An official website of the United States government
Cloud masking is both a fundamental and a critical task in the vast majority of Earth observation problems across social sectors, including agriculture, energy, water, etc. The sheer volume of satellite imagery to be processed has fast-climbed to a scale (e.g., >10 PBs/year) that is prohibitive for manual processing. Meanwhile, generating reliable cloud masks and image composite is increasingly challenging due to the continued distribution-shifts in the imagery collected by existing sensors and the ever-growing variety of sensors and platforms. Moreover, labeled samples are scarce and geographically limited compared to the needs in real large-scale applications. In related work, traditional remote sensing methods are often physics-based and rely on special spectral signatures from multi- or hyper-spectral bands, which are often not available in data collected by many -- and especially more recent -- high-resolution platforms. Machine learning and deep learning based methods, on the other hand, often require large volumes of up-to-date training data to be reliable and generalizable over space. We propose an autonomous image composition and masking (Auto-CM) framework to learn to solve the fundamental tasks in a label-free manner, by leveraging different dynamics of events in both geographic domains and time-series. Our experiments show that Auto-CM outperforms existing methods on a wide-range of data with different satellite platforms, geographic regions and bands.
more »
« less
This content will become publicly available on June 1, 2026
Comparing Cloud Mask Products for Seagrass Mapping Over Sentinel‐2 Imagery: Toward a First National Seagrass Map for Venezuela
Despite providing many valuable ecosystem services, seagrasses are a threatened habitat and their global distribution is not fully known. For example, Venezuela lacks a national seagrass map. An established regional mapping approach for seagrass exists for the Google Earth Engine (GEE) platform, but requires a long time window to obtain sufficient data to overcome cloud and other challenges. Recently, GEE has released a Cloud Score+ quality band product for the purpose of cloud masking. Cloud masking could potentially reduce the time window needed for a representative multitemporal composite, which would allow for temporal analyses. We compare the performance of Cloud Score+ derived products against previously established multitemporal image composites acquired in different time ranges, and the ACOLITE‐processed single image composite. The Sentinel‐2 (S2) Level‐1C (L1C) imagery for the whole Venezuelan coastline was processed following three different approaches: (a) using a multitemporal composition of the full S2 L1C archive available and processed in GEE using the Dark Object Subtraction; (b) integrating Cloud Score+ data set into the previous approach; and (c) using a single‐image offline approach applying ACOLITE atmospheric correction. Additional raster features were generated and a two‐step classification approach was performed with five classes, namely sand, seagrass, turbid water, deep water, and coral, and bootstrapped 20 times. Quantitatively, the performance within the Cloud Score+ derived products were largely similar. While the full archive approach had the best quantitative results, the ACOLITE approach produced the best maps qualitatively. With this, we produced the first national seagrass map for Venezuela.
more »
« less
- PAR ID:
- 10616256
- Publisher / Repository:
- AGU
- Date Published:
- Journal Name:
- Journal of Geophysical Research: Machine Learning and Computation
- Volume:
- 2
- Issue:
- 2
- ISSN:
- 2993-5210
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Quaking aspen is an important deciduous tree species across interior western U.S. forests. Existing maps of aspen distribution are based on Landsat imagery and often miss small stands (<0.09 ha or 30 m2), which rapidly regrow when managed or following disturbance. In this study, we present methods for deriving a new regional map of aspen forests using one year of Sentinel-1 (S1) and Sentinel-2 (S2) imagery in Google Earth Engine. Using observed annual phenology of aspen across the Southern Rockies and leveraging the frequent temporal resolution of S1 and S2, ecologically relevant seasonal imagery composites were developed. We derived spectral indices and radar textural features targeting the canopy structure, moisture, and chlorophyll content. Using spatial block cross-validation and Random Forests, we assessed the accuracy of different scenarios and selected the best-performing set of features for classification. Comparisons were then made with existing landcover products across the study region. The resulting map improves on existing products in both accuracy (0.93 average F1-score) and detection of smaller forest patches. These methods enable accurate mapping at spatial and temporal scales relevant to forest management for one of the most widely distributed tree species in North America.more » « less
-
Climate warming is occurring at an unprecedented rate in the Arctic due to regional amplification, potentially accelerating land cover change. Measuring and monitoring land cover change utilizing optical remote sensing in the Arctic has been challenging due to persistent cloud and snow cover issues and the spectrally similar land cover types. Google Earth Engine (GEE) represents a powerful tool to efficiently investigate these changes using a large repository of available optical imagery. This work examines land cover change in the Lower Yenisei River region of arctic central Siberia and exemplifies the application of GEE using the random forest classification algorithm for Landsat dense stacks spanning the 32-year period from 1985 to 2017, referencing 1641 images in total. The semiautomated methodology presented here classifies the study area on a per-pixel basis utilizing the complete Landsat record available for the region by only drawing from minimally cloud- and snow-affected pixels. Climatic changes observed within the study area’s natural environments show a statistically significant steady greening (~21,000 km2 transition from tundra to taiga) and a slight decrease (~700 km2) in the abundance of large lakes, indicative of substantial permafrost degradation. The results of this work provide an effective semiautomated classification strategy for remote sensing in permafrost regions and map products that can be applied to future regional environmental modeling of the Lower Yenisei River region.more » « less
-
Seagrass ecosystems are recognized for their capacity to sequester and store organic carbon, but there is large variability in soil organic carbon stocks associated with plant traits and environmental conditions, making the quantification and scaling of carbon storage and fluxes needed to contribute to climate change mitigation highly challenging. Here, we provide estimates of carbon stocks associated with seagrass systems (biomass and soil) through analyses of a comprehensive global database including 2700+ seagrass soil cores. The median global soil Corg stock estimate is 24.2 (12.4 – 44.9) Mg Corg ha−1 in the top 30 cm of soil, 27% lower than estimates from previous global syntheses, refining the IPCC Tier 1 soil Corg stock currently used for carbon accounting in places without local data. We estimate that seagrass carbon stocks at risk of degradation could emit 1,154 Tg (665 – 1699) CO2 with a social cost of $213 billion (2020 US dollars), if no action is taken to conserve these habitats.more » « less
-
Turtlegrass virus X, which infects the seagrass Thalassia testudinum, is the only potexvirus known to infect marine flowering plants. We investigated potexvirus distribution in seagrasses using a degenerate reverse transcription polymerase chain reaction (RT-PCR) assay originally designed to capture potexvirus diversity in terrestrial plants. The assay, which implements Potex-5 and Potex-2RC primers, successfully amplified a 584 nt RNA-dependent RNA polymerase (RdRp) fragment from TVX-infected seagrasses. Following validation, we screened 74 opportunistically collected, apparently healthy seagrass samples for potexviruses using this RT-PCR assay. The survey examined the host species T. testudinum, Halodule wrightii, Halophila stipulacea, Syringodium filiforme, Ruppia maritima, Zostera marina. Potexvirus PCR products were successfully generated only from T. testudinum samples and phylogenetic analysis of sequenced PCR products revealed five distinct TVX sequence variants. Although the RT-PCR assay revealed limited potexvirus diversity in seagrasses, the expanded geographic distribution of TVX shown here emphasizes the importance of future studies to investigate T. testudinum populations across its native range and understand how the observed fine-scale genetic diversity affects host-virus interactions.more » « less
