of a 1 km x 1 km SWE inference in critical snowsheds. While the model consists of twenty-three regionally specific sub-models tailored to the unique topography and hydroclimate phenomena in the Western U.S., participants will select a single domain of interest (e.g. Sierra Nevada Mountains) to explore the ML pipeline and participate in data processing, model training, model evaluation, and figure development.
DeepBucketLab is an interactive, hands-on tool designed to introduce students in civil engineering and Earth science disciplines to the fundamentals of neural network-based modeling, with a specific focus on hydrological processes. As neural networks become increasingly vital in predicting current and future hydrological conditions, there’s a growing need for formal education in this area at the upper division and graduate levels. DeepBucketLab aims to fill this gap by providing a practical and educational platform for training effective neural network models for hydrological prediction.
Physics-Informed Neural Network (PINN) and its applications in the Hydro fields
Day 4 Session 1
Abdol Mehdi Behroozi
In this workshop, we will begin by introducing the fundamental concepts of PINN, highlighting how this innovative framework integrates differential equations from physical laws with deep learning. Following the conceptual overview, we will dive into hands-on coding sessions. These sessions are designed to demonstrate how PINN can be applied to solve real-world problems in the hydro field, such as dealing with the shallow water equation, modeling sediment transport, or wave propagation.
Machine Learning: Decision Tree Models for Post-Processing National Water Model Streamflow Outputs
Day 4 Session 2
Savalan Neisary
The Decision-Tree workshop will explore simple Decision Trees, Random Forest, and XGBoost in streamflow modeling. The workshop will include a brief introduction to decision tree algorithms and transition to hands-on activities in which participants will engage in the ML development pipeline, including data processing, algorithm training, and model evaluation. The workshop will use the CIROH cloud and all of the Python code and data will be available through GitHub and public Amazon S3 buckets. Participants can expect an improved understanding of different decision tree algorithms and their applications within hydrological modeling, knowledge of data preprocessing, data visualization, and general skills in using Git and cloud computing workspace.