CMIX II Reference Data
HOME |
PROCESSORS |
REFERENCE DATA |
METRICS |
RESULTS |
CONTACT |
Reference Data
Algorithm outputs were evaluated against community generated, expert labeled, and optical based (“ground truth”) cloud mask datasets. Those datasets varied in the way they were sampled and geographically distributed, sample unit used (points or full image labels), and generated (experts, participants, sky images).
| Dataset | Spatial domain | Level of automatization | Purpose | Thematic depth | Satellites | Spatial resolution | #scenes |
| IRIS S2 | Subsets of Sentinel-2 Products | Manually selected and classified by an expert with machine assistance | Validation | Shallow (4 classes) | Sentinel-2 MSI | 20 m | 30 |
| IRIS Landsat | Subsets of Landsat 8 & 9 Products | Manually selected and classified by an expert with machine assistance | Validation | Shallow (4 classes) | Landsat 8 & 9 OLI | 30 m | 30 |
| Pixbox S2 | Single pixels | Manually selected and classified by an expert | Validation | Very high (10 and more categories with multiple classes) | Sentinel-2 MSI | 10 m | 100 |
| Pixbox Landsat | Single pixels | Manually classified by an expert | Validation | Very high (10 and more categories with multiple classes) | Landsat 8 and 9 OLI | 30 m | 99 |
| GSFC SkyCam | |||||||
| Multi-temporal S2 | Complete S2 products | Full year timeseries yielding cloud percentage and cloud sum | Validation | Shallow (4 classes) | Sentinel-2 MSI | 10 m | 437 S2 scenes over 4 AOIs |
| Multi-temporal Landsat | Complete LS products | Full year timeseries yielding cloud percentage and cloud sum | Validation | Shallow (4 classes) | Landsat 8 and 9 OLI | 30 m | 184 LS scenes split evenly over 4 AOIs |
