Multimodal Dataset Integration for Cloud Masking of ICESat-2
Abstract
The ICESat-2 mission provides higher precision elevation estimates and more complete coverage of the Earth's ice sheets than any previous mission. The ATL06 data product provides an along-track height dataset based on the retrieval of individual photons from the surface of the ice sheet, with a suite of quality flags that indicate the quality of observation. The top level flag, atl06_quality_summary, provides a single parameter to indicate whether a usable surface was identified. Multiple processes can cause the ATL06 algorithm to fail to find a surface, such as lakes with two reflecting surfaces, blowing snow, crevasses, or clouds. Alternatively, the top-level flag can indicate that a surface was found, but resulting surface elevations do not match expectations (e.g., when repeat track data varies significantly). In both cases, critical observations may be flagged and discarded or misinterpreted in an analysis. Thus, it is important to evaluate the accuracy of the ATL06 cloud flags and identify cases where cloud flagging or surface retrieval could fail. This project aims to tag cloud-contamination in ATL06 data by assessing coincident atmospheric products, such as VIIRS and MODIS, and, for cloud-free anomalous retrievals, identify the surface.
We implement automatic retrieval and fusion of MODIS and VIIRS products to ATL06, and account for temporal and spatial mismatches between the data acquisitions, as well as differences in resolution of the different products. We explore statistical and machine learning techniques (Bayesian networks and regularized regression) for modeling the relation between variables currently available in the ATL06 product and cloud flags from the other atmospheric products. We also combine atmospheric products to produce a high-quality cloud mask that can account for biases in ATL06 heights that result from scattering and refraction by clouds. Using the masked data, we identify spurious cloud-free retrievals with some distortions in the scattered photons that could lead to the automatic detection of crevasses and blowing snow. This work is part of the Jupyter meets the Earth project, supported by the NSF EarthCube program (awards 1928406 & 1928374).- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMC004.0010S
- Keywords:
-
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS;
- 0758 Remote sensing;
- CRYOSPHERE;
- 1910 Data assimilation;
- integration and fusion;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS