tags:
- time series
- time series classification
- monster
- satellite
pretty_name: LakeIce
size_categories:
- 100K<n<1M
license: cc-by-4.0
Part of MONSTER: https://arxiv.org/abs/2502.15122.
| LakeIce | |
|---|---|
| Category | Satellite |
| Num. Examples | 129,280 |
| Num. Channels | 1 |
| Length | 161 |
| Sampling Freq. | daily |
| Num. Classes | 3 |
| License | CC BY 4.0 |
| Citations | [1] [2] |
LakeIce consists of pixel-level backscatter (reflection) values from satellite images of an area of approximately 6,000 km2 in Yukon, Canada [1, 2]. The time series are extracted over three decades from ERS-1/2, Radarsat, and Sentinel-1 synthetic aperture radar satellites, which all use the C-band range of microwave frequencies (4-8GHz). This is a pixel-level dataset, such that each time series represents values over time for single pixel. The satellites used in this case have different spatial resolutions, resulting in a mixture of effective pixel sizes of between 12.5m, 30m, and 50m. The processed dataset contains 129,280 (univariate) time series each of length 161, representing daily data over near 6 months (October to March), with three classes, labelled manually, representing bedfast ice, floating ice, and land. (The original data has been calibrated and speckle-filtered, and then interpolated to provide daily values for the relevant period [1].) This version of the dataset has been split into stratified random cross-validation folds.
[1] Maria Shaposhnikova, Claude R Duguay, and Pascale Roy-Léveillée. (2023). Bedfast and floating-ice dynamics of rhermokarst lakes using a temporal deep-learning mapping approach: Case study of the Old Crow Flats, Yukon, Canada. The Cryosphere, 17(4):1697–1721.
[2] Maria Shaposhnikova, Claude R Duguay, and Pascale Roy-Léveillée. (2022). Annotated time series of lake ice C-band synthetic aperture radar backscatter created using Sentinel-1, ERS-1/2, and RADARSAT-1 imagery of Old Crow Flats, Yukon, Canada. https://doi.org/10.1594/PANGAEA.947789. CC BY 4.0.