Global Temperature Time Series Analysis

Singular Spectrum Analysis and ARIMA models implemented on Berkeley Earth Surface Time Series
Report Bugs

The Berkeley Earth Surface Temperature Study combines 1.6 billion temperature reports from 16 pre-existing archives. It is nicely packaged and allows for slicing into interesting subsets (for example by country). Time series analysis is performed on this dataset.

Link to the dataset

Kaggle Notebook for visualizations and ARIMA

Kaggle Notebook for SSA

Few points

• statsmodel.api is used for ARIMA implementation
• SSA (decomposition, reconstruction, forecasting) is implemented from scratch as presented in Golyandina, Nina & Zhigljavsky, Anatoly. (2013). Singular Spectrum Analysis for Time Series. 10.1007/978-3-642-34913-3.
• Python scientific stack is used to simplify all implementations - NumPy, Pandas, SciPy, Seaborn, Matplotlib
• PNG Output of all plots are in the Output folder

Decompositions and outputs

Decomposition of time series for ARIMA

Forecast for ARIMA - mse = 0.09 (on same data for last 12 points i.e. year 2015)

Decomposition of time series for SSA

Forecast for SSA - mse = 0.085 (on same data for last 12 points i.e. year 2015)

Contact

@LinkedIn - swaranjananayak@gmail.com

Project Link: https://github.com/sn2606/Global-Temperature-Time-Series

Acknowledgements

• This Kaggle tutorial notebook
• Github Rebository - pssa
• Deng, Cheng, "Time Series Decomposition Using Singular Spectrum Analysis" (2014). Electronic Theses and Dissertations. Paper 2352. https://dc.etsu.edu/etd/2352
• Golyandina, Nina & Zhigljavsky, Anatoly. (2013). Singular Spectrum Analysis for Time Series. 10.1007/978-3-642-34913-3.
• ARIMA Model Python Example — Time Series Forecasting
• Time Series Data Visulaization with Python