Data-driven modeling and in situ monitoring of river ice processes to support NWM NextGen capabilities in cold regions

Objective:

The goal is to develop machine learning techniques to model river ice processes, advance cold regions hydrology research, and overcome the complexity that limits the use of process-based models operationally for ice transport simulation.

Approach:

Here, we propose to address these operational gaps to simulate river ice processes, with a focus on ice jam formation, using data-driven methods to enhance NWM water prediction capabilities in cold regions. The NextGen framework will be used to integrate the developed techniques in the NWM. The main task consists of using a data-driven method that leverages multiyear river ice remote sensing observations generated from an ongoing FY22 CIROH-funded project to predict river ice dynamics. In addition, other meteorologic, hydrologic, hydraulic, and environmental variables will be used to train and test deep-learning, CNN-LSTM, models to predict ice dynamics. Observations from USGS will be used in the data-driven method. The potential of expanding river ice monitoring capabilities to better train and test the data-driven methods will be addressed through the testing of river ice in situ monitoring prototypes.

Impact:

Abstract:

The current version of the National Water Model (NWM) does not simulate river ice processes. The complexity of the interaction between river ice and the hydrodynamics in rivers made the coupling of both processes operationally in large-scale hydrologic models extremely challenging. In addition, the lack of continuous in situ observations of ice in rivers does not support the development and calibration/validation of such models. Here, we propose to address these operational gaps to simulate river ice processes, with a focus on ice jam formation, using data-driven methods to enhance NWM water prediction capabilities in cold regions. The NextGen framework will be used to integrate the proposed techniques into the NWM. The main task consists of using a data-driven method that leverages multiyear river ice remote sensing observations generated from an ongoing FY22 CIROH-funded project to predict river ice dynamics. In addition, other meteorologic, hydrologic, hydraulic, and environmental variables will be used to train and test deep-learning, CNN-LSTM, models to predict ice dynamics. Observations from USGS will be used in the data-driven method. The potential of expanding river ice monitoring capabilities to better train and test the data-driven methods will be addressed through the testing of river ice in situ monitoring prototypes. This project expands and improves water prediction and monitoring capabilities in cold regions through novel modeling and observation techniques and therefore concurrently serves the interests of NOAA, USGS, and USACE, three CIROH sponsors.