************************************************************************
************************************************************************
Code for the clustering-based forecasting method used in the paper:
“S2S Reboot: An Argument for Greater Inclusion of Machine Learning in
Subseasonal to Seasonal (S2S) Forecasts” Judah Cohen, Dim Coumou
Jessica Hwang, Lester Mackey, Paulo Orenstein, Sonja Totz and Eli
Tziperman
************************************************************************
************************************************************************

Python version: python 2.7

library Requirements:
 - numpy
 - netCDF4
 - scipy
 - matplotlib
 - operator
 - os
 - sklearn (for CCA)

In the file functions.py the observational data as well as the mask-file is loaded in the class "forecast_variable" (line 126 & 130) 
The precursors are loaded in the class "precursors" (l147 - l.193):


In line 521 the precursors for forecasting the temperature can be selected by numbers:
0="SIC", 1="SCE", 2="SST_tropics",3="SST_atl",4="SST_medi",5="SLP_nh",6="GPH500_EU"

In line 525 the number of clusters can be chosen. It is also possible to choose a range of number of clusters.

Uncomment the following lines to save forecast plots, correlation plots and forecast data(netcdf): l. 193 - 200 and adapt paths.

Program usage: python forecast_temperature_europe_class.py