Google Earth Engine JavaScript脚本运行需3-5分钟，求优化方案

我的GEE JavaScript脚本逻辑简单，但运行耗时3-5分钟。希望保留Landsat和Sentinel使用不同日期范围的灵活性，求提升运行速度的优化方法。原脚本如下：

// Set geometry of concern 
var pointJSP = ee.Geometry.Point([86.465263, 20.168076]);

// Center the map at that point.
Map.centerObject(pointJSP, 14);


// Set dates 
//Sentinel
// Time Block 1
var startDate1 = '2024-02-01';
var endDate1 = '2024-03-01';


// Landsat 8 
// Time Block 2
var startDate2 = '2024-02-01';
var endDate2 = '2024-03-01';



// Function for calculating and adding indices - NDVI, NDWI, and NDSI - to the image collections
var addIndices = function(image) {
  var ndvi = image.normalizedDifference(['NIR', 'RED']).rename('NDVI');
  var ndwi = image.normalizedDifference(['NIR', 'SWIR1']).rename('NDWI');
  var ndsi = image.normalizedDifference(['SWIR1', 'SWIR2']).rename('NDSI');
  return image.addBands(ndvi).addBands(ndwi).addBands(ndsi);
};


// Add permanent water mask
// JRC - Occurrence
var Water = ee.Image('JRC/GSW1_4/GlobalSurfaceWater').select('occurrence');
var visualization = {
  bands: ['occurrence'],
  min: 0.0,
  max: 100.0,
  palette: ['ffffff', 'ffbbbb', '0000ff']
};

Map.addLayer(Water, visualization, 'Water Occurrence', false);

// Implement a threshold.
var permWater = Water.gt(80).unmask();

// Map the threshold.
Map.addLayer(permWater,
    {
        min: 0,
        max: 1,
        palette: ['white', 'blue']
    },
    'Water vs Non-water', false);



// Visualisation Parameters 
var visTRUE = {
    bands: ['RED_p30', 'GREEN_p30', 'BLUE_p30'],
    min: 8500,
    max: 14000, 
    gamma: 1.3
};

var visTRUE_sent = {
  bands: ["RED_p30", "GREEN_p30", "BLUE_p30"], 
  min: 0.0,
  max: 0.2, 
  gamma: 1.3
}


// Define classification image visualization parameters.
var classificationVis = {
    min: 0,
    max: 2,
    palette: ['785a5a', 'ffc1b0','d63000' ] // Class 0 = others, class 1 = dry_ponds, class 2 = aquaculture
};




// ---------------------------------------------------------------------------
// Step 1.2 - Get relevant ImageCollections from Sentinel and Landsat8
// Mask and filter by location, date and cloud-free composite (ref. F4.3), calculate indices, add and rename bands

// 2.1 Sentinel 

var sentinel = sentinelBase.select
    (
        ['B2', 'B3', 'B4', 'B8', 'B11', 'B12'],
        ['BLUE', 'GREEN', 'RED', 'NIR', 'SWIR1', 'SWIR2']
    );

var sentinelIndices = sentinel.map(addIndices);

// Composite for Time Block 1
var sentinelImageCol_1 = ee.ImageCollection(sentinelIndices)
    .filterBounds(SamplingArea)
    .filterDate(startDate1, endDate1)
    .filter(ee.Filter.lessThan('CLOUDY_PIXEL_PERCENTAGE', 50));
    
print('Composite for Sentinel', sentinelImageCol_1);

var composite1 = sentinelImageCol_1.reduce(ee.Reducer.percentile([30])).clip(SamplingArea).mask(permWater.eq(0));
Map.addLayer(composite1, visTRUE_sent , 'Composite 1 Sentinel', false);


// 2.2 Landsat 8 
var landsat8 = landsat8base.select
    (
        ['SR_B2', 'SR_B3', 'SR_B4', 'SR_B5', 'SR_B6', 'SR_B7'],
        ['BLUE', 'GREEN', 'RED', 'NIR', 'SWIR1', 'SWIR2']
    );

var landsat8Indices = landsat8.map(addIndices);

// Composite for Time Block 2
var l8ImageCol_2 = ee.ImageCollection(landsat8Indices)
    .filterBounds(SamplingArea)
    .filterDate(startDate2, endDate2)
    .filter(ee.Filter.lessThan('CLOUD_COVER', 50));
    
print('Composite for Landsat 8', l8ImageCol_2);

var composite2 = l8ImageCol_2.reduce(ee.Reducer.percentile([30])).clip(SamplingArea).mask(permWater.eq(0));
Map.addLayer(composite2, visTRUE , 'Composite 2 Landsat', false);

优化建议 #

• 先过滤再计算指数：当前脚本先给整个影像集添加指数，再做空间、时间、云量过滤，会浪费算力处理大量无关图像。调整顺序：先对sentinelBase/landsat8base做filterBounds、filterDate、云量过滤，再执行select和map(addIndices)。比如Sentinel部分修改为：

  var sentinelImageCol_1 = sentinelBase
      .filterBounds(SamplingArea)
      .filterDate(startDate1, endDate1)
      .filter(ee.Filter.lessThan('CLOUDY_PIXEL_PERCENTAGE', 50))
      .select(['B2', 'B3', 'B4', 'B8', 'B11', 'B12'], ['BLUE', 'GREEN', 'RED', 'NIR', 'SWIR1', 'SWIR2'])
      .map(addIndices);
  

• 移除不必要的调试输出：print语句会触发影像集的计算，调试完成后建议注释掉，减少额外计算开销。
• 利用QA波段精准过滤云：仅靠CLOUDY_PIXEL_PERCENTAGE或CLOUD_COVER字段过滤不够精准，建议结合影像的QA波段（比如Sentinel-2的QA60、Landsat 8的QA_PIXEL）做像素级云掩膜，减少无效像素参与合成，提升计算效率。
• 优化掩码范围：permWater可以提前裁剪到SamplingArea范围，减少掩码数据的处理量：

  var permWater = Water.gt(80).unmask().clip(SamplingArea);
  

• 减少冗余波段计算：如果后续分析不需要所有指数（NDVI/NDWI/NDSI）的百分位数，在合成前筛选出需要的波段，避免计算无用数据。

内容的提问来源于stack exchange，提问作者Garima Jain