Advancing CONUS-scale Operational Snow Modeling Capabilities.

Objective:

The large-scale Snow modeling project is multifaceted with 4 key objectives, one from each institution. UVM: Advancing the modeling of snow-affected regions of the Northeastern US. UU: Advancing snow energy and mass balance physics in operational snow models will modernize hydrological modeling from a state of practice relying on simplified representations of mountain snow cover processes. UU: Integration technique goals will advance methods for updating snow state variables by incorporating novel ML methods that account for the spatial and temporal variability of snow state characteristics across accumulation, transformation, and melt phases in complex topography. Mines: Determining operational and forecasting benefits will evaluate the skill of different snow models within the NWM hydrofabric framework.

Approach:

The research objectives will be coordinated with one another and align modeling locations to montane catchments where there are high quantity observations, including essential water supply catchments in the Sierra Nevada, Rockies, and Green Mountains. Basin selection will also intersect with the CIROH Testbed (PI Wood) effort. Snow modeling and assessment will be carried out using meteorological forcings developed for Nextgen (i.e., the Analysis of Record, AORC), in situ observations, large-scale observation datasets, downscaled satellite remote sensing products, and LiDAR observations from the Airborne Snow Observatory (ASO), inc.

Task 1:Advancing Snow Modeling in Under-represented Snow-affected Regions will leverage existing snow models and apply them in historically underrepresented snow-affected regions of the Northeastern US. Research activities include data mining, model deployment, simulations, and evaluation to determine modeling needs, gaps, and overall performance for coupling with the NWM.
Task 2: Advancing Snow Energy and Water Balance Modeling Physics will focus on snow process representation driving peak SWE and the timing and magnitude of snowmelt contribution to catchment hydrology in areas with varying controls on snow accumulation, transformation, melt, and catchment residency. Task 3: ML and Data Assimilation Techniques will implement and refine ML-based snow modeling methods and assess data assimilation techniques for updating snow state variables in process models. Research activities focus on optimizing modeling domains and evaluating the impact of additional model inputs on SWE estimation accuracy. Task 4: Determining Operational Monitoring and Forecasting Benefits focuses on creating a snow-centric testbed to quantitatively assess the increase in snow modeling performance.

Impact:

Abstract:

The need for accurate characterization of accumulated snow water equivalent (SWE) and subsequent melt timing and rate are essential for water management efforts, specifically water supply forecasting. With the Next Generation water resources modeling framework (Nextgen), there is an opportunity to integrate distributed, process-based modeling and novel machine learning (ML) into large-scale hydrological operations, thereby advancing the representation of snow conditions in heterogeneous montane environments to enhance streamflow forecast skill. The impact of refine estimates of snow distribution is substantial, as it will enhance estimates of when, where, and how much surface water will be available for economic needs (e.g., hydropower, recreation, agriculture, etc.). To enhance snow distribution prediction, the project focuses on snow environments across CONUS (Sierra Nevada, Rockies, and Green Mountains), building on and expanding existing CIROH projects. Key objectives include creating a standardized workflow for evaluating the benefits of snow and hydrological modeling predictions, incorporating more physically explicit representations of snow processes, and advancing data assimilation and ML methods.

Project activities have matured ML methods for estimating real-time catchment-scale SWE based on the intrinsic relationships between in-situ observations and landscape characteristics. The ML team finished the second generation of SWEML (i.e., SWEMLv2.0), developed with three key objectives: 1) one ML model to leverage all available spatial snow observations, 2) enhanced prediction in locations with zero to minimal observations, and 3) support geospatial snow distribution relationships (e.g., train CNNs). Next steps include using different ML algorithms (LSTM, CNN), integration and fusion of precipitation products, including teleconnections, and rigorous model evaluation with the testbed group. The final product will be a semi-operational ML model producing accurate daily SWE maps for key ASO watersheds.

The testbed team developed a suite of datasets for snow water equivalent (SWE) based upon on the NextGen HydroFabric (v2.2), which serves as the core of a CIROH Hydrologic Prediction Testbed protocol for benchmarking catchment areal SWE. Associated with these datasets are proposed standards (metrics and test basins/periods) for community evaluation and tracking of progress as the final product.

The Northeast snow team conducted an initial test of iSnobal for a domain in northwestern Vermont, which captures a newly established snow monitoring network for model validation. Plans include extending the model evaluation to capture additional years through field observations and evaluating snowmelt simulations for an extreme rain-on-snow event in 2023. The final product will be a validated deployment of iSnobal for test watersheds in the snow monitoring network.

The physically-based snow modeling team is running the iSnobal model operationally (i.e. up to real time) over 9 watersheds spanning a range of snow environments in the Western US. Based on comparisons to observations (SNOTEL and ASO) the model framework is undergoing active development to; 1) modernize representation of shortwave and longwave radiation fluxes in mountain terrain (through updates to downscaling procedures for model forcings and model physics), 2) incorporate canopy interception and unloading, and 3) assess precipitation bias correction methods. The final product will be the expansion of iSnobal running operationally on additional watersheds (e.g., > 9)

The research has advanced the quality of CONUS scale modeling of montane snow distribution, created a standardized testbed supporting a research-to-operations (R2O) model transition in snow-dominated catchments, and created a HydroLearn module introducing snow monitoring infrastructure, tools, and modeling.