Exemplo de especificação de um workflow de classificação utilizando Common Workflow Language (CWL).

01_ExtractTimeSeries.R

# Autor: Felipe Menino Carlos
# Data: 25/04/2021

set.seed(777)
library(sits)

#
# Auxiliary function
#
extract_ts_by_sample_location <- function(collection, start_date, end_date, bands, sample_file) {
  cube <- sits_cube(
    type        = "BDC",
    name        = "cube_to_extract_sample",
    url         = "https://brazildatacube.dpi.inpe.br/stac/",
    collection  = collection,
    start_date  = start_date,
    end_date    = end_date,
    bands       = bands
  )
  
  samples <- sits_get_data(cube = cube, file = sample_file)
  samples
}

#
# User Defined parameters
#
args <- commandArgs(trailingOnly=TRUE)

collection       <- args[1]
collection_bands <- strsplit(args[2], ",")

#
# General definitions
#
start_date <- Sys.getenv("START_DATE")
end_date   <- Sys.getenv("END_DATE")

sample_file <- "https://brazildatacube.dpi.inpe.br/geo-knowledge-hub/bdc-article/training-samples/training-samples.csv"

#
# Output directory
#
output_dir <- paste0("training-samples/")
dir.create(
    path         = output_dir, 
    showWarnings = FALSE, 
    recursive    = TRUE
)

samples_with_ts <- extract_ts_by_sample_location(
  collection  = collection,
  start_date  = start_date,
  end_date    = end_date, 
  bands       = collection_bands[[1]],
  sample_file = sample_file
)
saveRDS(samples_with_ts, paste0(output_dir, "/", collection, ".rds"))

02_Classification.R

# Autor: Felipe Menino Carlos
# Data: 25/04/2021

set.seed(777)
library(sits)

#
# General definitions
#
args <- commandArgs(trailingOnly = TRUE)

collection <- args[1]
samples_file <- args[2]

classification_memsize    <- as.integer(Sys.getenv("CLASSIFICATION_MEMSIZE"))
classification_multicores <- as.integer(Sys.getenv("CLASSIFICATION_MULTICORES"))

start_date <- Sys.getenv("START_DATE")
end_date   <- Sys.getenv("END_DATE")

# define the roi and load samples file
samples <- readRDS(samples_file)
roi     <- readRDS(url("https://brazildatacube.dpi.inpe.br/geo-knowledge-hub/bdc-article/roi/roi.rds"))

#
# Output directory
#
output_dir <- collection
dir.create(
    path         = output_dir, 
    showWarnings = FALSE, 
    recursive    = TRUE
)

#
# Load data cube from BDC-STAC
#
cube <- sits_cube(
  type        = "BDC",
  name        = "cube_to_classify",
  url         = "https://brazildatacube.dpi.inpe.br/stac/",
  collection  = collection,
  start_date  = start_date,
  end_date    = end_date,
  roi         = roi$search_roi
)

#
# Defining tree Model
#
tree_model <- sits_rfor(num_trees = 1000)

#
# Training model
#
treined_model <- sits_train(samples, tree_model)

#
# Classify using the data cubes
#
probs <- sits_classify(data       = cube,
                       ml_model   = treined_model,
                       memsize    = classification_memsize,
                       multicores = classification_multicores,
                       roi        = roi$classification_roi,
                       output_dir = output_dir)

#
# Post-processing
#
probs_smoothed <- sits_smooth(probs, type = "bayes", output_dir = output_dir)
labels         <- sits_label_classification(probs_smoothed, output_dir = output_dir)

#
# Saving results 
#

# labels
saveRDS(
    labels, file = paste0(output_dir, "/labels.rds")
)

# probs
saveRDS(
    probs, file = paste0(output_dir, "/probs_cube.rds")
)

# smoothed probs
saveRDS(
    probs_smoothed, file = paste0(output_dir, "/probs_smoothed_cube.rds")
)

classification.tool

#!/usr/local/bin/env cwl-runner
# Autor: Felipe Menino Carlos
# Data: 25/04/2021

cwlVersion: v1.0
class: CommandLineTool

baseCommand: Rscript

inputs:
    classify_datacube_script:
        type: File
        inputBinding:
            position: 0

    collection_name:
        type: string
        inputBinding:
            position: 1

    samples_tseries_rds:
        type: File
        inputBinding:
            position: 2

    outputfile:
        type: string
        inputBinding:
            position: 3 # not used in script

outputs:
    classification:
        type:
            type: array
            items: File
        outputBinding:
            glob: $(inputs.outputfile)

extractseries.tool

#!/usr/local/bin/env cwl-runner
# Autor: Felipe Menino Carlos
# Data: 25/04/2021

cwlVersion: v1.0
class: CommandLineTool

baseCommand: Rscript

inputs:
    extractseries_script:
        type: File
        inputBinding:
            position: 0

    collection_name:
        type: string
        inputBinding:
            position: 1
    
    collection_bands:
        type: string
        inputBinding:
            position: 2

    outputfile:
        type: string
        inputBinding:
            position: 3 # not used in script

outputs:
    result:
        type: File
        outputBinding:
            glob: $(inputs.outputfile)

lulc-classification-job.yml

#
# General definitions
#
extractseries_script:
  class: File
  path: 01_ExtractTimeSeries.R

classify_datacube_script:
  class: File
  path: 02_Classification.R

#
# CBERS-4/AWFI definitions
#
cb4_collection_name: CB4_64_16D_STK-1
cb4_collection_bands: BAND15,BAND14,BAND13,BAND16,NDVI,EVI

cb4_series_output: training-samples/CB4_64_16D_STK-1.rds
cb4_classification_output: CB4_64_16D_STK-1/*.tif

lulc-classification.cwl

#!/usr/bin/env cwl-runner
# Autor: Felipe Menino Carlos
# Data: 25/04/2021

cwlVersion: v1.0
class: Workflow

hints:
    DockerRequirement:
      dockerPull: m3nin0/sits:0.10.0

requirements:

  ResourceRequirement:
    ramMin: 8192
    coresMin: 6

  EnvVarRequirement:
    envDef:
        START_DATE: "2018-09-01"
        END_DATE:   "2019-08-31"

        CLASSIFICATION_MEMSIZE: "8"
        CLASSIFICATION_MULTICORES: "6"

        BDC_ACCESS_KEY: ""

inputs:
  extractseries_script: File
  classify_datacube_script: File
  
  # CBERS-4/AWFI inputs
  cb4_series_output: string
  cb4_classification_output: string

  cb4_collection_name: string
  cb4_collection_bands: string

outputs:
  result_extract_ts_cb4:
    type: File
    outputSource: cb4_extractseries/result

  result_cb4classification:
    type: File[]
    outputSource: cb4_classification/classification

steps:
  cb4_extractseries:
    run: extractseries.tool
    in:
      extractseries_script: extractseries_script

      collection_name: cb4_collection_name
      collection_bands: cb4_collection_bands

      outputfile: cb4_series_output
    out: [result]

  cb4_classification:
    run: classification.tool
    in:
      classify_datacube_script: classify_datacube_script
      
      samples_tseries_rds: cb4_extractseries/result
      
      collection_name: cb4_collection_name
      outputfile: cb4_classification_output
    out: [classification]