Research

Machine Learning (ML) & eXplainable Artificial Intelligence (XAI) for Arctic Sea Ice

Machine learning (ML) models for sea ice in the Arctic have fewer complexities and are more computationally efficient than traditional physics-based models. This work uses remote sensing data and model output from CMIP6 to build ML models that make skillful , predictions of sea ice properties (sea-ice motion, sea-ice extent, etc.) in the Arctic on various time scales (daily, seasonal, interannual). I also implement eXplainable Artificial Intelligence (XAI) techniques to understand the predictions made by the ML models and gain valuable insight into the drivers of variability for sea ice properties.

Related publications:

Hoffman, L., Mazloff, M. R., Gille, S. T., Giglio, D., Bitz, C. M., Heimbach, P. (2024, in prep). Machine learning for evaluating trends in the drivers of variability in Arctic sea-ice dynamics.
Hoffman, L., Mazloff, M. R., Gille, S. T., Giglio, C. M., Heimbach, P. (2024, submitted). Evaluating the trustworthiness of explainable artificial intelligence (XAI) methods applied to regression predictions of Arctic sea-ice motion

Hoffman, L., Mazloff, M. R., Gille, S. T., Giglio, D., Bitz, C. M., Heimbach, P., Matsuyoshi, K. (2023). Machine learning for daily forecasts of Arctic sea-ice motion: an attribution assessment of model predictive skill.

Modeling Sea-Ice Dynamics

Machine learning is a useful tool to predict and understand sea-ice motion in the Arctic. Predictability: I build ML models to make one-day predictions of Arctic sea-ice motion for operational forecasting. Understanding: In a changing Arctic, sea ice is becoming more responsive to wind forcing. Historically, wind velocity has been shown to explain 70% of the variability in sea-ice velocity. However, the response of sea-ice to the dynamic mechanisms that drive its motion is changing as it melts. I use XAI techniques to understand the nuances in this changing relationship between wind and ice motion.

Related publications:

Hoffman, L., Mazloff, M. R., Gille, S. T., Giglio, D., Bitz, C. M., Heimbach, P. (2024, in prep). Machine learning for evaluating trends in the drivers of variability in Arctic sea-ice dynamics.

Modeling Upper Ocean Salinity Response to Atmospheric Rivers

Atmospheric rivers (ARs) are responsible for a large fraction of the precipitation in the California Current System. Rainfall on the ocean can generate freshwater lenses (i.e. a buoyant layer of fresh water at the surface) that impact ocean stratification and inhibit exchanges between the surface and the mixed layer. This work employed a combination of observations and a dynamical model (the MITgcm) to understand the surface salinity response to atmospheric river precipitation on both event and seasonal timescales. Seasonally, greater precipitation over the rainy season is linked to a larger decrease in salinity over time. On event timescales, AR precipitation events commonly produce a surface salinity response that is detectable by ocean instruments.

Related publications:

Hoffman, L., Mazloff, M. R., Gille, S. T., Giglio, D., Varadarajan, A. (2022). Ocean Surface Salinity Response to Atmospheric River Precipitation in the California Current System.