Project Methodology


The goal of the project was to develop a method to provide 3-week to seasonal (S2S) sea ice guidance for a set of predefined locations based on the historical output from the operational NCEP Climate Forecast System Version 2 (CFSv2) coupled with an experimental sea ice model (CPC).

Training data

CFSv2 Experimental is a long-term, dynamic model that predicts a number of variables relatively far into the future including air temperature, ice concentration, and salinity. Historical model results for sea ice concentration and surface temperature were ingested from the model for the years 2012 to 2020 to provide the basis of the sea ice guidance model development. Time series of model results at 315 points of interest identified by the National Weather Service were extracted from the model and saved to netCDF files.

Model development

After building and evaulating a number of (S)ARIMA based models, we zeroed in on the GAM-based time-series prediction library Prophet (Taylor and Letham, 2017). Prophet has been shown to be accurate for time-series that have known seasonality and at least a few years of annual data. This fit the needs for this project, which had ~9 years of historical inputs from CFSv2, and sea ice generation has a strong seasonality based largely on the calendar year. Prophet is generally straightforward to set up and use, generates predictions based on a linear logistic curve, and handles outliers and missing data well, and is generally much faster to train compared to (S)ARIMA models.

By default, Prophet merely uses historical trends in a single variable to model future trends in that variable, and these are the models published in the accompanying github page. However, it can also include one or more regressors (e.g., air temperature) to drive the future model. I.e., if there are long-term predictions of driving variables (e.g., air temperature, wind, etc.) but ice has not been modeled, Prophet could be trained to predict ice values from the other variables. This could be useful in some circumstances, which we discuss in the evaluation portion of the project.

Daily point-based sea ice inputs showed an inherent, non-seasonal noise that seemed to contribute to unrealistic trends in the output models. We compared the prophet models produced at daily and weekly inputs and multiple stations and found that using weekly inputs produced models that produced more realistic trends.

Model assessment

We evaluated the fit of the models in a few ways, and these can be found in the evaluation portion of this site.

Data, models, and outputs

All data, models, and outputs are available on GitHub.