Bias corrected CORDEX AFR-44 data. Bias correction done using QDM approach (Cannon et al. 2015) with downscaleR package (https://github.com/SantanderMetGroup/downscaleR). Bias correction to earth2observe data (https://wci.earth2observe.eu/)
 
############################
data organization:

fullperiod - contains files for individual projections, models, variables covering the entire period of 1951-2099, with the following data model:

fullperiod/{scenario}/{variable}/{variable}_{basetime}_{rcm}_{scenario}_{gcm}_{cordex_domain}_{startdate}-{enddate}_bc_{bcmethod}.nc

where:
variable - is variable name
basetime - is temporal resolution (day)
rcm - is Regional Climate Model that generated the data
scenario - is the GHG emission scenario, here it denotes data from historical and RCP experiments merged together
gcm - is GCM from which boundary conditions for the RCM were derived
cordex_domain - is CORDEX domain (here AFR44)
startdate - is the first date of the file
enddate - is the last date of the file
bcmethod - is bias correction method applied (here qdm)



gwl - contains data processed to represent global warming levels
details of gwl dataset in gwl/readme_gwl.txt

############################
bias correction procedure

1. Preprocessing
- E2o data, originally at 3h time step were processed to calculate daily (0-24h)
 totals or means, depending on variable, and individual monthly files were merged to a single file spanning 1979-2014, and subset to southern Africa domain (10E to 52E, 35S to 5S).
- for the purpose of bias correction, files covering the entire domain were split into quarters:
    NE 31E,52E,20S,4S
    SE 31E,52E,36S,20S
    SW 10E,31E,36S,20S
    NW 10E,31E,20S,4S

- CORDEX data for all listed variables were subset to the southern Africa domain, maintaining original resolution
- historical data were merged with rcp data creating dataset spanning 1951-2099
- files covering the entire domain were split into quarters as described above

2. Bias correction
Bias correction was implemented using downscaleR package (https://github.com/SantanderMetGroup/downscaleR), quantile delta mapping (QDM) approach, with the following parameters:

reference period: 1980-2014
target period: 1951-2099


tas, tasmax, tasmin were processed as absolute variables
remaining variables (pr, ps, rsds, rlds, huss, sfcWind) were processed as "ratio" variables

threshold for filling with random numbers: 
for precipitation: 1mm/day
for other variables: 0

moving window parameters: 30,10

3. Postprocessing
- bias corrected files in the four quarters were merged to a single file for the entire domain



############################
variables:
huss
pr
ps
rlds
rsds
sfcWind
tas
tasmax
tasmin

############################
spatial domain:

southern Africa
10E to 52E
35S to 5S

############################
spatial resolution:

0.25 deg


############################
temporal resolution:

daily

############################
temporal coverage:

19510101 to 20991231

############################
scenarios:

rcp45 merged with historical 
rcp85 merged with historical

inventory of models and variables available for each model in inventory files in each sub-directory