Random_Forest(GEE)

//Code Written by Fauzan Muzakki, ITC, University of Twente
//Internship at Arizona State University 2021

//Zoom in to the study area
Map.centerObject(Iburi, 11)

//DEM Extraction
var elevation = SRTM.select('elevation')
.clip(JapanAOI)
var terrain = ee.Terrain.products(elevation)
function threshold(image) {  
  var slope = terrain.select('slope')
  var thres = slope.gte(8).rename('thres')
  
  return image.addBands(thres);
}

var slope = terrain.select('slope')

//cloudmasking Sentinel 2
function maskS2clouds(image) {
  var qa = image.select('QA60');

  // Bits 10 and 11 are clouds and cirrus, respectively.
  var cloudBitMask = 1 << 10;
  var cirrusBitMask = 1 << 11;

  // Both flags should be set to zero, indicating clear conditions.
  var mask = qa.bitwiseAnd(cloudBitMask).eq(0)
      .and(qa.bitwiseAnd(cirrusBitMask).eq(0));

  return image.updateMask(mask).divide(10000);
}

//map vegetation indices
function addindices(image) {
  var params= {'NIR':image.select('B8'),
      'RED':image.select('B4'),
      'BLUE':image.select('B2'),
      'SWIR':image.select('B11'),
      'GREEN': image.select('B3') }
  var ndvi =  image.normalizedDifference(['B8', 'B4']).rename(['ndvi'])
  var bsi = image.expression('((SWIR+RED)-(NIR+BLUE))/((SWIR+RED)+(NIR+BLUE))', params).rename(['bsi'])
  var br  = image.expression ('RED/(RED+GREEN+BLUE)', params).rename(['br'])
  var ndwi = image.normalizedDifference(['B8', 'B3']).rename(['ndwi'])
  
  return image.addBands(ndvi).addBands(bsi).addBands(br).addBands(ndwi);

}

////////////////////////////////////////////
//image collection Sentinel 2
/////////////////////////////////////////////
var pre_event_s2 = Sentinel2.filterDate ('2016-05-01', '2016-09-30')
                  .filterMetadata("CLOUDY_PIXEL_PERCENTAGE", "less_than", 5)
                  .filterBounds(JapanAOI)
                  .map(maskS2clouds)
                  .map(addindices)
                  .map(threshold)
                  //.median()
                  print(pre_event_s2, 'total_image_pre' )

var post_event_s2 = Sentinel2.filterDate('2018-09-01', '2018-10-30')
                   .filterMetadata("CLOUDY_PIXEL_PERCENTAGE", "less_than", 5)
                   .filterBounds(JapanAOI)
                   .map(maskS2clouds)
                   .map(addindices)
                   .map(threshold)
                   //.median()
                   print(post_event_s2, 'total_image_post' )

var visualization = {
  min: 0.0,
  max: 0.5,
  bands: ['B4', 'B3', 'B2'],
}

//////////////////////////////////////////////////////////////////
//Change Detection
//////////////////////////////////////////////////////////////////
var change_detection = post_event_s2.median().subtract(pre_event_s2.median()) // the visualization of change compared before and after landslide 
var change_ratio =change_detection.divide(pre_event_s2.median()) //the proportion of change caused by landslide 

//////////////////////////////////////////////////////////////////
//Change Detection
//////////////////////////////////////////////////////////////////
var change_detection = post_event_s2.median().subtract(pre_event_s2.median()) // the visualization of change compared before and after landslide 
var change_ratio =change_detection.divide(pre_event_s2.median()) //the proportion of change caused by landslide 

//////////////////////////////////////////////////////////////
///Random Forest
/////////////////////////////////////////////////////////////

///Samples preparation 

//Assigning a property as numeric value ('class') to landslide and non landslide class
var landslide = ee.FeatureCollection.randomPoints(Iburi, 500)
    .map(function(f){
      return f.set('class',1)
    })
print(landslide.first())

var nonlandslide = ee.FeatureCollection.randomPoints(outlandslide, 3000)
    .map(function(f){
      return f.set('class',0)
    })
print(nonlandslide.first())

var samples = landslide.merge(nonlandslide) //merging two samples
print(samples)

/////Random Forest Classification
var bands1 = ['ndvi', 'bsi', 'br']
var detailAOI = change_ratio.clip(JapanAOI2)

var training = change_ratio.select(bands1).sampleRegions({
 collection : samples,
 properties : ['class'],
 scale :10
})

var classifier = ee.Classifier.smileRandomForest(500).train({
  features : training,
  classProperty : 'class',
  inputProperties : bands1
})

var classified = detailAOI.select(bands1).classify(classifier)


////////////////////////////
Download
//////////////////////////////
Export.image.toDrive({
  image: classified, 
  description: 'RF classfied',
  folder: 'Landslide_Aufar',
  region: JapanAOI,                 
  scale: 10,
  maxPixels : 1e8
})

}) 

//////////////////////////////////////////////
//Data Visualization
/////////////////////////////////////////////
Map.addLayer(change_ratio.clip(JapanAOI2), {}, 'change');
Map.addLayer(classified, {min : 0 , max: 1 ,pallete:['red','blue']}, 'RFresult')