An official website of the United States government
Arctic amplification has altered the climate patterns both regionally and globally, resulting in more frequent and more intense extreme weather events in the past few decades. The essential part of Arctic amplification is the unprecedented sea ice loss as demonstrated by satellite observations. Accurately forecasting Arctic sea ice from sub-seasonal to seasonal scales has been a major research question with fundamental challenges at play. In addition to physics-based Earth system models, researchers have been applying multiple statistical and machine learning models for sea ice forecasting. Looking at the potential of data-driven approaches to study sea ice variations, we propose MT-IceNet – a UNet-based spatial and multi-temporal (MT) deep learning model for forecasting Arctic sea ice concentration (SIC). The model uses an encoder-decoder architecture with skip connections and processes multi-temporal input streams to regenerate spatial maps at future timesteps. Using bi-monthly and monthly satellite retrieved sea ice data from NSIDC as well as atmospheric and oceanic variables from ERA5 reanalysis product during 1979-2021, we show that our proposed model provides promising predictive performance for per-pixel SIC forecasting with up to 60% decrease in prediction error for a lead time of 6 months as compared to its state-of-the-art counterparts.
more »
« less
Sea Ice Forecasting using Attention-based Ensemble LSTM
Accurately forecasting Arctic sea ice from sub- seasonal to seasonal scales has been a major scientific effort with fundamental challenges at play. In addition to physics-based earth system models, researchers have been applying multiple statistical and machine learning models for sea ice forecast- ing. Looking at the potential of data-driven sea ice forecasting, we propose an attention-based Long Short Term Memory (LSTM) ensemble method to predict monthly sea ice extent up to 1 month ahead. Using daily and monthly satellite retrieved sea ice data from NSIDC and atmospheric and oceanic variables from ERA5 reanalysis product for 39 years, we show that our multi-temporal ensemble method outperforms several baseline and recently proposed deep learning models. This will substantially improve our ability in predicting future Arctic sea ice changes, which is fundamental for forecasting transporting routes, resource development, coastal erosion, threats to Arctic coastal communities and wildlife.
more »
« less
- PAR ID:
- 10303962
- Date Published:
- Journal Name:
- Proceedings of Tackling Climate Change with Machine Learning Workshop at International Conference on Machine Learning (ICML 2021)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Anthropogenic warming has led to an unprecedented year-round reduction in Arctic sea ice extent. This has far-reaching consequences for indigenous and local communities, polar ecosystems, and global climate, motivating the need for accurate seasonal sea ice forecasts. While physics-based dynamical models can successfully forecast sea ice concentration several weeks ahead, they struggle to outperform simple statistical benchmarks at longer lead times. We present a probabilistic, deep learning sea ice forecasting system, IceNet. The system has been trained on climate simulations and observational data to forecast the next 6 months of monthly-averaged sea ice concentration maps. We show that IceNet advances the range of accurate sea ice forecasts, outperforming a state-of-the-art dynamical model in seasonal forecasts of summer sea ice, particularly for extreme sea ice events. This step-change in sea ice forecasting ability brings us closer to conservation tools that mitigate risks associated with rapid sea ice loss.more » « less
-
Important natural resources in the Arctic rely heavily on sea ice, making it important to forecast Arctic sea ice changes. Arctic sea ice forecasting often involves two connected tasks: sea ice concentration at each pixel and overall sea ice extent. Instead of having two separate models for two forecasting tasks, in this report, we study how to use multi-task learning techniques and leverage the connections between ice concentration and ice extent to improve accuracy for both prediction tasks. Because of the spatiotemporal nature of the data, we designed two novel multi-task learning models based on CNNs and ConvLSTMs, respectively. We also developed a custom loss function which trains the models to ignore land pixels when making predictions. Our experiments show our models can have better accuracies than separate models that predict sea ice extent and concentration separately, and that our accuracies are better than or comparable with results in the state-of-the-art studies.more » « less
-
This dataset contains data used in the paper: Revisiting the Last Ice Area Projections from a High-Resolution Global Earth System Model - Fol et al (2025). Results are organized in excel files or numpy arrays with the dataset name, variable and ensemble member (for simulations) in the name of the file. See below for more information on what variables are included in the files and their structure. CESM_HR : CESM_HR_SIAFluxes - per ensemble member: Timeseries of monthly SIA flux per gate. CESM_HR_fluxesCAADiv.npy, _fluxesQEIDiv.npy, _fuxesQEIDivMeltSeason.npy : Timeseries of annual divergence over the Queen Elizabeth Islands and the Southern Canadian Arctic Archipelago derived from monthly SIA fluxes at the entry and exit gates. CESM_HR_ThicknessDistribution.xlsx : Thickness distribution for the LIA-N, QEI, and CAA-S computed from the simulated thickness distribution (aicen001, aicen002, aicen003, aicen004, aicen005). CESM_HR_tendencies - per ensemble member- per region (LIA-N, QEI, CAA-S). Timeseries of melt season integrated thermodynamic, dynamic (advection and ridging terms) sea ice area loss. CESH_HR_sitPanArctic - per ensemble member: Timeseries of pan-Arctic mean may sea ice thickness. CESM_HR_sieSept and CESM_HR_sieMarch - per ensemble member- per region (LIA-N, QEI, CAA-S) and pan-Arctic : Timeseries of March or September sea ice extent, sea ice area. CESM_HR_sic - per ensemble member- per region (LIA-N, QEI, CAA-S): Timeseries of mean sea ice concentration for grid cells having more than 15% of SIC (no open water). CESM_HR_meltSeason - per ensemble member- per region (LIA-N, QEI, CAA-S): Timeseries of annual freeze and melt onset dates allowing the definition of the melt season based on the thermodynamic sea ice area tendency crossing 0. CESM_HR_mean_mayThickness.npy and CESM_HR_meanseptconc.npy: Results for map of the mean september sea ice concentration and may ice thickness for 1981-2000, 2001-2020, 2021-2040 and 2041-2060. CESM_LR : CESM_LR_sieSept - per region (LIA-N, QEI, CAA-S) and pan-Arctic: Timeseries of September sea ice extent and sea ice area. CESM_LR_sitPanArctic.xlsx: Timeseries of pan-Arctic mean May sea ice thickness. CESM_LR_tendencies - per region (LIA-N, QEI, CAA-S). Timeseries of melt season integrated thermodynamic, dynamic (advection and ridging terms) sea ice area loss. CESM_LR_meltSeason - per region (LIA-N, QEI, CAA-S): Timeseries of annual freeze and melt onset dates allowing the definition of the melt season based on the thermodynamic sea ice area tendency crossing 0. CESM2_LE: CESM2_LE_CAA_sept, _LIAN_sept, _QEI_sept, panArctic: Mean September sea ice extent and sea ice area per region. There is one excel tab per ensemble member in each file. CESM2_LE_pan_Arctic_hi_may: Pan-Arctic mean May sea ice thickness. There is one excel tab per ensemble member in each file. PIOMAS: PIOMAS_panArctic_hi.xlsx: Timeseries of mean may sea ice thickness. PIOMAS_mean_1981_2000_mean_mayThickness.npy : Results for map of the mean May ice thickness for 1981-2000 and 2001-2020. Observations: CIS_marchSept_1982_1990_sie - per region (QEI and CAA-S) : Timeseries of March and September mean sea ice extent and area. CIS ice charts do not fully cover the LIA-N. NSDICCDR_1979_2023_sia - per region (LIA-N, QEI, CAA-S) and pan-Arctic : Timeseries of March and September mean sea ice extent and area. NSDICCDR_1981_2000_mean_septConc.npy: Results for map of the mean September sea ice concentration for 1981-2000 and 2001-2020. NSDICCDR_1981_2000_LIAN_sic.xlsx: Timeseries of monthly mean sea ice concentration in the LIA-N. NSDICCDR_1981_2000_QEI_sic.xlsx: Timeseries of monthly mean sea ice concentration in the QEI. These results are derived from the following datasets: Ensemble members 1 and 3 of simulations from the high-resolution Community Earth System Model version 1.3 (CESM1.3-HR) produced for the International Laboratory for High-Resolution Earth System Prediction (iHESP) by the Qingdao National Laboratory for Marine Science and Technology (QNLM), Texas A&M University (TAMU), and the U.S. National Center for Atmospheric Research (NCAR). The lower resolution simulation is also used (CESM1.3-LR). (Chang et al., 2020; Zhang et al., 2020). The 100-ensemble members Community Earth System Model (version 2) Large Ensemble (CESM2-LE) (Danabasoglu et al., 2020). Satellite-derived monthly mean SIA fluxes through entry and exit gates of the CAA and Nares Strait (Howell et a., 2019; 2021; 2023; 2024; Smedsrud et al., 2017; Kwok, 2006). The National Snow and Ice Data Center (NSIDC) Climate Data Record (CDR) (version 4) sea ice concentration, stored on a 25 x 25 km polar stereographic grid centered on the North Pole from 1979 to 2023 (Meier et al., 2021). The gridded version of the regional Canadian Ice Service (CIS) Digital Archive ice charts from the Eastern and Western Arctic regions (Tivy et al., 2011). The Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS) assimilated sea ice concentration and derived ice thickness distribution estimates in the Arctic from 1978 to 2022 (Zhang et al., 2000).more » « less
-
Abstract In this study the impact of extreme cyclones on Arctic sea ice in summer is investigated. Examined in particular are relative thermodynamic and dynamic contributions to sea ice volume budgets in the vicinity of Arctic summer cyclones in 2012 and 2016. Results from this investigation illustrate sea ice loss in the vicinity of the cyclone trajectories during each year were associated with different dominant processes: thermodynamic (melting) in the Pacific sector of the Arctic in 2012, and both thermodynamic and dynamic processes in the Pacific sector of the Arctic in 2016. Comparison of both years further suggests that the Arctic minimum sea ice extent is influenced by not only the strength of the cyclone, but also by the timing and location relative to the sea ice edge. Located near the sea ice edge in early August in 2012, and over the central Arctic later in August in 2016, extreme cyclones contributed to comparable sea ice area (SIA) loss, yet enhanced sea ice volume loss in 2012 relative to 2016. Central to a characterization of extreme cyclone impacts on Arctic sea ice from the perspective of thermodynamic and dynamic processes, we present an index describing relative thermodynamic and dynamic contributions to sea ice volume changes. This index helps to quantify and improve our understanding of initial sea ice state and dynamical responses to cyclones in a rapidly warming Arctic, with implications for seasonal ice forecasting, marine navigation, coastal community infrastructure and designation of protected and ecologically sensitive marine zones.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
