Feature selection

[aaronspring]

Here we can discuss and link experiments and experiences related to input features:

• which spatial regions to select for EOFs?
• use rotated EOFs and how many?
• filter features by tigramite causal links?
• use EOFs at all or use CNNs?
• more variables like soil moisture

[marpyr]

I would say: -add soil moisture and SST gradient maps as predictors -discuss all together for the regions -probably already implement the regions for SSTs suggested by Chris. -keep eons that sum to 90% of explained variance and select the relevant ones with tigramite. -get prediction with 2 implementations. 1: EOFs, LSTM and 2: CNN and LSTM.