Forecasting arctic sea ice extent trend using time series models: NNAR, SARIMA and SARIMAX using the data prior to the COVID-19 pandemic

This research paper aims to examine the patterns of Arctic sea ice extent (ASIE) between 1979 and 2020 by using monthly data acquired from NSIDC. Our study employs seasonal time series models such as SARIMA, SARIMAX with global temperature anomalies as an exogenous variable, as well as the neural network (NNAR) models to forecast the future of ASIE in the next 100 years. To evaluate the accuracy of these models, we conduct fittings and analyze various prediction error metrics, MSE, MAE, and MAPE. Our findings reveal that the SARIMAX model outperforms others in out-of-sample forecasting (testing set), effectively integrating data components and accurately capturing both trends and seasonal variations, thus enhancing predictive accuracy for the ASIE relative to other models. According to the SARIMAX model, the overall coverage of sea ice is diminishing, and it is projected that the Arctic will experience its first ice-free month no later than September of 2076, with September of 2074 being a potential milestone. Additionally, the NNAR model suggests that ASIE will not disappear within the next 100 years.