#STAGE 5: Output analyses: assessment of results for specific dates...

#

#AUTHOR: Benoit Parmentier                                                                       

#PROJECT: NCEAS INPLANT: Environment and Organisms --TASK#363, TASK$568--   

#source(file.path(script_path,"download_and_produce_MODIS_LST_climatology_06112013.R"))

source(file.path(script_path,"covariates_production_temperatures_08052013.R"))

source(file.path(script_path,"Database_stations_covariates_processing_function_06112013.R"))

source(file.path(script_path,"results_interpolation_date_output_analyses_08052013.R"))

#source(file.path(script_path,"results_covariates_database_stations_output_analyses_04012013.R")) #to be completed

#FUNCTIONS CALLED FROM GAM ANALYSIS RASTER PREDICTION ARE FOUND IN...

source(file.path(script_path,"interpolation_method_day_function_multisampling_07052013.R")) #Include GAM_day

source(file.path(script_path,"GAM_fusion_function_multisampling_validation_metrics_08062013.R"))

stages_to_run<-c(0,2,3,4,5) #MRun only raster fitting, prediction and assessemnt (providing lst averages, covar brick and met stations)

#If stage 2 is skipped then use previous covar object

covar_obj_file<-"/data/project/layers/commons/data_workflow/output_data_365d_gam_fus_lst_test_run_07172013/covar_obj__365d_gam_fus_lst_test_run_07172013.RData"

#If stage 3 is skipped then use previous met_stations object

met_stations_outfiles_obj_file<-"/data/project/layers/commons/data_workflow/output_data_365d_gam_fus_lst_test_run_07172013/met_stations_outfiles_obj_gam_fusion__365d_gam_fus_lst_test_run_07172013.RData"

var<-"TMAX" # variable being interpolated

out_suffix_modis <-"_05302013"                       #pattern to find tiles produced previously     

#interpolation_method<-c("gam_fusion","gam_CAI","gam_daily") #other otpions to be added later

#interpolation_method<-c("gam_fusion") #other otpions to be added later

#interpolation_method<-c("kriging_fusion") #other otpions to be added later

#interpolation_method<-c("gwr_fusion") #other otpions to be added later

#interpolation_method<-c("gwr_CAI") #other otpions to be added later

#interpolation_method<-c("kriging_CAI") 

#interpolation_method<-c("kriging_daily") #other otpions to be added later

#interpolation_method<-c("gwr_daily") #other otpions to be added later

names(list_param_data_prep) <- c("infile_monthly","infile_daily","infile_locs","infile_covariates","covar_names","var","out_prefix","CRS_locs_WGS84")

#Set additional parameters

seed_number<- 100  #if seed zero then no seed?     

nb_sample<-1           #number of time random sampling must be repeated for every hold out proportion

step<-0         

constant<-0             #if value 1 then use the same samples as date one for the all set of dates

prop_minmax<-c(0.3,0.3)  #if prop_min=prop_max and step=0 then predictions are done for the number of dates...

#dates_selected<-c("20100101","20100102","20100103","20100901") # Note that the dates set must have a specific format: yyymmdd

#dates_selected<-c("20100101","20100102","20100301","20100302","20100501","20100502","20100701","20100702","20100901","20100902","20101101","20101102")

dates_selected<-"" # if empty string then predict for the full year specified earlier

#LC1: Evergreen/deciduous needleleaf trees

#Combination 3: for paper baseline=s(lat,lon)+s(elev)

#Combination 4: for paper baseline=s(lat,lon)

#list_models<-c("y_var ~ lat*lon + elev_s")

#Collect all parameters in a list

list_param_raster_prediction<-list(list_param_data_prep,screen_data_training,

                                seed_number,nb_sample,step,constant,prop_minmax,dates_selected,

                                list_models,lst_avg,out_path,script_path,

                                interpolation_method)

names(list_param_raster_prediction)<-c("list_param_data_prep","screen_data_training",

                                "seed_number","nb_sample","step","constant","prop_minmax","dates_selected",

                                "list_models","lst_avg","out_path","script_path",

                                "interpolation_method")

raster_prediction_obj <-raster_prediction_fun(list_param_raster_prediction)

############## STAGE 5: OUTPUT ANALYSES ##################

sampling_training_testing<-function(list_param_sampling){

  #It requires 6 inputs:                                           

  # 1) seed_number: allow comparison across runs, if seed zero then no seed number is used

  # 2) nb_sample: number of time random sampling must be repeated for every hold out proportion 

  # 1) sampling_dat: sampling information for every run by date and sampling combintation

  # 2) sampling_index: list of indexes for training and testing for every dates

  # 3) sampling_stat_id: list of station ID for training and testing for every dates

  #AUTHOR: Benoit Parmentier                                                                       

  #PROJECT: NCEAS INPLANT: Environment and Organisms --TASK#363, TASK#558--     

  #Comments and TODO

  #

  prop_minmax<-list_param_sampling$prop_minmax

  dates<-list_param_sampling$dates

  #ghcn_name<-list_param_sampling$ghcn_name

  #ghcn<-get(ghcn_name) 

  ### BEGIN FUNCTION ####

  dates_list<-vector("list",nel) #list of one row data.frame

  prop_min<-prop_minmax[1]

  prop_max<-prop_minmax[2]

  prop_range<-(seq(from=prop_min,to=prop_max,by=step))*100     #range of proportion to run

  sn<-length(dates)*nb_sample*length(prop_range)               #Number of samples to run

  }

  sampling_obj<-list(sampling_dat,sampling,sampling_station_id,ghcn.subsets)

  return(sampling_obj)