当前位置：网站首页>Google Earth engine (GEE) - real time global 10 meter land use / land cover (LULC) data set based on S2 images

Google Earth engine (GEE) - real time global 10 meter land use / land cover (LULC) data set based on S2 images

2022-06-10 15:02:00 【Hua Weiyun】

The global dynamic land classification data set is a 10 Meters of near real time （NRT） Land use / Land cover （LULC） Data sets , It includes nine categories of probability and label information .

The prediction of the dynamic world applies to 2015-06-27 So far Sentinel-2 L1C aggregate .Sentinel-2 The frequency of revisits is 2-5 God , Depending on latitude . Dynamic world prediction is aimed at CLOUDY_PIXEL_PERCENTAGE<=35% Of Sentinel-2 L1C Image generation . The prediction results are masked , To remove clouds and cloud shadows , Use S2 Cloud probability 、 The combination of cloud displacement index and direction distance transformation .

The names of the images in the dynamic world collection and the single... From which they come Sentinel-2 L1C The asset names are consistent , for example

ee.Image('COPERNICUS/S2/20160711T084022_20160711T084751_T35PKT')

There is a matching dynamic world image , be known as ：ee.Image('GOOGLE/DYNAMICWORLD/V1/20160711T084022_20160711T084751_T35PKT') .

except " label " belt , The sum of all probability bands is 1.

To learn more about " Dynamic world " Data set information , And look at the resulting compounds 、 Examples of calculating regional statistics and processing time series , Please see the " Dynamic world " Introduction to the series .

Whereas " Dynamic world " The level estimation of is obtained from a single image through the spatial background of a small moving window , therefore , If there is no obvious distinguishing feature , Predicted land cover top-1 " probability " Will be relatively low , This part is defined by the cover that changes over time , Like crops . High returns in arid climates 、 sand 、 Sunlight, etc. may also show this phenomenon .

In order to select only pixels that are sure to belong to the dynamic world category , It is suggested that the former 1 An estimate of a forecast " probability " Perform threshold processing to mask the dynamic world output .

Dataset Availability

2015-06-23T00:00:00 -

Dataset Provider

World Resources Institute Google

Collection Snippet

ee.ImageCollection("GOOGLE/DYNAMICWORLD/V1")

Resolution

10 meters

Bands Table

Name	Description	Max
water	Estimated probability of complete coverage by water	1
trees	Estimated probability of complete coverage by trees	1
grass	Estimated probability of complete coverage by grass	1
flooded_vegetation	Estimated probability of complete coverage by flooded vegetation	1
crops	Estimated probability of complete coverage by crops	1
shrub_and_scrub	Estimated probability of complete coverage by shrub and scrub	1
built	Estimated probability of complete coverage by built	1
bare	Estimated probability of complete coverage by bare	1
snow_and_ice	Estimated probability of complete coverage by snow and ice	1
label	Index of the band with the highest estimated probability	8

Class Table: label

Value	Color Value	Description
0	#419BDF	water
1	#397D49	trees
2	#88B053	grass
3	#7A87C6	flooded_vegetation
4	#E49635	crops
5	#DFC35A	shrub_and_scrub
6	#C4281B	built
7	#A59B8F	bare
8	#B39FE1	snow_and_ice

attribute ：

Name	Type	Description
dynamicworld_algorithm_version	String	The version string uniquely identifying the Dynamic World model and inference process used to produce the image.
qa_algorithm_version	String	The version string uniquely identifying the cloud masking process used to produce the image.

Other official website links APP：

App: https://www.dynamicworld.app

Thesis link ：

https://doi.org/10.1038/s41597-022-01307-4

Code ：

//  Build a corresponding dynamic world and Sentinel-2 Set , For inspection . Filter by region and date DW and S2 Set .var COL_FILTER = ee.Filter.and(    ee.Filter.bounds(ee.Geometry.Point(20.6729, 52.4305)),    ee.Filter.date('2021-04-02', '2021-04-03'));// The two datasets are filtered according to the same results var dwCol = ee.ImageCollection('GOOGLE/DYNAMICWORLD/V1').filter(COL_FILTER);var s2Col = ee.ImageCollection('COPERNICUS/S2').filter(COL_FILTER);//  Add the corresponding DW and S2 Images （ Through the system ： Indexes ）.var DwS2Col = ee.Join.saveFirst('s2_img').apply(dwCol, s2Col,    ee.Filter.equals({leftField: 'system:index', rightField: 'system:index'}));//  Pick up a DW Examples of images and their sources S2 Images .var dwImage = ee.Image(DwS2Col.first());var s2Image = ee.Image(dwImage.get('s2_img'));// Create a will DW Class label and probability fusion visualization . Definition DW LULC A list of labels and colors .var CLASS_NAMES = [    'water', 'trees', 'grass', 'flooded_vegetation', 'crops',    'shrub_and_scrub', 'built', 'bare', 'snow_and_ice'];var VIS_PALETTE = [    '419BDF', '397D49', '88B053', '7A87C6',    'E49635', 'DFC35A', 'C4281B', 'A59B8F',    'B39FE1'];//  stay [0, 1] Create a label on （ Most likely category ） Of RGB Images .var dwRgb = dwImage    .select('label')    .visualize({min: 0, max: 8, palette: VIS_PALETTE})    .divide(255);//  Get the most likely classification probability .var top1Prob = dwImage.select(CLASS_NAMES).reduce(ee.Reducer.max());//  stay [0, 1] Create a mountain chart of the most likely category probability on .var top1ProbHillshade =    ee.Terrain.hillshade(top1Prob.multiply(100))    .divide(255);//  take RGB The image is combined with the hilly area .var dwRgbHillshade = dwRgb.multiply(top1ProbHillshade);//  Use source Sentinel-2 Images show the visualization of the dynamic world .Map.setCenter(20.6729, 52.4305, 12);Map.addLayer(    s2Image,    {min: 0, max: 3000, bands: ['B4', 'B3', 'B2']},    'Sentinel-2 L1C');Map.addLayer(    dwRgbHillshade,    {min: 0, max: 0.65},    'Dynamic World');