Predicting the Spatial Distribution of Snow Water Equivalent using Transfer Learning
Abstract/Contents
- Abstract
- Globally, water resources are under increasing stress that risks both water availability and quality. Hydrologic and watershed models have been used to forecast both water flow and quality. Furthermore, these models serve as decision making tools for sustainable water management. These models are subject to high uncertainties that can be reduced by accurate estimates of high-resolution spatial distribution of Snow Water Equivalent (SWE). The Airborne Snow Observatory generates Lidar maps that quantify SWE at a 50 m resolution. Although these maps provide detailed information, their infrequent temporal availability limits their use for modeling. For instance, in the Rocky Mountains of Colorado, only 12 maps are present in total for the period covering 2016 to 2019. In this work, we develop a neural network (NN) framework to model this temporally sparse Lidar data, which enables the estimation of SWE at unmapped time points in the Rocky Mountains of Colorado. To achieve this, we first incorporate meteorological and topographic data that describe the physical processes governing SWE. This data is used to extract seven predictor variables: accumulated snow, accumulated precipitation, mean seasonal temperature, the sum of positive degree days, elevation, slope, and aspect. Next, we tackle data scarcity by using transfer learning to integrate information from 80 Lidar maps in the Sierra Nevada mountains in California. An explanatory factor analysis is conducted to justify and reinforce the use of transfer learning. Results show that our NN framework, using the seven predictor variables, models SWE with a mean R2 of 0.44 prior to transfer learning. After applying transfer learning, we improve the mean R2 results to 0.56. Overall, transfer learning increases model predictability while decreasing both scatter and bias.
Description
Type of resource | text |
---|---|
Publication date | December 15, 2022 |
Creators/Contributors
Author | El Halabi, Lama |
---|---|
Thesis advisor | Tartakovsky, Daniel |
Researcher | Mital, Utkarsh |
Advisor | Dwivedi, Dipankar |
Degree granting institution | Stanford University |
Subjects
Subject | Snow water equivalent |
---|---|
Subject | Transfer learning (Machine learning) |
Subject | Deep learning (Machine learning) |
Subject | Factor analysis |
Genre | Text |
Genre | Capstone |
Genre | Report |
Genre | Thesis |
Genre | Student project report |
Bibliographic information
Access conditions
- Use and reproduction
- User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 4.0 International license (CC BY-NC).
Preferred citation
- Preferred citation
- El Halabi, L. (2022). Predicting the Spatial Distribution of Snow Water Equivalent using Transfer Learning. Stanford Digital Repository. Available at https://purl.stanford.edu/xv296vk9757. https://doi.org/10.25740/xv296vk9757.
Collection
Master's Theses, Doerr School of Sustainability
View other items in this collection in SearchWorksContact information
- Contact
- lie08@stanford.edu
Also listed in
Loading usage metrics...