Presenters-0665
From CSDMS
CSDMS 2025: Exploring Earth's Surface with Models, Data & AI
Integrating CNN with LSTM Models for Capturing Spatiotemporal Dynamics of Flood Modeling
Abstract
This clinic will introduce the concept and implementation of a hybrid Deep Learning (DL) framework, that integrates: Convolutional Neural Network (CNN) and Long-short Term Model (LSTM), to simulate two-dimensional flood scenarios. This advanced DL framework enables spatiotemporal predictions of hydrodynamic parameters, with a particular focus on predicting water depths of flood events. Participants will first acquire a brief introduction to both models and their integration concept, followed by hands-on experience in developing a simple hybrid model using the PyTorch library. The training process will utilize a small-scale 2D flume as a practical and time-efficient exercise; however, this technique can be scaled up and applied to large stream segments in real-world scenarios. The workshop will highlight the capabilities, applications, and best practices of the developed model within the water resources domain. Notably, it will showcase the DL models’ ability to generate predictions significantly faster than traditional hydrodynamic models like HEC-RAS, which face substantial computational challenges in simulating 2D flood scenarios, especially for large-scale or long-term simulations. This workshop will equip the participants with the necessary background and technical skills for various spatiotemporal applications, encompassing geomorphological processes, groundwater dynamics, and wave-driven simulations. For this workshop, a basic understanding of machine learning in Python is recommended to maximize the benefits of this session.
Please acknowledge the original contributors when you are using this material. If there are any copyright issues, please let us know (CSDMSweb@colorado.edu) and we will respond as soon as possible.
Of interest for:
