R/ccsearch.R

#' cc_search options
#'
#' @name ccafs-search
#' @export
#' @details \code{cc_params} is a list with slots matching the
#' parameters in \code{\link{cc_search}} - so you can more easily
#' get to the options you want to pass in to that function.
#'
#' The options are also listed below.
#'
#' @section file set options:
#' \itemize{
#'  \item 12: Delta Method IPCC AR5
#'  \item 4: Delta Method IPCC AR4
#'  \item 9: MarkSim Pattern Scaling
#'  \item 10: Eta South America
#'  \item 7: PRECIS Andes
#'  \item 8: CORDEX
#'  \item 11: Disaggregation IPCC AR4
#'  \item 3: Delta Climgen
#'  \item 2: Delta Method IPCC AR4 (Climgen Data)
#'  \item 5: Delta Method IPCC AR4 (Stanford Data)
#'  \item 6: Delta Method IPCC AR3
#' }
#'
#' @section scenario options:
#' \itemize{
#'  \item 1: Baseline
#'  \item 2: SRES A1B
#'  \item 3: SRES A2A
#'  \item 4: SRES B2A
#'  \item 5: SRES A2
#'  \item 6: SRES B1
#'  \item 7: RCP 2.6
#'  \item 8: RCP 4.5
#'  \item 9: RCP 6.0
#'  \item 10: RCP 8.5
#' }
#'
#' @section model options:
#' \itemize{
#'  \item 1: baseline
#'  \item 42: bcc_csm1_1
#'  \item 43: bcc_csm1_1_m
#'  \item 2: bccr_bcm2_0
#'  \item 44: bnu_esm
#'  \item 45: cccma_cancm4
#'  \item 46: cccma_canesm2
#'  \item 3: cccma_cgcm2_0
#'  \item 4: cccma_cgcm3_1_t47
#'  \item 5: cccma_cgcm3_1_t63
#'  \item 47: cesm1_bgc
#'  \item 48: cesm1_cam5
#'  \item 49: cesm1_cam5_1_fv2
#'  \item 50: cesm1_fastchem
#'  \item 51: cesm1_waccm
#'  \item 52: cmcc_cesm
#'  \item 53: cmcc_cm
#'  \item 54: cmcc_cms
#'  \item 6: cnrm_cm3
#'  \item 55: cnrm_cm5
#'  \item 56: csiro_access1_0
#'  \item 57: csiro_access1_3
#'  \item 7: csiro_mk2
#'  \item 8: csiro_mk3_0
#'  \item 25: csiro_mk3_5
#'  \item 58: csiro_mk3_6_0
#'  \item 59: ec_earth
#'  \item 60: fio_esm
#'  \item 9: gfdl_cm2_0
#'  \item 61: gfdl_cm2_1
#'  \item 10: gfdl_cm2_1
#'  \item 62: gfdl_cm3
#'  \item 63: gfdl_esm2g
#'  \item 64: gfdl_esm2m
#'  \item 11: giss_aom
#'  \item 65: giss_e2_h
#'  \item 66: giss_e2_h_cc
#'  \item 67: giss_e2_r
#'  \item 68: giss_e2_r_cc
#'  \item 26: giss_model_eh
#'  \item 27: giss_model_er
#'  \item 38: hadcm_cntrl
#'  \item 39: hadcm_high
#'  \item 40: hadcm_low
#'  \item 41: hadcm_midi
#'  \item 12: hccpr_hadcm3
#'  \item 13: iap_fgoals1_0_g
#'  \item 28: ingv_echam4
#'  \item 37: inm_cm3_0
#'  \item 69: inm_cm4
#'  \item 14: ipsl_cm4
#'  \item 70: ipsl_cm5a_lr
#'  \item 71: ipsl_cm5a_mr
#'  \item 72: ipsl_cm5b_lr
#'  \item 73: lasg_fgoals_g2
#'  \item 74: lasg_fgoals_s2
#'  \item 15: miroc3_2_hires
#'  \item 16: miroc3_2_medres
#'  \item 75: miroc_esm
#'  \item 76: miroc_esm_chem
#'  \item 77: miroc_miroc4h
#'  \item 78: miroc_miroc5
#'  \item 17: miub_echo_g
#'  \item 34: mohc_hadam3p_2
#'  \item 35: mohc_hadam3p_3
#'  \item 79: mohc_hadcm3
#'  \item 31: mohc_hadcm3q0
#'  \item 33: mohc_hadcm3q16
#'  \item 32: mohc_hadcm3q3
#'  \item 80: mohc_hadgem2_cc
#'  \item 81: mohc_hadgem2_es
#'  \item 29: mpi_echam4
#'  \item 18: mpi_echam5
#'  \item 30: mpi_echam5
#'  \item 82: mpi_esm_lr
#'  \item 83: mpi_esm_mr
#'  \item 84: mpi_esm_p
#'  \item 19: mri_cgcm2_3_2a
#'  \item 85: mri_cgcm3
#'  \item 20: ncar_ccsm3_0
#'  \item 86: ncar_ccsm4
#'  \item 21: ncar_pcm1
#'  \item 87: ncc_noresm1_m
#'  \item 88: ncc_noresm1_me
#'  \item 36: ncep_r2
#'  \item 22: nies99
#'  \item 89: nimr_hadgem2_ao
#'  \item 23: ukmo_hadcm3
#'  \item 24: ukmo_hadgem1
#' }
#'
#' @section period options:
#' \itemize{
#'  \item 1: 1970s
#'  \item 10: 1990s
#'  \item 2: 2000s
#'  \item 3: 2020s
#'  \item 4: 2030s
#'  \item 5: 2040s
#'  \item 6: 2050s
#'  \item 7: 2060s
#'  \item 8: 2070s
#'  \item 9: 2080s
#' }
#'
#' @section variable options:
#' \itemize{
#'  \item 1: Bioclimatics
#'  \item 6: Diurnal Temperature Range
#'  \item 3: Maximum Temperature
#'  \item 4: Mean Temperature
#'  \item 5: Minimum Temperature
#'  \item 2: Precipitation
#'  \item 7: Solar Radiation
#'  \item 9999: Other
#' }
#'
#' @section resolution options:
#' \itemize{
#'  \item 1: 30 seconds
#'  \item 2: 2.5
#'  \item 3: 5 minutes
#'  \item 4: 10 minutes
#'  \item 5: 30 minutes
#'  \item 6: 25 minutes
#'  \item 7: 20 minutes
#' }

cc_params <- list(
  file_set = list("Delta method IPCC AR5" = 12,
                  "Delta method IPCC AR4" = 4,
                  "MarkSim Pattern Scaling" = 9,
                  "Eta South America" = 10,
                  "PRECIS Andes" = 7,
                  "CORDEX" = 8,
                  "Disaggregation IPCC AR4" = 11,
                  "Delta Climgen" = 3,
                  "Delta Method IPCC AR4 (Climgen Data)" = 2,
                  "Delta Method IPCC AR4 (Stanford Data)" = 5,
                  "Delta Method IPCC AR3" = 6),
  scenario = list("Baseline" = 1,
                  "SRES A1B" = 2,
                  "SRES A2A" = 3,
                  "SRES B2A" = 4,
                  "SRES A2" = 5,
                  "SRES B1" = 6,
                  "RCP 2.6" = 7,
                  "RCP 4.5" = 8,
                  "RCP 6.0" = 9,
                  "RCP 8.5" = 10),
  model = list("baseline" = 1,
               "bcc_csm1_1" = 42,
               "bcc_csm1_1_m" = 43,
               "bccr_bcm2_0" = 2,
               "bnu_esm" = 44,
               "cccma_cancm4" = 45,
               "cccma_canesm2" = 46,
               "cccma_cgcm2_0" = 3,
               "cccma_cgcm3_1_t47" = 4,
               "cccma_cgcm3_1_t63" = 5,
               "cesm1_bgc" = 47,
               "cesm1_cam5" = 48,
               "cesm1_cam5_1_fv2" = 49,
               "cesm1_fastchem" = 50,
               "cesm1_waccm" = 51,
               "cmcc_cesm" = 52,
               "cmcc_cm" = 53,
               "cmcc_cms" = 54,
               "cnrm_cm3" = 6,
               "cnrm_cm5" = 55,
               "csiro_access1_0" = 56,
               "csiro_access1_3" = 57,
               "csiro_mk2" = 7,
               "csiro_mk3_0" = 8,
               "csiro_mk3_5" = 25,
               "csiro_mk3_6_0" = 58,
               "ec_earth" = 59,
               "fio_esm" = 60,
               "gfdl_cm2_0" = 9,
               "gfdl_cm2_1" = 61,
               "gfdl_cm2_1" = 10,
               "gfdl_cm3" = 62,
               "gfdl_esm2g" = 63,
               "gfdl_esm2m" = 64,
               "giss_aom" = 11,
               "giss_e2_h" = 65,
               "giss_e2_h_cc" = 66,
               "giss_e2_r" = 67,
               "giss_e2_r_cc" = 68,
               "giss_model_eh" = 26,
               "giss_model_er" = 27,
               "hadcm_cntrl" = 38,
               "hadcm_high" = 39,
               "hadcm_low" = 40,
               "hadcm_midi" = 41,
               "hccpr_hadcm3" = 12,
               "iap_fgoals1_0_g" = 13,
               "ingv_echam4" = 28,
               "inm_cm3_0" = 37,
               "inm_cm4" = 69,
               "ipsl_cm4" = 14,
               "ipsl_cm5a_lr" = 70,
               "ipsl_cm5a_mr" = 71,
               "ipsl_cm5b_lr" = 72,
               "lasg_fgoals_g2" = 73,
               "lasg_fgoals_s2" = 74,
               "miroc3_2_hires" = 15,
               "miroc3_2_medres" = 16,
               "miroc_esm" = 75,
               "miroc_esm_chem" = 76,
               "miroc_miroc4h" = 77,
               "miroc_miroc5" = 78,
               "miub_echo_g" = 17,
               "mohc_hadam3p_2" = 34,
               "mohc_hadam3p_3" = 35,
               "mohc_hadcm3" = 79,
               "mohc_hadcm3q0" = 31,
               "mohc_hadcm3q16" = 33,
               "mohc_hadcm3q3" = 32,
               "mohc_hadgem2_cc" = 80,
               "mohc_hadgem2_es" = 81,
               "mpi_echam4" = 29,
               "mpi_echam5" = 18,
               "mpi_echam5" = 30,
               "mpi_esm_lr" = 82,
               "mpi_esm_mr" = 83,
               "mpi_esm_p" = 84,
               "mri_cgcm2_3_2a" = 19,
               "mri_cgcm3" = 85,
               "ncar_ccsm3_0" = 20,
               "ncar_ccsm4" = 86,
               "ncar_pcm1" = 21,
               "ncc_noresm1_m" = 87,
               "ncc_noresm1_me" = 88,
               "ncep_r2" = 36,
               "nies99" = 22,
               "nimr_hadgem2_ao" = 89,
               "ukmo_hadcm3" = 23,
               "ukmo_hadgem1" = 24),
  extent = list("global" = "global",
                "region" = "region"),
  format = list("ascii" = "ascii",
                "esri" = "esri"),
  period = list("1970s" = 1,
                "1990s" = 10,
                "2000s" = 2,
                "2020s" = 3,
                "2030s" = 4,
                "2040s" = 5,
                "2050s" = 6,
                "2060s" = 7,
                "2070s" = 8,
                "2080s" = 9),
  variable = list("Bioclimatics" = 1,
                  "Diurnal Temperature Range" = 6,
                  "Maximum Temperature" = 3,
                  "Mean Temperature" = 4,
                  "Minimum Temperature" = 5,
                  "Precipitation" = 2,
                  "Solar Radiation" = 7,
                  "Other" = 9999),
  resolution = list("30 seconds" = 1,
                    "2.5 minutes" = 2,
                    "5 minutes" = 3,
                    "10 minutes" = 4,
                    "30 minutes" = 5,
                    "25 minutes" = 6,
                    "20 minutes" = 7),
  tile = list("A1" = "A1",
              "A2" = "A2",
              "A3" = "A3",
              "A4" = "A4",
              "A5" = "A5",
              "A6" = "A6",
              "B1" = "B1",
              "B2" = "B2",
              "B3" = "B3",
              "B4" = "B4",
              "B5" = "B5",
              "B6" = "B6",
              "C1" = "C1",
              "C2" = "C2",
              "C3" = "C3",
              "C4" = "C4",
              "C5" = "C5",
              "C6" = "C6")
)

Try the ccafs package in your browser

Any scripts or data that you put into this service are public.

ccafs documentation built on May 2, 2019, 6:34 a.m.