GEE:使用MODIS数据和随机森林方法完成生物量估算

///                   PROSAIL inversion by means of RF regression.                               //////////
///                                  Important !!!!!!                                            //////////
///   Results are not realiable due to the reduced amount of training data (500samples), the     //////////
///   origial training data set contains around 15000 samples. This code is only intended for    //////////
///   ilustrative purposes only.                                                                 //////////
/// 通过 RF 回归进行 PROSAIL 反演。 //////////
/// 重要的 ！！！！！！ //////////
/// 由于训练数据量减少（500 个样本），结果不可靠， //////////
/// 原始训练数据集包含大约 15000 个样本。此代码仅适用于 //////////
/// 仅供参考。 //////////



//Geographical Location of the selected pixel
//所选像素的地理位置

var exportarea = /* color: #ffc82d */ee.Geometry.Point([-1.10893, 42.65113]);
//The default location conrresponds with a forested area in northern part of spain.


Map.centerObject(exportarea, 14)
var modisLC = ee.Image('MODIS/051/MCD12Q1/2010_01_01')
    .select(['Land_Cover_Type_5']).reduceResolution(ee.Reducer.mode(),true,200);
    
var maskLC=modisLC.gt(0).and(modisLC.lt(9));

var mcdscale=0.0001;



Map.addLayer(exportarea,null,'Location',true)

var trainingfeatures=ee.FeatureCollection('ft:1POjOubAf8ZSvmMW3Ml3Ka3EUBt-5wY1eUg_J-04E'); //FAPAR LAI (500 samples)
var trainingfeatures2=ee.FeatureCollection('ft:1goDR2acd9mOOaY3ha092MU_NOOMRvTh0sIApn_od'); //FVC CWC EWT (500 samples)






//INPUT: Ref Nadir MODIS 
var MODIS_MCD43A4 = ee.ImageCollection("MODIS/MCD43A4").filterDate('2012-01-01','2016-12-31')//.select([0,3,6,9,12,15,18])//.first();
var MODIS_MDC43A2 = ee.ImageCollection("MODIS/MCD43A2").filterDate('2012-01-04','2016-12-31').select(0);




var idJoin = ee.Filter.equals({leftField: 'system:time_end', rightField: 'system:time_end'});
var innerJoin = ee.Join.inner('NBAR', 'QA');
var combinedCol = innerJoin.apply(MODIS_MCD43A4, MODIS_MDC43A2, idJoin);
var addQABands = function(image){
    var nbar = ee.Image(image.get('NBAR'));
    var qa = ee.Image(image.get('QA')).select(['BRDF_Albedo_Quality']);
    return nbar.addBands(qa);
};

var MODISFiltered = ee.ImageCollection(combinedCol.map(addQABands));






var inputnames=['B1','B2','B3','B4','B5','B6','B7'];


//Let's train the classifier
var RFtrained = ee.Classifier.randomForest({
 numberOfTrees: 74,
 minLeafPopulation: 3,
 bagFraction: 1, 
 outOfBagMode: false,
 seed: 123
})
.setOutputMode('REGRESSION') 
    .train({
    features: trainingfeatures,
    inputProperties: inputnames,
    classProperty: 'LAI'});   
    
print(RFtrained)      

////END of the training part of the code:


var maps =  MODISFiltered.map(function(img){
  var scaled=img.select([0,1,2,3,4,5,6],['B1','B2','B3','B4','B5','B6','B7']).toFloat().multiply(mcdscale);
  var estimation=scaled.mask(maskLC)
  .mask((img.select(7).eq(0)).or(img.select(7).eq(1)))
  .classify(RFtrained)
  return estimation.copyProperties(img, ['system:time_start']) ;
});




// Create an image time series chart.
var chart = ui.Chart.image.series({
  imageCollection: maps,
  region: exportarea,
  reducer: ee.Reducer.first(),
  scale: 500
}).setOptions({
          title: 'LAI temporal profile',
          vAxis: {title: 'LAI'},
          hAxis: {title: 'Time'},
          series: {
            0: {color: '008000', lineWidth: 2, pointSize: 4, curveType: 'function'}}
});

// Add the chart to the map.
chart.style().set({
  position: 'bottom-right',
  width: '300px',
  height: '200px'
});
Map.add(chart);




/////////////////Let's do the same with the FAPAR



var inputnames1=['B1','B2','B3','B4','B5','B6','B7'];


//Let's train the classifier
var RFtrained1 = ee.Classifier.randomForest({
 numberOfTrees: 58,
 minLeafPopulation: 3,
 bagFraction: 1, 
 outOfBagMode: false,
 seed: 123
})
.setOutputMode('REGRESSION') 
    .train({
    features: trainingfeatures,
    inputProperties: inputnames1,
    classProperty: 'FAPAR'});   
      
print(RFtrained1)     



var maps1 =  MODISFiltered.map(function(img){
  var scaled=img.select([0,1,2,3,4,5,6],['B1','B2','B3','B4','B5','B6','B7']).toFloat().multiply(mcdscale);
  var estimation=scaled.mask(maskLC).mask((img.select(7)).eq(0)).classify(RFtrained1)
  return estimation.copyProperties(img, ['system:time_start']) ;
});


////END of the training part of the code:






// Create an image time series chart.
var chart = ui.Chart.image.series({
  imageCollection: maps1,
  region: exportarea,
  reducer: ee.Reducer.first(),
  scale: 500
}).setOptions({
          title: 'FAPAR temporal profile',
          vAxis: {title: 'FAPAR'},
          hAxis: {title: 'Time'},
          series: {
            0: {color: 'B22222', lineWidth: 2, pointSize: 4, curveType: 'function'}}
});

// Add the chart to the map.
chart.style().set({
  position: 'bottom-left',
  width: '300px',
  height: '200px'
});
Map.add(chart);




/////////////////Let's do the same with the FVC


var inputnames1=['B1','B2','B3','B4','B5','B6','B7'];


//Let's train the classifier
var RFtrained2 = ee.Classifier.randomForest({
 numberOfTrees: 75,
 minLeafPopulation: 3,
 bagFraction: 1, //1.0 MATLAB
 outOfBagMode: false,
 seed: 123
})
.setOutputMode('REGRESSION') 
    .train({
    features: trainingfeatures2,
    inputProperties: inputnames1,
    classProperty: 'FVC'});   
      
print(RFtrained2)     


var maps2 =  MODISFiltered.map(function(img){
  var scaled=img.select([0,1,2,3,4,5,6],['B1','B2','B3','B4','B5','B6','B7']).toFloat().multiply(mcdscale);
  var estimation=scaled.mask(maskLC).mask((img.select(7)).eq(0)).classify(RFtrained2)
  return estimation.copyProperties(img, ['system:time_start']) ;
});


////END of the training part of the code:







// Create an image time series chart.
var chart2 = ui.Chart.image.series({
  imageCollection: maps2,
  region: exportarea,
  reducer: ee.Reducer.first(),
  scale: 500
}).setOptions({
          title: 'FVC temporal profile',
          vAxis: {title: 'FVC'},
          hAxis: {title: 'Time'},
          series: {
            0: {color: 'B22222', lineWidth: 2, pointSize: 4, curveType: 'function'}}
});

// Add the chart to the map.
chart2.style().set({
  position: 'bottom-center',
  width: '300px',
  height: '200px'
});
Map.add(chart2);





/////////////////Let's do the same with the CWC



var inputnames1=['B1','B2','B3','B4','B5','B6','B7'];


//Let's train the classifier
var RFtrained3 = ee.Classifier.randomForest({
 numberOfTrees: 74,
 minLeafPopulation: 2,
 bagFraction: 1,
 outOfBagMode: false,
 seed: 123
})
.setOutputMode('REGRESSION') 
    .train({
    features: trainingfeatures2,
    inputProperties: inputnames1,
    classProperty: 'CWC'});   
      
print(RFtrained3)     


var maps3 =  MODISFiltered.map(function(img){
  var scaled=img.select([0,1,2,3,4,5,6],['B1','B2','B3','B4','B5','B6','B7']).toFloat().multiply(mcdscale);
  var estimation=scaled.mask(maskLC).mask((img.select(7)).eq(0)).classify(RFtrained3)
  return estimation.copyProperties(img, ['system:time_start']) ;
});


////END of the training part of the code:






// Create an image time series chart.
var chart3 = ui.Chart.image.series({
  imageCollection: maps3,
  region: exportarea,
  reducer: ee.Reducer.first(),
  scale: 500
}).setOptions({
          title: 'CWC temporal profile',
          vAxis: {title: 'CWC'},
          hAxis: {title: 'Time'},
          series: {
            0: {color: 'B22222', lineWidth: 2, pointSize: 4, curveType: 'function'}}
});

// Add the chart to the map.
chart3.style().set({
  position: 'top-left',
  width: '300px',
  height: '200px'
});
Map.add(chart3);