Temporal Convolutional Networks for the Advance Prediction of ENSO

Accessed on Jan 2, 2021

  • citation:

    Yan, J., Mu, L., Wang, L. et al. Temporal Convolutional Networks for the Advance Prediction of ENSO. Sci Rep 10, 8055 (2020). https://doi.org/10.1038/s41598-020-65070-5

  • notes from a blog here - accessed on Jan 2, 2021. The following are copied from the blog:

    motivation - the paper selected the following two indicators for predictions

    1. The Niño 3.4 index as an indicator of ENSO events in the ocean.
    2. The SOI(southern oscillation index ) as a measure of ENSO events in the atmosphere.

existing approaches in prediction the aforementioned indicators:

  1. Statistics-based methods: Holt-Winters (HW) method and ARIMA method(poor ftting)
  2. ML-based methods: SVR, ANNs, LSTM (complex and computationally time-consuming)
  3. Hybrid approach: ARIMA-ANNs and ensemble empirical mode decomposition (EEMD)-convolutional long short-term memory (ConvLSTM). (depend largely on the statistical decomposition model)

models proposed by the authors: EEMD-TCN: ensemble empirical mode decomposition-temporal convolutional network

  • my takeaways:
    1. this paper seems to be an application of the models described in Bai, Shaojie & Kolter, J. & Koltun, Vladlen. (2018). An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling.
    2. the authors used PCC (Pearson correlation coefficient) as a model evaluation metric.
      • Pearson correlation coefficient (PCC): a measure of the linear correlation between the predicted value and the actual value
      • The formulas for calculating the PCC is as follows:
      • title

        where m is the length of the time-series, p is the prediction results and p¯¯¯ is its mean value, o represents the actual value and o¯¯¯ represents its mean value.