Poster Abstracts

This work compares the nature of persistence forecasts in the Arctic and Antarctic. We use a diagnostic, lag correlation approach to look at predictability in total ice area and volume on monthly timescales in CESM(CAM5). The model runs analyzed include the CESM Large Ensemble control runs (fully coupled and slab ocean) and a high resolution, slab ocean version of CESM. These results point out several noteworthy differences between the poles. Typically, lag correlations of model sea ice area in the Arctic show reemergence of correlations related to sea surface temperature (SST) and ice thickness anomalies. SST related reemergence was found in the Antarctic as well, but thickness related reemergence was not. Comparing the fully coupled and slab ocean runs showed interhemispheric differences in how ocean dynamics affected the strength of area correlations, with Antarctic predictability being diminished in the presence of ocean dynamics. Taking a closer look at the Antarctic, we suggest that ice advection is a driver of low predictability in several regions. In the Arctic, fluctuations in area and volume correlations during different periods of the runs reveal low frequency climate variability changing the effectiveness of a monthly persistence forecast.

The decline in extent and thickness of Arctic sea ice is an active area of scientific effort and one with significant implications for ecosystems and communities. Forecasting for seasonal timescales is of particular interest. However, the variable and complex nature of Arctic weather and ocean conditions combined with the limits of data availability make seasonal forecasting particularly challenging. The Sea Ice Prediction Network (SIPN: https://www.arcus.org/sipn), launched in late 2013, is a network of scientists and others collaborating to improve Arctic sea ice prediction knowledge and tools. The project’s objectives—to coordinate and evaluate predictions; integrate, assess, and guide observations; synthesize predictions and observations; and disseminate predictions and engage key stakeholders—are supported by a range of activities, including the Sea Ice Outlook, post-melt season analysis and reports, workshops, webinars, and other communication tools such as a mailing list and website. This poster will present information about levels of participation in SIPN activities, the geographic and institutional range of participants, and how participation is changing over time.

Since 2001, we have been forecasting the climatic fields of the Antarctic sea ice (SI) fields and surface air temperature, surface pressure and precipitation anomalies for the Southern Hemisphere at the Meteorological Department of the Argentine Naval Hydrographic Service. Sea Ice forecast are used principally to prepare Navy Summer campaigns to Antarctica but the atmospheric forecasts are used to make the monthly forecast three months ahead. Forecast is based on the results of S-Mode Principal Components Analysis applied on SI series that gives patterns of temporal series with validity areas. These series are important to determine which areas in Antarctica will have positive or negative SI anomalies based on what happen in the atmosphere. On the other hand, T-Mode Principal Components Analysis applied on SI fields gives us the form of the SI fields anomalies based on a classification of 16 patterns. Each pattern has unique atmospheric fields associated to them. Therefore, it is possible to forecast whichever atmosphere variable we decide for the Southern Hemisphere. When the forecast is obtained, each pattern has a probability of occurrence and sometimes it is necessary to compose more than one of them to obtain the final result. S-Mode and T-Mode are monthly updated with new data, for that reason the forecasts improved with the increase of cases since 2001. We used the Monthly Polar Gridded Sea Ice Concentrations database derived from satellite information generated by NASA Team algorithm provided monthly by the National Snow and Ice Data Center of USA that begins in November 1978.
Recently, we have been experimenting with multilayer Perceptron (neuronal network) with supervised learning and a back-propagation algorithm to improve the forecast. The Perceptron is the most common Artificial Neural Network topology dedicated to image pattern recognition. It was implemented through the use of temperature and pressure anomalies field images that were associated with a group of sea ice anomaly patterns. The variables analyzed included only composites of surface air temperature and pressure anomalies to simplify the density of input data and avoid a non-converging solution. The obtained results are encouraging.

Global climate models continue to exhibit sensitivity to uncertainty in poorly-defined values required for parameterisations relevant to sea ice. These average values are chosen based on observations, and are then generally tuned by hand during model development, a process which may introduce human error. We report on an attempt to automate this process by treating model differences from observations as a non-linear optimisation problem. While there are limitations due to observational uncertainty, we find that this automated approach is successful in tuning the seasonal cycle of sea ice in HadCM3. There are significant improvements in both the spatial distribution and overall trends of simulated Arctic sea ice. Results for the Antarctic are poorer and indicate missing physical processes and/or Southern Ocean biases.

To improve snowpack estimates, a new snow data assimilation system (SNODAS) is established, which consists of the Community Land Model version 4 (CLM4), the land component of the Community Earth System Model (CESM), and the Data Assimilation Research Testbed (DART). The SNODAS is capable of assimilating multi-sensor satellite observations including the Moderate Resolution Imaging Spectroadiometer (MODIS) snow cover fraction (SCF) and the Gravity Recovery and Climate Experiment (GRACE) terrestrial water storage (TWS) anomalies. Results of the data assimilation of MODIS SCF and GRACE TWS observations are presented. This study also assesses the influence of uncertainties from multiple sources on the SNODAS performance. The Los Alamos sea ice model (CICE), the sea ice component of CESM, also can be linked with DART to do sea ice data assimilation with the purpose of investigating sea ice predictability stemmed from initial values of the sea ice cover. This study presents a prototype of sea ice data assimilation via DART and CICE.

Data assimilation is a convenient approach to reconstruct the history of the ocean-sea ice system, or to initialize seasonal-to-decadal climate predictions. Advanced techniques, such as the ensemble Kalman filter (EnKF), have already proven useful for Arctic and Antarctic reconstructions/predictions. However, important design choices are often overlooked when it comes to implement ensemble data assimilation methods in the models. These choices are for example: the assimilation window (interval between assimilation steps); the number of ensemble members; the radius of localization; the nature and magnitude of perturbations to apply in order to create the ensemble. Taking advantage of the recent release of the NEMO3.6/LIM3.5 ocean-sea ice model, we aim at rethinking how these parameters should be chosen before the next assimilation exercise is started. To this end, we take a deeper look than usual into the main modes of variability of the ocean-sea ice system as simulated by NEMO3.6/LIM3.5, the relationships between observable and non-observable variables and the characteristic number of degrees of freedom of the system. These results are put in the perspective of assimilation of sea ice concentration and thickness retrievals.

Recently, the development of sea ice prediction models has been extensively studied, but the predictability for four months ahead or longer remains poor due to the complexity of the Arctic climate system. Here we show a new and simple seasonal prediction of the Arctic sea ice in the late summer/early autumn. Since the radiative forcing is initially important in the sea ice melting, our focus is to predict the summer sea ice concentration (SIC) by using CERES satellite observations of absorbed solar radiation (ASR) at the top of the atmosphere that determines the heat input into the open sea and sea ice surface. In the observation, the decrease in SIC anomalies in late summer/early autumn (August-September-October) was followed by the increase in the ASR in June. Based on the lagged ASR-SIC relationship, two simple statistical models were established: the Markovian stochastic model and the linear regression model. As results, the predictability of the 4-month-ahead area-averaged Arctic SIC by both models is high (the correlation coefficients between the predicted SIC and the observed SIC were 0.29 and 0.82 by the Markovian and regression models, respectively). The results from the models still have a limitation on predicting the extreme SIC anomalies on a regional scale. Nevertheless, the statistical models designed with simplicity have a great advantage as the predictions from both models have captured the inter-annual variation of the arctic SIC.

The US Navy is developing a fully coupled global Earth System Model as part of the US national Earth System Prediction Capability (ESPC) project. Two large foci of ESPC include (1) improving operational capabilities in Polar Regions and (2) extending forecasting capabilities to 90 days. These two foci require a tightly coupled atmosphere-ocean-sea ice system. The Navy is developing a tightly coupled system based on existing mature global models that run in Navy operations. The Navy Earth System Model (NESM) includes the NAVy Global Environmental Model (NAVGEM) for the atmosphere, the HYbrid Coordinate Ocean Model (HYCOM) for the ocean, and the Community Ice CodE (CICE) for sea ice. This presentation describes the current development state of NESM focusing on polar prediction. Coupling design, physical parameterizations, and comparisons with in-situ data will be described. Hindcast testing has been performed as part of the Sea Ice Prediction Network (SIPN) Sea Ice Outlook (SIO) comparison project. These runs are multi-month predictions of Northern Hemisphere sea ice extent, and NESM produced results similar to other modeling centers. Additional hindcast testing has focused on the development of loosely-coupled data assimilation utilizing the current mature atmospheric and oceanic data assimilation systems. Two CICE versions (4 and 5) have been tested within data assimilative runs. Polar two-meter temperatures in the coupled model configuration using CICE version 5 are generally warmer over sea ice compared to the coupled configuration using CICE version 4 and stand-alone NAVGEM.

We present a robust assessment of the impact of sea ice initialisation from reconstructions of the real state on the sea ice and atmosphere prediction skill. We ran two ensemble seasonal prediction experiments from 1979 to 2012 : one using realistic sea ice initial conditions and another where sea ice is initialized from a climatology, with two forecast systems. During the melting season in the Arctic Ocean, sea ice forecasts become skilful with sea ice initialization until three to five months ahead, thanks to the memory held by sea ice thickness. During the freezing season in both the Arctic and Antarctic Oceans, sea ice forecasts are skilful for seven and two months respectively with negligible differences between the two experiments, the memory being held by the ocean heat content. A weak impact on the atmosphere prediction skill is obtained.

Linear kinematic features (LKFs) in sea ice, potentially important for short-term forecast users as for climate simulations, emerge as viscous-plastic sea ice models are used at high (<10km) resolution. Here we analyze the short-range (up to 10 days) potential predictability of LKFs in Arctic sea ice using an ocean/sea-ice model with a grid point separation of ~4.5 km. We analyze the sensitivity of predictability to idealized initial perturbations, resembling uncertainties in sea ice analyses, and to growing uncertainty of the atmospheric forcing caused by the chaotic nature of the atmosphere. For the latter we use different members of ECMWF ensemble forecasts to drive ocean/sea-ice forecasts. For our analysis, we diagnose LKFs occurrence and investigate different sea ice characteristics. On the 10-day-time scale, the model has lower predictive skill for LKFs and deformation than for sea-ice thickness and concentration.

Recent reduction of summer sea ice area in the Arctic has accelerated maritime transport using the Arctic sea route. Sea ice prediction is essential to realize safe and sustainable use of the route. Since 2010, we have predicted the summer Arctic sea-ice cover by a statistical method. For this prediction, daily ice-velocity, ice concentration, and ice thickness products are prepared using data from satellite passive microwave sensors AMSR-E and AMSR2. We found that the winter ice divergence/convergence is strongly related to the summer ice cover in some regions. This relation implies that the winter ice redistribution controls the spring ice thickness and the summer ice cover. Based on this relation, we predicted the summer ice extent using the winter sea-ice motion and initial ice thickness (http://www.1.k.u-tokyo.ac.jp/YKWP/2015arctic_e.html). As a next step, we are trying to predict the summer ice thickness distribution.

The last couple of decades has seen dramatic change in various aspects of the Arctic cryosphere. Two notable cases have been reductions in June snow cover extent and the September sea ice extent minimum. Hence, questions have been prompted about the relationship between northern hemisphere spring snow cover extent and the annual minimum sea ice extent. Using techniques adapted from machine learning our preliminary results are able to identify regions of terrestrial snow cover that produce surprisingly high correlations with sea ice extent in various sectors of the Arctic. Correlations between snow and sea ice extents are calculated in raw form, as well as using linear detrending and second order polynomial detrending so as to remove the broader climate signal. As an example, at a lead time of over three months a linearly detrended correlation of over 0.75 has been found for Laptev Sea ice extent. The spatial and temporal patterns of the relationship between terrestrial snow and sea ice extent are investigated, as are the possibilities of exploiting predictive power.

The Study of Environmental Arctic Change (SEARCH) aims to develop scientific knowledge to help society understand and respond to the rapidly changing Arctic. In September 2015, the SEARCH Sea Ice Action Team (SIAT), with a focus on science communication, developed a strategy for mobilizing the research community to organize, synthesize, and disseminate scientific knowledge related to the societal impacts of Arctic sea ice loss. Key elements are to (1) facilitate research collaboration on synthesis products that are accessible to broad and interdisciplinary audiences, (2) engage stakeholder communities through sustained and sophisticated dialogues related to impacts and their information needs, and (3) co-communicate the importance and state-of-the-art of Arctic research. The core product of the strategy will be a website to comprehensively communicate why and how "sea ice matters". This resource will provide tiered access to sea ice information, organized across a series of high-level topics using a hierarchical, pyramid structure based on increasing levels of scientific complexity. This resource will depend on collaboratively developed, peer-reviewed, and concisely edited scientific content, which will serve to coordinate the scientific community, disseminate important findings to broad audiences, and provide a take-away “go-to” resource for decision-makers and the media. In addition, "Sea Ice Matters" will host guest perspectives from across the science and stakeholders communities and provide timely information on emerging high-interest topics, such as notable weather events or recent science publications. Sea ice prediction is one of many societally relevant topics around which the SIAT is looking to collaboratively communicate the state of science and engage stakeholder communities.

Substantial error exists in current sea ice concentration predictions. Comparing predictions to observations over time exposes spatial and temporal patterns in these errors. This indicates that inaccuracies in sea ice prediction stem, at least partially, from systematic model biases. Consequently, identification of and correction for these biases has the potential to improve predictions of sea ice. In this work, we focus on the largest contiguous region in the Arctic with sea ice concentration higher than 15%. We develop a method to isolate spatial and temporal trends in the discrepancies between the observed and predicted contours surrounding this region. We then propose a statistical post-processing technique to correct for model bias that explicitly accounts for spatiotemporal structure in errors. We apply this method to sea ice concentrations predictions made by the NOAA Geophysical Fluid Dynamics Laboratory (GFDL) and satellite observations produced by NASA. This approach generally results in more accurate forecasts, suggesting that spatiotemporal bias correction can yield improved predictions.

A joint World Weather Research Programme/ World Climate Research Programme (WWRP/WCRP initiative on subseasonal to seasonal (S2S) prediction has recently been launched to foster collaboration and research in the weather and climate communities, with the goals of improving forecast skill and physical under-standing, promoting forecast uptake by operational centres, and exploitation by the applications community. A key component of the project is to create an archive of sub-seasonal operational forecasts from global producing centres (GPCs), to facilitate research and development of forecast products and solutions for early warning and managing weather risks on the time scale from 2 weeks to a season. This database became operational at ECMWF in 2015, and now includes an archive of forecasts (3 weeks behind real time), and reforecasts from 9 GPCs. The database is currently being reproduced at the Chinese Meteorological Agency (CMA) as a second archiving centre.

Polar prediction on S2S scales an important component of the S2S project that the project seeks to encourage, in collaboration with WWRP’s PPP project. Besides a large number of daily instantaneous atmospheric variables, S2S database includes several surface variables of relevance to polar prediction research on S2S timescales, including several daily snow variables and daily sea ice cover. Work is ongoing to include additional variables in the S2S archive including ice thickness.