File size: 8,323 Bytes
f2973fb
 
e7b469c
f2973fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2c4a5fe
f2973fb
 
 
 
 
 
 
4beb26f
f2973fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
471dee3
 
 
 
 
 
f2973fb
 
 
 
 
 
fb273f3
 
 
 
 
 
 
f2973fb
 
 
 
 
 
 
fb273f3
 
 
 
 
 
 
f2973fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5bd9c0d
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
---
language: []
pretty_name: "DASP"
tags:
  - satellite-imagery
  - remote-sensing
  - earth-observation
  - sentinel-2
  - geospatial
license: "cc-by-sa-3.0"
task_categories:
  - image-segmentation
  - image-classification
  - object-detection
  - other
---

# Dataset Card for DASP

## Dataset Description

The DASP **(Distributed Analysis of Sentinel-2 Pixels)** dataset consists of cloud-free satellite images captured by Sentinel-2 satellites. Each image represents the most recent, non-partial, and cloudless capture from over 30 million Sentinel-2 images in every band. The dataset provides a near-complete cloudless view of Earth's surface, ideal for various geospatial applications. Images were converted from JPEG2000 to **JPEG-XL** to improve storage efficiency while maintaining high quality.

**Huggingface page:** https://huggingface.co/datasets/RichardErkhov/DASP

**Github repository:** https://github.com/nicoboss/DASP

**Points of Contact:**
- [Richard's Discord](https://discord.gg/pvy7H8DZMG)
- [Richard's GitHub](https://github.com/RichardErkhov)
- [Richard's website](https://erkhov.com/)
- [Nico Bosshard's website](https://www.nicobosshard.ch)
- [Nico Bosshard's github](https://github.com/nicoboss)

### Dataset Summary
- Full cloudless satellite coverage of Earth.
- Sourced from Sentinel-2 imagery, selecting the most recent cloud-free images.
- JPEG2000 images transcoded into JPEG-XL for efficient storage.
- Cloudless determination based on B1 band black pixel analysis.
- Supports AI-based image stitching, classification, and segmentation.

### Use cases
- **Image Stitching:** Combines individual images into a seamless global mosaic.
- Enables high-resolution satellite mosaics for academic and commercial applications.
- Supports AI-driven Earth observation projects.
- Facilitates urban planning, climate research, and environmental monitoring.
- Land Use Classification: Enables categorization of land cover types.

## Download a band (folder) 

```sh
huggingface-cli download RichardErkhov/DASP --include TCI/* --local-dir DASP --repo-type dataset
```

## Dataset Structure

### Data Instances

The resulting image are in separate folders named after their band. The image names can be collated to the provided metadata. The ZStatandard compression algorithm was used to compress the metadata.

### File: Sentinel_B1_black_pixel_measurements.txt

Header: 
```
URL, total black pixels, black pixels top, black pixels right, black pixels bottom, black pixels left, average grayscale value of all non-black pixels
```

Sample data:
```
http://storage.googleapis.com/gcp-public-data-sentinel-2/tiles/43/N/CA/S2A_MSIL1C_20220401T051651_N0400_R062_T43NCA_20220401T075429.SAFE/GRANULE/L1C_T43NCA_A035380_20220401T053643/IMG_DATA/T43NCA_20220401T051651_B01.jp2: 62262 0,747,166,0 20
http://storage.googleapis.com/gcp-public-data-sentinel-2/tiles/36/M/XD/S2B_MSIL1C_20190716T074619_N0208_R135_T36MXD_20190716T104338.SAFE/GRANULE/L1C_T36MXD_A012316_20190716T080657/IMG_DATA/T36MXD_20190716T074619_B01.jp2: 0 0,0,0,0 20
http://storage.googleapis.com/gcp-public-data-sentinel-2/tiles/20/V/LJ/S2A_MSIL1C_20200629T154911_N0209_R054_T20VLJ_20200629T193223.SAFE/GRANULE/L1C_T20VLJ_A026220_20200629T155413/IMG_DATA/T20VLJ_20200629T154911_B01.jp2: 2293175 876,1830,1630,0 35
```

### File: index_Sentinel.csv

Header: 
```
GRANULE_ID,PRODUCT_ID,DATATAKE_IDENTIFIER,MGRS_TILE,SENSING_TIME,TOTAL_SIZE,CLOUD_COVER,GEOMETRIC_QUALITY_FLAG,GENERATION_TIME,NORTH_LAT,SOUTH_LAT,WEST_LON,EAST_LON,BASE_URL
```

Sample data:
```
L1C_T42UWG_A041401_20230527T062703,S2A_MSIL1C_20230527T062631_N0509_R077_T42UWG_20230527T071710,GS2A_20230527T062631_041401_N05.09,42UWG,2023-05-27T06:33:56.700000Z,764715852,0.597667731340191,,2023-05-27T07:17:10.000000Z,55.94508401564941,54.947111902793566,68.99952976138768,70.75711635116411,gs://gcp-public-data-sentinel-2/tiles/42/U/WG/S2A_MSIL1C_20230527T062631_N0509_R077_T42UWG_20230527T071710.SAFE
L1C_T33XWB_A021112_20190708T105646,S2A_MSIL1C_20190708T105621_N0208_R094_T33XWB_20190708T113743,GS2A_20190708T105621_021112_N02.08,33XWB,2019-07-08T11:00:35.000000Z,197594271,0.0,,2019-07-08T11:37:43.000000Z,73.86991541093971,72.88068077877183,16.368773276100033,18.540242190343452,gs://gcp-public-data-sentinel-2/tiles/33/X/WB/S2A_MSIL1C_20190708T105621_N0208_R094_T33XWB_20190708T113743.SAFE
L1C_T23LLJ_A028635_20201215T132230,S2A_MSIL1C_20201215T132231_N0209_R038_T23LLJ_20201215T151022,GS2A_20201215T132231_028635_N02.09,23LLJ,2020-12-15T13:25:11.367000Z,721319047,62.8896,,2020-12-15T15:10:22.000000Z,-9.946873284601002,-10.942725175756962,-46.83018842375086,-45.82296488039833,gs://gcp-public-data-sentinel-2/tiles/23/L/LJ/S2A_MSIL1C_20201215T132231_N0209_R038_T23LLJ_20201215T151022.SAFE
```

## Dataset Creation

### Collection and Processing
The dataset was curated by selecting the latest cloud-free images from **Sentinel-2** data archives. The **B1 spectrum** black pixel count was analyzed to determine partial or full images. Images with black pixels exceeding a threshold were discarded. The selected images were then transcoded from **JPEG2000 to JPEG-XL** for optimized storage.

### Source Data
- **Satellite**: Sentinel-2 (ESA)
- **Selection Criteria**:
  - Cloud coverage < 1% (from metadata)
  - Most recent full image per tile (based on B1 black pixel analysis)
    - Less than 10000 total black pixels and no more than 6 black pixels on each side of the image
- **Data Transformation**: JPEG2000 → JPEG-XL conversion

### Annotation Process
No additional annotations are provided beyond the provided metadata and B1 black pixel measurements

### Sensitive Information
The dataset contains only satellite images and does not include personal or sensitive data.

## Code used to filter images

### Filtering out partial images based ouer B1 black pixel measurments

```python
# Function to parse the data and filter URLs
def parse_and_filter_data(file_path, output_path):
    with open(file_path, 'r') as file:
        with open(output_path, 'w') as output_file:
            for line in file:
                if "Error decoding JPEG2000 image" in line:
                    continue
                if "manifest.safe does not contain B01.jp2" in line:
                    continue
                url, data = line.split(': ')
                first_number, comma_separated, _ = data.split(' ')
                first_number = int(first_number)
                comma_separated_numbers = list(map(int, comma_separated.split(',')))
                
                if first_number < 10000 and all(num <= 6 for num in comma_separated_numbers):
                    output_file.write(url + '\n')
                    #print(line)

# Example usage
file_path = 'Sentinel_B1_black_pixel_measurements.txt'
output_path = 'filteredUrls.txt'
parse_and_filter_data(file_path, output_path)
```

### Extracting URLs of Cloudless Images

```python
import csv
from datetime import datetime

data = {}
print("Reading index_Sentinel.csv...")
with open('index_Sentinel.csv', 'r') as csvfile:
    reader = csv.DictReader(csvfile)
    for row in reader:
        try:
            cloud_cover = float(row['CLOUD_COVER'])
        except ValueError:
            continue
        if cloud_cover < 1:
            mgrs_tile = row['MGRS_TILE']
            sensing_time = datetime.fromisoformat(row['SENSING_TIME'].replace('Z', '+00:00'))
            if mgrs_tile not in data or sensing_time > data[mgrs_tile]['SENSING_TIME']:
                data[mgrs_tile] = {
                    'SENSING_TIME': sensing_time,
                    'GRANULE_ID': row['GRANULE_ID']
                }
print("Finished reading index_Sentinel.csv.")

filtered_urls = []
with open('filteredUrls.txt', 'r') as urlfile:
    for line in urlfile:
        granule_id = line.split('/')[10]
        if granule_id in data:
            filtered_urls.append(line.strip().replace('_B01.jp2', '_TCI.jp2'))

print(f"Number of filtered URLs: {len(filtered_urls)}")
with open('noCloudURLs.txt', 'w') as outfile:
    outfile.write('\n'.join(filtered_urls))
print("Filtered URLs saved.")
```



## Citation
If you use this dataset, please cite:

```
@misc{DASP,
  author    = {Richard Erkhov and Nico Bosshard},
  title     = {DASP},
  year      = {2025},
  url       = {https://huggingface.co/datasets/RichardErkhov/DASP}
}
```