Setup after downloading

Build package from command line:
R CMD build solarbenchmarks

Install within R:
# Install dependencies
install.packages(c("ggplot2", "ggfan", "here", "ncdf4", "pracma", "reshape2", "tibble", "tidyr", "truncnorm" ))
# Install solarbenchmarks
install.packages(<solarbenchmarks_0.1.0.tar.gz in local directory>, repos=NULL)

From command line within the solarbenchmarks directory:
Rscript benchmark_forecast_comparison.R 

----------------------------------------------------------------------------


This package contains the data and functions necessarily to perform 9 benchmark 
probabilistic solar power forecasts at two temporal resolutions: climatology, 
CH-PeEn, PeEn, NWP ensemble, and Gaussian error distributions. The functions to 
perform the ECMWF-based forecasts are included, but the data is proprietary 
and cannot be directly shared; instead, batch scripts and .R scripts used to access 
the data are included for those with access to ECMWF's MARS database. 

There are 3 entry points to the package. Most users will want to jump right to #3:

1. If you would like to replicate or change the SURFRAD and/or CAMS McClear data preprocessing:
 - Unzip CAMS_McClear_files.zip
 - Unzip each of the 7 site-specific folders in SURFRAD_files
 - Execute or modify preprocess_SURFRAD.R. This will re-generate the preproessed files in the
 	GHI_files/ (already made available with the Git repo).

2. If you would like to replicate or change the ECMWF data preprocessing:
 - Follow the instructions in ECMWF_files/readme.txt to download the GRIB files from the 
 	ECMWF MARS website and translate to NetCDF files
 - Execute preprocess_ECMWF.R 

3. If you would like to jump right in to replicating benchmark forecasts with the preprocessed 
data made available with the package in the GHI_files/ folder:
 - Execute benchmark_forecast_comparison.R
 - This will generate a Results/ folder containing 200+ plots comparing the benchmarks
   for each of the 7 SURFRAD sites, .csv files containing the forecast quantiles at each
   time-step through the year 2018, and .csv files comparing the CRPS summary statistics
   for each site across the methods at each temporal scale. If the ECMWF data is not available,
   the two ECMWF-based forecasts will be skipped.