Attention-based Models for Snow-Water Equivalent Prediction

Krishu K Thapa¹, Bhupinderjeet Singh¹, Supriya Savalkar¹, Alan Fern², Kirti Rajagopalan¹, Ananth Kalyanaraman¹

¹ Washington State University, Pullman, WA. ² Oregon State University, Corvallis, OR.

Predicting the SWE value for multiple SNOTEL locations in the Western US using the Attention Models

Model Folder

• Spatial_Attention.py - This file has the code for the spatial attention implementation along with training and testing. The data is loaded from the SDL.py file inside the DataLoader folder.

• Temporal_Attention.py - This file has the code for the temporal attention implementation along with training and testing. The data is loaded from the TDL.py file inside the DataLoader folder.

DataLoader

• Data: This has all the data we have used in our model implementation for the SNOTEL locations.
• SDL.py: This is the data loader file for the spatial model. It returns the training and testing data for the Spatial Attention model.
• TDL.py: This is the data loader file for the temporal model. It returns the training and testing data for the Temporal Attention model.
• feature_prep.py: This file processes all the raw data and generates the processed csv files of data which are used by the data loaders for spatial and attention model.