Land Cover Classification (Sentinel-2)

Finish setting up your layer

Describe your item below. Add fields on the Data tab. Configure editing on the Settings tab. Configure drawing and pop-ups through Map Viewer or Visualization tab.

Open

Deep learning model to perform land cover classification on Sentinel-2 imagery. A brief summary of the item is not available. Add a brief summary about the item.

‎Deep learning package from

Managed by

Item created: Feb 17, 2021 Item updated: Dec 27, 2024 Number of downloads: 71,769

Description

Land cover describes the surface of the earth. Land cover maps are useful in urban planning, resource management, change detection, agriculture, and a variety of other applications in which information related to earth surface is required. Land cover classification is a complex exercise and is hard to capture using traditional means. Deep learning models are highly capable of learning these complex semantics, giving superior results.

Using the model

Follow the guide to use the model. Before using this model, ensure that the supported deep learning libraries are installed. For more details, check Deep Learning Libraries Installer for ArcGIS.

Fine-tuning the model

This model can be fine-tuned using the Train Deep Learning Model tool. Follow the guide to fine-tune this model.

Input

Raster, mosaic dataset, or image service. (Preferred cell size is 10 meters.)

Note: This model is trained to work on Sentinel-2 Imagery datasets which are in WGS 1984 Web Mercator (auxiliary sphere) coordinate system (WKID 3857).

Output

Classified raster with the same classes as in Corine Land Cover (CLC) 2018.

Applicable geographies

This model is expected to work well in Europe and the United States.

Model architecture

This model uses the UNet model architecture implemented in ArcGIS API for Python.

Accuracy metrics

This model has an overall accuracy of 82.41% with Level-1C imagery and 84.0% with Level-2A imagery, for CLC class level 2 classification (15 classes). The table below summarizes the precision, recall and F1-score of the model on the validation dataset.

Class	Level-2A Imagery			Level-1C Imagery
Class	Precision	Recall	F1 Score	Precision	Recall	F1 Score
Urban fabric	0.81	0.83	0.82	0.82	0.84	0.83
Industrial, commercial and transport units	0.74	0.65	0.69	0.73	0.66	0.7
Mine, dump and construction sites	0.63	0.52	0.57	0.69	0.55	0.61
Artificial, non-agricultural vegetated areas	0.70	0.46	0.55	0.67	0.47	0.55
Arable land	0.86	0.90	0.88	0.86	0.89	0.87
Permanent crops	0.76	0.73	0.74	0.75	0.71	0.73
Pastures	0.75	0.71	0.73	0.74	0.71	0.73
Heterogeneous agricultural areas	0.61	0.56	0.58	0.62	0.51	0.56
Forests	0.88	0.93	0.90	0.88	0.92	0.9
Scrub and/or herbaceous vegetation associations	0.74	0.69	0.72	0.73	0.67	0.7
Open spaces with little or no vegetation	0.87	0.84	0.85	0.85	0.82	0.84
Inland wetlands	0.81	0.78	0.80	0.82	0.77	0.79
Maritime wetlands	0.74	0.76	0.75	0.87	0.89	0.88
Inland waters	0.94	0.92	0.93	0.94	0.91	0.92
Marine waters	0.98	0.99	0.98	0.97	0.98	0.98

This model has an overall accuracy of 90.79% with Level-2A imagery for CLC class level 1 classification (5 classes). The table below summarizes the precision, recall and F1-score of the model on the validation dataset.

Class	Precision	Recall	F1 Score
Artificial surfaces	0.85	0.81	0.83
Agricultural areas	0.90	0.91	0.91
Forest and semi natural areas	0.91	0.92	0.92
Wetlands	0.77	0.70	0.73
Water bodies	0.96	0.97	0.96

Training data

This model has been trained on the Corine Land Cover (CLC) 2018 with the same Sentinel 2 scenes that were used to produce the database. Scene IDs for the imagery were available in the metadata of the dataset.

Sample results

Here are a few results from the model. To view more, see this story.

An in-depth description of the item is not available.

Layers

Ground Layers

Tools

Tables

Basemap

Project Contents:

Solution Contents

Layers

Screenshots

Terms of Use

esri logo This work is licensed under the Esri Master License Agreement.

View Summary | View Esri Terms of Use

No special restrictions or limitations on using the item's content have been provided.

Comments (40) Sort by

Leave a comment.

Felix.Soltau_unisiegen_verm Item Owner commented a year ago Delete Reply

Hello, I downloaded Sentinel-2 data (12 Bands, 8-bit, TIFF) from EO Browser (https://apps.sentinel-hub.com/eo-browser/?zoom=14&lat=50.90669&lng=8.11701&themeId=DEFAULT-THEME&visualizationUrl=https%3A%2F%2Fservices.sentinel-hub.com%2Fogc%2Fwms%2F42924c6c-257a-4d04-9b8e-36387513a99c&datasetId=S2L1C&fromTime=2023-09-09T00%3A00%3A00.000Z&toTime=2023-09-09T23%3A59%3A59.999Z&layerId=1_TRUE_COLOR&demSource3D=%22MAPZEN%22) I already tried different solutions from the comments below (e.g. L1C or L2A). I always get a classification with a few vegetated areas and the rest is "waters", as some others already commented below. Do we already have a solution for this? Thanks in advance

Felix.Soltau_unisiegen_verm Item Owner commented a year ago Delete

Hi, thanks again! The problem is that I did not have a .SAFE file or MTD_MSIL2A.xml. I only get 12 .tiff images (Band 01-12) when I download the data as described from the eo-browser website. I changed to copernicus browser where I found another download button which gives me the files you mentioned. With these files I was able to do the classification. Can you tell what is the difference between the data via the .xml file and the 12 Bands I usually download at eo browser? What is missing in my .tiff files, that no classification is possible with this tool? Thanks a lot!

spathak_deldev Item Owner commented a year ago Delete

Hi, I downloaded the same tile which use mentioned in your comment. As the image is from 2023 we need to use radiometric offset correction of -1000. Instead of manually creating the composite raster, Follow the below steps: Open Pro Project -> Click on Add data -> Browse to your .SAFE file directory -> you will see a file named "MTD_MSIL2A.xml " -> Double click on the xml file -> Select "BOA Reflectance" raster -> Right click on the raster and Add it to the current map. After adding the BOA Reflectance product in the map -> Open Raster Functions -> Search for "Minus" function -> Use the BOA Reflectance raster as Raster and write 1000 as Raster 2. Use this minus raster in Classify Pixels Using Deep Learning tool. Please try this workflow as I am getting correct LULC classes.

Felix.Soltau_unisiegen_verm Item Owner commented a year ago Delete

Hi, of couse, thanks! I used this website: https://apps.sentinel-hub.com/eo-browser/ I searched for "Siegen, Deutschland". On the "Discover" tab I ticked "Sentinel-2", activated "Advanced search" and chose "L2A" and set "Max. cloud coverage" to 20%. I set the "Time range" from 2023-08-14 until today and used the result of 2023-09-26 (8.1% cloud coverage, map tile 32UMB). I drew a rectangle with the following coordinates: [8.085594,50.879102], [8.085594,50.929332], [8.171339,50.929332], [8.171339,50.879102]. I clicked the "Download image" button on the right hand site and chose all 12 "Raw" Bands on the "Analytical" tab. I used the "Composite Bands" tool from the Data Management Tools in ArcGIS Pro 3.1.0 and then tried to classify the result with the "Land Cover Classification (Sentinel-2)" tool from this website. Thank you so much for your help.

spathak_deldev Item Owner commented a year ago Delete

Hi, can you please share the details of area for which you are trying, I can download that tile and also try it on my end.

Felix.Soltau_unisiegen_verm Item Owner commented a year ago Delete

Thanks for your quick response. I tried the Minus GP Tool today as well as using data from 2020. I always get the same result of water classification almost everywhere. The few other small areas of different classes in the classified image seem not to fit to any objects of the original image. Do you have any idea, what else I could try? Thank you!

spathak_deldev Item Owner commented a year ago Delete

Hi, for which year you are trying to create LULC, if it is after Jan 2022 you will need to do the radiometric offset correction. For radiometric offset correction you have to minus 1000 from all the pixels you can use "Minus" GP tool or Raster Function. You can try this with Sentinel L2A BOA Reflectance product. We are also adding the radiomentric offset correct in the model as a parameter, will release it soon in this month.

wpdthinkenergy Item Owner commented a year ago Delete Reply

A possible solution has been found: In order to be able to use the dlpk (Landcover Classification Sentinal-2), the following parameters must be observed: It must be a Sentinal-2 grid with at least 12 bands. The following Sentinal-2 can be used successfully: "BOA Reflectance" This can be downloaded with an account at this e.g. page: https://dataspace.copernicus.eu/browser/?zoom=11&lat=52.38682&lng=8.06214&themeId=DEFAULT-THEME&visualizationUrl=https%3A%2F%2Fsh.dataspace.copernicus.eu%2Fogc%2Fwms%2F274a990e-7090-4676-8f7d-f1867e8474a7&datasetId=S2_L1C_CDAS&fromTime=2023-09-24T00%3A00%3A00.000Z&toTime=2023-09-24T23%3A59%3A59.999Z&layerId=1_TRUE_COLOR Extract the zip and then load the "BOA Reflectance" from the xml "MTD_MSIL2A" into ArcPro. Then work as usual with the tool "ClassifyPixelsUsingDeepLearning" and the dlpk Land Cover Classification (Sentinel-2) from the Living Atlas. The result is ok, you may have to retrain this dlpk: (tutorial https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Flearn.arcgis.com%2Fen%2Fpaths%2Ftry-deep-learning-in-arcgis%2F&data=05%7C01%7Co.goepfert%40wpd.de%7C01709a7d5f104f634da108dbdeb81f65%7Ca56e04c0ecfc4ac7a1e423eacc14ee24%7C0%7C0%7C638348654609817758%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=tjKVQMEgmTOaKHMoAgkL1U5BafP%2Flmu1kkfg%2BwPUnVM%3D&reserved=0).

[Deleted User] Item Owner commented a year ago Delete Reply

I am trying to use the model on a Sentinel-2 L2A tiff, where cell size is put to 10 as described in the documentation. ERROR 999999: Something unexpected caused the tool to fail. Contact Esri Technical Support (https://esriurl.com/support) to Report a Bug, and refer to the error help for potential solutions or workarounds. Invalid cell size Invalid cell size Failed to execute (ClassifyPixelsUsingDeepLearning).

wpdthinkenergy Item Owner commented a year ago Delete Reply

And next: ERROR 999999: Something unexpected caused the tool to fail. Contact Esri Technical Support (https://esriurl.com/support) to Report a Bug, and refer to the error help for potential solutions or workarounds. Unable to prepare input raster(s) for the python raster function. [Failed to generate table] Unable to prepare input raster(s) for the python raster function. [Failed to open dataset: Raster] Unable to prepare input raster(s) for the python raster function. Unable to open the specified raster. [raster] Client tried to access password-protected page without proper authorization. Unable to prepare input raster(s) for the python raster function. [Failed to generate table] Unable to prepare input raster(s) for the python raster function. [Failed to open dataset: Raster] Unable to prepare input raster(s) for the python raster function. Unable to open the specified raster. [raster] Client tried to access password-protected page without proper authorization. Failed to execute (DetectObjectsUsingDeepLearning).

wpdthinkenergy Item Owner commented a year ago Delete Reply

The error messages that are output after approx. 2% are as follows: ERROR 999999: Something unexpected caused the tool to fail. Contact Esri Technical Support (https://esriurl.com/support) to Report a Bug, and refer to the error help for potential solutions or workarounds. Unable to prepare input raster(s) for the python raster function. Unable to open the specified raster. [raster] Unable to extract and reorder raster bands. Invalid band index encountered. [requested band 3 at position 3] Unable to prepare input raster(s) for the python raster function. Unable to open the specified raster. [raster] Unable to extract and reorder raster bands. Invalid band index encountered. [requested band 3 at position 3] Failed to execute (ClassifyPixelsUsingDeepLearning).

Details

Dashboard views: Desktop

Source: (Image Collection)

Creating data in:

Published as:

Other Views:

Dependent items in the recycle bin

Applicable: 2d

Size: 440.047 MB

Attachments size: 0 KB

ID: afd124844ba84da69c2c533d4af10a58

Image Count: 0

Owner

Esri Managed by:

esri_analytics

Folder

Credits (Attribution)

No acknowledgements.

Esri

ArcGIS Navigator link

URL View

URL

All items were exported successfully

${numberOfItems} item(s) were exported successfully. Some item(s) skipped or failed to export.

See description for more information

Cannot import

Export packages from newer portal versions cannot be imported to older versions.