Project: improving coastal sea level predictions with ocean process insight guided by ML sea level along the united states coastline is predicted to rise 0.25 to 0.3 M by 2050, SIM- ilar to what…

Project: improving coastal sea level predictions with ocean process insight guided by ML sea level along the united states coastline is predicted to rise 0.25 to 0.3 M by 2050, SIM- ilar to what previously occurred over the last 100 years (sweet and ET AL, 2022). High impact weather events and decadal variability will compound increasing flood hazards in the coming CEN- tury (cazenave and cozannet, 2014). Among the many different processes that drive sea level variability, ocean dynamics are a leading order source of model error (couldrey ET AL., 2021). Recent work in developing forecast systems of sea level anomalies on subseasonal to annual timescales (e.g. (frederikse ET AL., 2022)) demonstrate the predictive power in coupled seasonal forecast systems, including the gfdl seamless system for prediction and earth system research model (spear, 1 ocean resolution). Here, we will determine the oceanic sources of variability and provide tools for model improvement. This proposed work will identify the physical ocean processes and their interactions that contribute to sea level variability on subseasonal to decadal timescales, and thus predictability, in the gfdl spear model. Machine learning techniques will be used to 1) develop regimes or geographical regions/clusters tailored to processes that locally govern sea level variability in a hierarchy of coupled models with increasing ocean resolution (0.5, 0.25, and 0.125) developed at gfdl. The identification of the regimes will be rooted in physical constraints that arise from the individual terms of the depth-integrated barotropic vorticity (BV) equation following methods from (sonnewald ET AL., 2019, 2020). The complex spatial and temporal sea level variability patterns will be decomposed through this framework, and process-level contributions to the sea level variability and their interactions will be diagnosed. 2) the regimes of sea level will be used to train a hierarchy of neural network models of increasing complexity that can determine the dynamical regimes based on surface fields (e.g. wind stress, depth, sea surface height) following methods including uncertainty quantification and skill attribution described in (sonnewald and lguensat, 2021; clare ET AL., 2022). 3) once the neural network models are constructed, surface satellite observations will be used as input to derive the dynamical regimes of the real ocean. A series of process-oriented diagnostics will be developed that can assess the models performance in simulating the regimes and their associated processes. The diagnostics will be delivered to gfdl community and improvement of the spear model, and the scientific community through the model diagnostics task force (MDTF) framework. Finally, 4) the overall performance of the neural networks ability to recover the dynamical sea level regimes for different forecast initialization and lead times will be used to develop novel quantitative metrics of forecast uncertainty. These metrics can be applied to real-time spear predictions and observations of monthly sea level anomalies currently available through noaas physical sciences laboratory website. Understanding the skill of regime representation throughout the forecast system will highlight where where a better ocean representation would improve the forecast and where targeted data collection could aid the spear framework. Relevance to competition: the proposal will leverage resultsfrom the spear model and other coupled models developed at gfdl to inform and improve efforts to predict highly impactful sea level anomalies on subseasonal to decadal timescales. Leveraging machine learning tools and techniques, this work will characterize key sea level related ocean processes, quantifying their contributions to sea level variability, and informing development to improve their representation in ocean models. This work addresses noaas long term objectives of improving stakeholder access to quality information