This workbook builds on the ongoing Cooperative Institute for Research-to-Operations in Hydrology (CIROH) machine learning snow modeling project to provide operational, near-real-time snow-water-equivalent (SWE) estimates for updating the state of the snowpack within the National Water Model (NWM).
The Getting started file will help new users create a virtual environment and install the correct packages using the environment.yml.
Snow-derived water is a critical hydrological component for characterizing the quantity of water available for domestic, recreation, agriculture, and power generation in the western United States. Advancing the efficiency and optimization of these aspects of water resources management requires an enhanced characterization of the snow state variable, particularly the essential global inputs of snow-water-equivalent (SWE), peak SWE, and snowmelt onset for hydrological models. While physically-based models that characterize the feedbacks and interactions between influencing factors predict SWE values well in homogeneous settings, these models exhibit limitations for CONUS-scale deployment due to challenges attributed to spatial resolution, landscape heterogeneity, and computational intensity. Leveraging a collaborative partnership between the Alabama Water Institute (AWI) at the University of Alabama (UA) and the University of Utah (UU), we address these limitations through 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 model consists of twenty-three regionally specific sub-models tailored to the unique topography and hydroclimate phenomena in the Western U.S., exhibiting an RMSE less than 8 cm and a coefficient of determination approaching 0.99 on predictions spanning the 2013-2017 training period. The model 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.
The purpose of this workbook is to provide users with an introduction to snow modeling through leveraging machine learning and observation data. After running the example notebook, users are encouraged to explore and create their own modifications to the code to refine SWE prediction performance in the Sierra Nevada mountains. The current Sierra Nevada core ML algorithm is a deep neural network, and while it demonstrates satisfactory prediction skill, there still are large errors in predction and room for improvement identified by a rigorous evaluation of the model. The most notable error appear at locations below 1,500m and above 2,900, ephemeral and alpine regions, respectively, and during the snow melt period. We encourage users to explore other ML algorithms, such as other neural networks, Long-Short-Term-Memory, tree-based, and even simple regression algorithms such as Ordinary Least Squares. Model evaluation will form a critical element of determing differences in algorithm performance, and the Standardized Snow Water Equivalent Evaluation Tool (SSWEET) will serve as a standardized method to evaluate the different algorithm architectures.
All data will be provided to participants with cloud access to either the CIROH publicly accessible Amazon Web Services (AWS) S3 storage or via Box. Project data includes:
- NASA Airborne Snow Observator (ASO) LiDAR-derived SWE estimates
- Copernicus 90m Digitial Elevation Model (DEM)
- Natural Resources Conservation Service (NRCS) Snow Telemetry (SNOTEL) monitoring station SWE observations
- Visible Infrared Imaging Radiometer Suite (VIIRS) fraction snow covered area (fSCA)
The data folder contains pre-processed model training dataframes of approximately 7,000 1-km grid locations to limit the amount of time spent on data processing tasks. While a key and essential component of any ML objective, data pre-processing and feature engineering are not a focus of this workbook.