Advancing River Ice Climatology through Multivariate Satellite and In-situ Observations: A Nexus for Enhanced Streamflow Prediction in the Northeastern United States

In the watersheds of the Northeastern United States, streamflow simulation by NOAA’s National Water Model (NMW) routinely overlooks the influence of river ice, leading to potential forecast inaccuracies, especially during critical periods such as ice break-up and freeze-up. Despite the recognized challenge, there is a gap in comprehensive research that effectively combines hydrological, climatological, and geomorphological data. This study aims to address this gap by intricately weaving these data strands to understand ice phenology nuances in regional rivers.

This study leveraged remote sensing data from the Suomi NPP Visible Infrared Imaging Radiometer Suite (VIIRS) spanning water years from 2012 to 2023. A U-Net deep learning algorithm was employed for image segmentation to extract dates of ice formation and break-up, along with quantifying ice concentrations. To provide a deeper historical context, in-situ measurements from USGS streamflow gauging stations were utilized, encompassing water years from 1985 to 2023. The measurement flag present in the streamflow series from USGS gauging stations was employed to identify the periods of ice formation and breakup. Furthermore, when water temperature data were available, this information served as a validation metric, reinforcing the reliability of our method that relies on the measurement flag to identify ice cover periods. The study focused on ice-prone rivers selected using the Cold Regions Research and Engineering Laboratory (CRREL) database. In complementing the above, the ERA5 dataset from the ECMWF Climate Reanalysis was employed to discern hydroclimatological drivers possibly influencing ice dynamics, encompassing factors like soil freeze-thaw cycles, air temperature, and precipitation – all aggregated at the HUC-8 drainage basin level. The in-situ data was analyzed using the Modified Mann-Kendall (MMK) test to highlight trends, while a Bayesian multiple change point detection approach was employed to detect possible trend shifts. Further, correlations with global climatic oscillations were explored, providing insights into their influence on river ice dynamics and peak flows during break-up.

Through a combination of VIIRS segmented imagery and the NHDplus dataset – the latter functioning as the hydrofabric for NWM streamflow simulations – critical insights into spatial variation of ice dynamics were obtained. These findings, in turn, hold potential in refining the calibration of pivotal model parameters, including channel roughness coefficients. Our climatological analysis illuminated the intricate interplay between river ice dynamics and their climatic triggers, thus offering pathways for NWM enhancements during the break-up and freeze-up seasons. Notably, the study underscored the overarching influence of teleconnection patterns on river ice dynamics in the northern watersheds, elucidating the significant variability of break-up, freeze-up, and peak flows therein. The obtained results revealed a strong coherence between El Niño-Southern Oscillation (ENSO) indices and ice formation and breakup dates, with higher streamflow observed during ice breakups in the El Nino phase. Furthermore, while the MMK test did not manifest any notable temporal trends in the ice duration and break-up peak flow series, the Bayesian multiple change point detection elucidated shifts in the linear regression. These shifts were notably correlated with extreme ENSO phases.

In the evolving landscape of hydrological modeling, our study underscores the power of integrating diverse datasets to enhance predictive accuracy. With river ice playing a pivotal role in the hydrological regime of the northeastern watersheds, its comprehensive understanding becomes non-negotiable. Our methodological approach and findings set a benchmark for future endeavors in transitioning next-generation water resources models to operations.