ImageCollection の反復処理

map() はコレクション内のすべての画像に関数を適用しますが、関数はコレクション内のすべての画像を個別に訪問します。たとえば、時系列から時刻 t の累積異常(At)を計算するとします。At = f(Imaget, At-1) という形式で再帰的に定義されたシリーズを取得するには、関数(f)が前の結果(At-1)に依存するため、マッピングは機能しません。たとえば、ベースラインを基準として、一連の累積正規化植生指標(NDVI)異常画像を計算するとします。A0 = 0、f(Imaget, At-1) = Imaget + At-1 とします。ここで、At-1 は時刻 t-1 までの累積異常で、Imaget は時刻 t の異常です。imageCollection.iterate() を使用して、この再帰的に定義された ImageCollection を作成します。次の例では、accumulate() 関数はコレクション内の画像と、以前のすべての出力のリストの 2 つのパラメータを受け取ります。iterate() が呼び出されるたびに、異常値が累積合計に追加され、結果がリストに追加されます。最終的な結果は ImageCollection コンストラクタに渡され、新しい画像シーケンスが取得されます。

コードエディタ(JavaScript)

// Load MODIS EVI imagery.
var collection = ee.ImageCollection('MODIS/006/MYD13A1').select('EVI');

// Define reference conditions from the first 10 years of data.
var reference = collection.filterDate('2001-01-01', '2010-12-31')
  // Sort chronologically in descending order.
  .sort('system:time_start', false);

// Compute the mean of the first 10 years.
var mean = reference.mean();

// Compute anomalies by subtracting the 2001-2010 mean from each image in a
// collection of 2011-2014 images. Copy the date metadata over to the
// computed anomaly images in the new collection.
var series = collection.filterDate('2011-01-01', '2014-12-31').map(function(image) {
    return image.subtract(mean).set('system:time_start', image.get('system:time_start'));
});

// Display cumulative anomalies.
Map.setCenter(-100.811, 40.2, 5);
Map.addLayer(series.sum(),
    {min: -60000, max: 60000, palette: ['FF0000', '000000', '00FF00']}, 'EVI anomaly');

// Get the timestamp from the most recent image in the reference collection.
var time0 = reference.first().get('system:time_start');

// Use imageCollection.iterate() to make a collection of cumulative anomaly over time.
// The initial value for iterate() is a list of anomaly images already processed.
// The first anomaly image in the list is just 0, with the time0 timestamp.
var first = ee.List([
  // Rename the first band 'EVI'.
  ee.Image(0).set('system:time_start', time0).select([0], ['EVI'])
]);

// This is a function to pass to Iterate().
// As anomaly images are computed, add them to the list.
var accumulate = function(image, list) {
  // Get the latest cumulative anomaly image from the end of the list with
  // get(-1).  Since the type of the list argument to the function is unknown,
  // it needs to be cast to a List.  Since the return type of get() is unknown,
  // cast it to Image.
  var previous = ee.Image(ee.List(list).get(-1));
  // Add the current anomaly to make a new cumulative anomaly image.
  var added = image.add(previous)
    // Propagate metadata to the new image.
    .set('system:time_start', image.get('system:time_start'));
  // Return the list with the cumulative anomaly inserted.
  return ee.List(list).add(added);
};

// Create an ImageCollection of cumulative anomaly images by iterating.
// Since the return type of iterate is unknown, it needs to be cast to a List.
var cumulative = ee.ImageCollection(ee.List(series.iterate(accumulate, first)));

// Predefine the chart titles.
var title = {
  title: 'Cumulative EVI anomaly over time',
  hAxis: {title: 'Time'},
  vAxis: {title: 'Cumulative EVI anomaly'},
};

// Chart some interesting locations.
var pt1 = ee.Geometry.Point(-65.544, -4.894);
print('Amazon rainforest:',
    ui.Chart.image.series(
      cumulative, pt1, ee.Reducer.first(), 500).setOptions(title));

var pt2 = ee.Geometry.Point(116.4647, 40.1054);
print('Beijing urbanization:',
    ui.Chart.image.series(
      cumulative, pt2, ee.Reducer.first(), 500).setOptions(title));

var pt3 = ee.Geometry.Point(-110.3412, 34.1982);
print('Arizona forest disturbance and recovery:',
    ui.Chart.image.series(
      cumulative, pt3, ee.Reducer.first(), 500).setOptions(title));

Python の設定

Python API とインタラクティブな開発で geemap を使用する方法については、 Python 環境のページをご覧ください。

import ee
import geemap.core as geemap

Colab(Python)

import altair as alt
# Load MODIS EVI imagery.
collection = ee.ImageCollection('MODIS/006/MYD13A1').select('EVI')

# Define reference conditions from the first 10 years of data.
reference = collection.filterDate('2001-01-01', '2010-12-31').sort(
    # Sort chronologically in descending order.
    'system:time_start',
    False,
)

# Compute the mean of the first 10 years.
mean = reference.mean()

# Compute anomalies by subtracting the 2001-2010 mean from each image in a
# collection of 2011-2014 images. Copy the date metadata over to the
# computed anomaly images in the new collection.
series = collection.filterDate('2011-01-01', '2014-12-31').map(
    lambda image: image.subtract(mean).set(
        'system:time_start', image.get('system:time_start')
    )
)

# Display cumulative anomalies.
m = geemap.Map()
m.set_center(-100.811, 40.2, 5)
m.add_layer(
    series.sum(),
    {'min': -60000, 'max': 60000, 'palette': ['FF0000', '000000', '00FF00']},
    'EVI anomaly',
)
display(m)

# Get the timestamp from the most recent image in the reference collection.
time_0 = reference.first().get('system:time_start')

# Use imageCollection.iterate() to make a collection of cumulative anomaly over time.
# The initial value for iterate() is a list of anomaly images already processed.
# The first anomaly image in the list is just 0, with the time_0 timestamp.
first = ee.List([
    # Rename the first band 'EVI'.
    ee.Image(0)
    .set('system:time_start', time_0)
    .select([0], ['EVI'])
])

# This is a function to pass to Iterate().
# As anomaly images are computed, add them to the list.
def accumulate(image, list):
  # Get the latest cumulative anomaly image from the end of the list with
  # get(-1).  Since the type of the list argument to the function is unknown,
  # it needs to be cast to a List.  Since the return type of get() is unknown,
  # cast it to Image.
  previous = ee.Image(ee.List(list).get(-1))
  # Add the current anomaly to make a new cumulative anomaly image.
  added = image.add(previous).set(
      # Propagate metadata to the new image.
      'system:time_start',
      image.get('system:time_start'),
  )
  # Return the list with the cumulative anomaly inserted.
  return ee.List(list).add(added)

# Create an ImageCollection of cumulative anomaly images by iterating.
# Since the return type of iterate is unknown, it needs to be cast to a List.
cumulative = ee.ImageCollection(ee.List(series.iterate(accumulate, first)))

# Predefine the chart titles.
title = 'Cumulative EVI anomaly over time'

# Chart some interesting locations.
def display_chart(region, collection):
  reduced = (
      collection.filterBounds(region)
      .sort('system:time_start')
      .map(
          lambda image: ee.Feature(
              None,
              image.reduceRegion(ee.Reducer.first(), region, 500).set(
                  'time', image.get('system:time_start')
              ),
          )
      )
  )
  reduced_dataframe = ee.data.computeFeatures(
      {'expression': reduced, 'fileFormat': 'PANDAS_DATAFRAME'}
  )
  alt.Chart(reduced_dataframe).mark_line().encode(
      alt.X('time:T').title('Time'),
      alt.Y('EVI:Q').title('Cumulative EVI anomaly'),
  ).properties(title=title).display()

pt_1 = ee.Geometry.Point(-65.544, -4.894)
display('Amazon rainforest:')
display_chart(pt_1, cumulative)

pt_2 = ee.Geometry.Point(116.4647, 40.1054)
display('Beijing urbanization:')
display_chart(pt_2, cumulative)

pt_3 = ee.Geometry.Point(-110.3412, 34.1982)
display('Arizona forest disturbance and recovery:')
display_chart(pt_3, cumulative)

これらのシーケンスをグラフにすると、NDVI が以前の擾乱に比べて安定しているかどうか、または NDVI が新しい状態に向かって変化しているかどうかがわかります。Earth Engine のグラフについて詳しくは、グラフのセクションをご覧ください。

反復関数で実行できるオペレーションには制限があります。具体的には、関数の外部で変数を変更したり、何も出力したり、JavaScript の「if」ステートメントや「for」ステートメントを使用したりすることはできません。収集する結果や、次の反復処理に持ち越す中間情報を関数の戻り値に含める必要があります。条件付きオペレーションを実行するには、ee.Algorithms.If() を使用します。