Navy PIPS Model Reports Sea Ice Mass Increase Since 2007

If this picture from the Navy isreliable, it flies in the face of a lot of evidence coming out over the pastthree years.  Yet this ice mass issupposed to be around forty percent of the original ice mass measured back in1959.

The report makes no mention, orperhaps I missed it, of a history of hard measurements set to establish groundtruth.  Only the Navy has had thecapability to do that and to have done it annually for years.  It is a natural task for a submarine mission.

It should be even possible to usesonar to confirm ice depth.  Thus it wasprobably done.

Curiously, the total area appearsto be the lowest in 12 years while the actual mass has recovered to the highsof five years ago.  I would hate to do anerror calculation for the data involved.

Arctic Ice Recovering In Three Dimensions

Frank Lansner generated this graph from Navy PIPS data, showing thesignificant increase in Arctic ice thicknessover the last few years.

You can see why NSIDC only likes to talk about 5+ year old ice.(Because it hasn’t yet been five years since 2007.)

Navy Sea Ice Prediction Systems

Ruth H. Preller, Pamela G. Posey

Naval Research Laboratory • Stennis Space Center, Mississippi  USA
Wieslaw Maslowski, Donald Stark

Naval Postgraduate SchoolMonterey, California  USA
Thomas Thang C. Pham

Fleet Numerical Meteorology and OceanographyCenterMonterey, California  USA

The availability of real-time information on sea ice conditions in icecovered seas has always been important, not only to strategic militaryoperations, but to the economies of those countries that border the Arctic and itsmarginal seas. Knowledge of the thickness and movement of sea ice as well asthe locations of open water is required for traversing the Arcticwhether in a drill ship, in a cargo vessel or in an ice strengthened ship suchas a Coast Guard ice breaker.

Forecasting these conditions is a difficult task at best. The ice andsnow that cover the cold Arctic Ocean arehighly variable on short time scales, such as days to weeks, and longer timescales of years to decades. This variability in the sea ice cover is due to acombination of dynamic and thermodynamic effects.

Surface stresses on the top and bottom of the ice cause the movement ofsea ice, or ice drift, as well as the deformation of the ice.

Heating and cooling from the atmosphere and the ocean are largelyresponsible for the growth and decay of sea ice, in combination with the ice motion.

Sea ice has a strong seasonal variability. The thinnest sea ice andlargest amount of open water appear in the summer months from June toSeptember.

Ice begins to grow back in the fall and builds to a maximum thicknessin the late winter and early spring, March-April. Many of the marginal seas,such as the Barents and Greenland Seas are nearly ice freein the summer.

Other marginal seas like the Bering Sea and the Sea of Okhotskare completely ice free in the summer.

These seasonal patterns vary inter-annually as well. This variabilityis often represented by a seesaw effect when one part of the Arctic experiencesa “mild” ice year while another part of the Arctichas an increase inice extent and/or ice thickness. A general trend overlyingthis inter-annual variability has been seen by several scientists who haveexamined long records of satellite data. The trend is for an overall decreasein ice extent from the late 1970s through the mid-1990s (Parkinson et al.,1999; Cavalieri et al., 1997;
Johannessen et al., 1995). Whether or not this decrease will continueand what effect it may have on Arctic economics, operations and climatic changeis a topic of great international interest.

Sea ice forecasts usually focus on short time scales of 5–7 days. Inorder to provide an accurate prediction, forecast systems are most often basedon a combination of models and data. Modeling sea ice can be a difficultproblem, as it exists in many different forms
(Figure 1).

It can appear as a field of disjointed ice floes, as a level,continuous field of ice or as a landscape of ice hills and ridges. Interfacial stresses,from the atmosphere and the ocean, interacting with natural coastal boundariescause the ice to be in a nearly constant state of motion and deformationresulting in the formation of ice rubble, ridges, leads and ice floes. Most seaice models make the assumption that ice exists as a “continuum”. The ice is modeled as having some combination of viscous, plastic, or elasticmaterial characteristics. The way that the ice behaves as a material when actedupon by stress is called ice rheology. Ice within a model grid cell is definedas having a certain thickness and coverage, also defined as a percent of iceconcentration. Thickness and concentration are allowed to vary in timedepending upon the atmospheric and oceanographic conditions.

The Arctic presents a hostileenvironment for those trying to gather data on the atmosphere, ice or ocean.

Sub-zero temperatures, strong winds, limited daylight and someintimidating wildlife (Figure 2) make fieldwork in the Arctica difficult task.

A majority of observations of the Arctic ice cover,both in real-timeand over periods of decades, have been provided by satellite imagery anddrifting buoys.

Satellite data derived from the Defense Meteorological SatelliteProgram (DMSP) Operational Line-Scan (OLS) Sensor, the National Oceanic andAtmospheric Administration (NOAA) Advanced Very High Resolution Radiometer(AVHRR), the DMSP Special Sensor Microwave/Imager (SSM/I), and the Canadian RADARSATare used by forecasting centers like the U.S. Navy/NOAA National Ice Center(NIC) to provide a real-time picture of conditions in the Arctic. A network ofautomatic drifting data buoys monitoring surface air temperature, pressure andice motion, has existed in the Arctic basin since early 1979, first under theArctic Ocean Buoy Program and then in 1991, apart of the International ArcticBuoy Program (IABP).

These data have been used to support real-time operations in the Arctic as well as meteorological and oceanographicresearch of the Arctic basin. More information on the IABP is available at

Forecasts of ice conditions are produced by numerical ice-ocean modelsthat use these observations to help specify an “initial” state and then runforward in time. The length of the ice forecasts depends on the atmosphericforcing that drives the model. Usually, the forcing is derived from anatmospheric forecast model, and extends about seven days into the future.Longer forecasts (to 30 days) are sometimes generated using persistentatmospheric conditions. The main forecast products that describe sea iceconditions are ice thickness, ice concentration, and ice drift.

Real-time sea ice observations, analyses and forecasts are nowavailable from ice centers around the world (i.e. Canadian, Swedish, Danish,Japanese, Icelandic centers). These centers are usually responsible forproviding information on ice conditions closest to their own coastlines. The U.S. NIC isactively providing global sea ice information to Navy fleet operators andcommercial users. In addition to satellite observations, aerial icereconnaissance, ship/shore station observations, drifting buoy reports,meteorological guidance products, climatology and information frominternational partners, the NIC uses the forecasts from the Navy’s sea iceforecasting system, the Polar
Ice Prediction System (PIPS) version 2.0 to generate analyses and forecastsof sea ice conditions in the ArcticThe Polar Ice Prediction System (PIPS)

Since the late 1980s, the Naval Research Laboratory (NRL) has beendeveloping sea ice forecast systems for operational use by the U.S. Navy.These models run operationally on computer systems at the Fleet NumericalMeteorology and Oceanography Center (FNMOC). Productsare disseminated by FNMOC for use at the NIC. There have been several versionsof the Polar Ice Prediction System in the past that have led up to the present PIPS2.0. The first version of PIPS (PIPS 1.0), operational in 1987, covered theArctic basin, Barents and Greenland Seas using a horizontalgrid resolution of 127 km.

 PIPS 1.0 was followed by twohigher resolution regional models for the Barents Sea, theRegional Polar IcePrediction System—Barents (RPIPSB) and for the Greenland Sea, the RegionalPolar Ice Prediction System—Greenland Sea (RPIPS-G). The regional modelsdiffered from the PIPS 1.0 model in that they covered a smaller area and usedhigher grid resolution (25 km vs. 127 km). The higher resolution was necessaryto provide a better estimate of ice edge and coastlines. These forecast systemsmade use of an ocean climatology to provide ocean stresses and heat fluxes tothe ice model.

 Atmospheric stresses and fluxeswere provided to the ice model by the Navy Operational Global AtmosphericPrediction System (NOGAPS; see Rosmond et al. in this issue). Once these forecastsystems were in place, additional requirements were identified. High-resolutionforecasts were needed in all of the ice-covered regions of the Northern Hemisphereand sea ice models required coupling to an ocean model to provide daily oceanvariability. In response to these requirements, NRL developed the Polar IcePrediction System 2.0 (PIPS 2.0). This system became operational in 1996 andreplaced PIPS 1.0, RPIPS-B and RPIPS-G.

The core of the PIPS 2.0 is a coupled ice-ocean model that consists ofthe Hibler ice model (Hibler, 1979; 1980) and the Bryan and Cox ocean model(Cox, 1984). PIPS 2.0 coverage extends from the North Pole south, toapproximately 30°N latitude and includes
marginal seas such as the Sea of Okhotsk, the Sea of Japan and theYellow Sea in the Pacific and the Gulf of St. Lawrence and the Labrador Sea inthe Atlantic (Cheng and Preller, 1996). The grid resolution, 0.28°,varies from17–33 km depending on the location of the grid square within the sphericalcoordinate system (Figure 3). PIPS 2.0 uses a rotated coordinate system to avoidthe problem of a numerical singularity at the pole.

For both the ice and ocean models, the lateral boundaries are definedas solid walls and placed away from any sea ice covered regions in order tominimize their effect on sea ice forecast areas of interest.

Following the technique of Sarmiento and Bryan (1982), the ocean modeltemperature and salinity fields are loosely constrained to the Levitus (1982)climatological data set using a 250 day restoring time at all levels of theocean. This weak constraint on the ocean temperature and salinity does notadversely affect the PIPS 2.0 ability to forecast ocean variability on thedaily to weekly time scales. The bathymetry used by the ocean model is derivedfrom a U.S. Navy 5 minute data base called the Navy Digital Bathymetry DataBase 5' x 5' (DBDBV; NAVOCEANO, 1997).

In this coupled system, the ocean model provides input to the ice modelin terms of ocean currents, salinity and heat fluxes (temperatures). The icemodel provides salinity and temperature changes due to the growth and decay ofsea ice and surface stresses to the ocean model. The ocean model contains 15vertical levels, each level increasing in thickness with depth. Direct interactionbetween the ice and ocean model occurs in the first level that is 30 m deep andrepresents the ocean mixed layer.

Figure 4 is a schematic of the design of PIPS 2.0. At the center of thesystem, is the coupled Hibler ice model/Cox ocean model. Atmospheric forcing isprovided by NOGAPS. The atmospheric forcing fields used by the forecast systemare surface wind stress, surface air temperature, surface pressure, surfacevapor pressure, shortwave radiation, sensible plus latent heat flux, and thetotal heat flux. The system produces a 120- hour forecast each day. In additionto the output of the model (ice drift, thickness, concentration, growth/decayof ice, ocean currents, temperature and salinity), restart fields consisting ofthe model’s 24-hour forecast are written to a file to be used to initialize thenext day’s forecast. If the model restart fields are not available, amodel-derived climatology is used to initialize the forecast.

The quality of each forecast depends strongly on the accuracy of theinitial conditions and the atmospheric forecasts, as well as on the inherentaccuracy of the forecast model. If the unaltered 24-hour forecast is used toinitialize the forecast, these errors will inevitably compound eventuallyleading to unacceptably large forecast errors. To reduce this effect, the 24- hourforecast fields are used in combination with observed ice concentration data toinitialize the model.

The PIPS 2.0 assimilates ice concentration data from the DMSP SpecialSensor Microwave/Imager (SSM/I).

These data are chosen for three reasons: they are available daily, inreal time, at FNMOC; the resolution of the data is similar to the resolution ofthe model; and the data cover the entire model domain. The SSM/I brightnesstemperatures are converted into ice concentration data using the Navy CAL/VALalgorithm (Hollinger et al., 1991) at FNMOC. This algorithm is sensitive to theice/water boundary and to thin ice. As such it provides a good estimate of thelocation of the ice edge and a better estimate of thin ice than most otheralgorithms.

However, due to its sensitivity to thin ice, it often saturates tooquickly to 100% ice concentration, producing an overestimate of high iceconcentrations. For more information on algorithm descriptions and comparisonssee

After the PIPS 2.0 restart fields are read into the model, the dailySSM/I ice concentration data, interpolated to the model grid, are read in.Based on the characteristics of the CAL/VAL algorithm, the SSM/I data replacesthe model-derived ice concentration only at locations where observedconcentration is greater than 80% or less than 50% and, the difference between thetwo fields is greater than 10% or 5% respectively.

The model-derived ice thickness field and the ocean surface temperaturefield are then adjusted to be consistent with the concentration data. That is,if theSSM/I data indicate there is no ice where the model had produced ice, iceis removed from the model field and the ocean temperature is raised one degreeabove freezing to restrict immediate ice growth in this location. If the modeldid not have ice in a location where the SSM/I sees ice, a small amount of ice(0.3–0.6 m) is added to the model thickness field and the ocean temperature isset to the freezing temperature of sea ice.

These adjustment values were determined from a series of modelexperiments. As the adjustments are usually confined to a few grid cells nearthe ice edge, the experiments showed no serious dynamical/numerical problemsassociated with the technique.

Forecast Products

Since the end of July 1996, PIPS 2.0 has been producing operational120-hour forecasts daily. Each day a set of 11 products is generated from thePIPS 2.0 forecast (Table 1). These products, and the interval at which they aregenerated, are determined by the NIC. They are transmitted to the NIC in GRIBformat and may be viewed as a graphic product on NIC workstations. The NIC usesPIPS 2.0 products as guidance in generating their ice analysis and forecastproducts.

Figures 5b–d are examples of standard products from the system: icedisplacement (based on the ice drift), ice concentration and ice thickness.Figure 5a represents the observed ice motion derived from the available IABPdrifting buoys. Figures 5a and 5b show the qualitative similarities between thePIPS 2.0 forecast and the observed ice motion. PIPS 2.0 also has the capabilityto forecast ocean currents, ocean temperature and salinity. Recently thesefields were requested by the NIC as input to a marginal ice zone specific ice modelthat is being developed and tested for the center’s use.

Future Sea Ice Forecast Systems—PIPS 3.0

In the pre-operational evaluation of PIPS 2.0 products (Preller andPosey, 1995) the original PIPS and PIPS 2.0 ice drift were evaluated against Arctic buoy data. These results showed that PIPS 2.0produced an improved ice drift forecast over PIPS with a decrease in the RootMean Squared (RMS) Error of about 10–15%.

Due to the combination of the increased resolution and theincorporation of an ocean model versus an ocean climatology, the PIPS 2.0 iceedge was a substantial improvement over the existing PIPS models, including thehigher resolution regional models.

A recent study by a group of scientists from the NIC and NOAA (VanWoert et al., 2001), showed that although the PIPS 2.0 forecasts (48-hour) werebetter than persistence on average, there were still substantial biases in itsprediction of the growth and decay of sea ice in the marginal ice zone. PIPS2.0 often over-predicts the amount of ice in the Barents Sea and therefore often places the ice edge too far south. Incontrast, PIPS 2.0 often under-predicts the ice extent in the Labrador Sea and Hudson Bay. As forecasts are a combination ofmanydifferent components (Figure 4)—atmospheric forcing, oceanographic forcing,model parameterizations and initialization—the combination of even small errorsfrom each of these pieces could add up to a substantial error in the forecast.

In recent years, there has been a request for the capability to predictthe location of regions of open water within the ice pack. Approximately 1% ofthe central Arctic ice pack is estimated to be open water in winter. In summer,the amount increases to 10–20% (Gow and Tucker, 1990). Openings in the ice packusually appear in two forms: leads and polynyas. Leads are crack-like openingsin the ice pack, usually tens to hundreds of meters wide, while polynyas arelarger, more permanent openings. Leads form in regions where large-scaledivergent wind patterns produce divergent stresses in the ice causing the iceto break apart. Leads appear as linear features in compact ice that often have apreferential orientation (direction). Polynyas form near coastlines wheredivergence or shearing of the pack ice separates the ice from the coast or froma fast ice boundary. Leads and polynyas are significant because they providetraversable areas in the ice pack for ships. In addition they are importantbecause they are regions of very large heat-loss to the atmosphere. Winterleads rapidly refreeze, due to the heat loss to the atmosphere, forming youngthin ice that is easily deformed. Leads and polynyas are a major source of newice growth in winter. Although PIPS 2.0 can predict large-scale polynyas, itdoes not have the capability to produce smaller polynyas or leads or to provideguidance on lead orientation.

PIPS 2.0 is based on the scientific research efforts of late 1970s andearly 1980s. During the past 10 years (1990s), great strides have been made inunderstanding sea ice dynamics and thermodynamics as well as observing iceconditions. An additional and very important factor in the improvement of icemodeling and forecast capabilities is the advance in computer technology overthe past 10 years. Computer codes now make use of multiple processors and canperform more extensive computations in operationally acceptable time periods.

In 1998, the Office of Naval Research (ONR) and the Oceanographer ofthe Navy via the Space andNaval Warfare Systems Command (SPAWAR) joined forcesto fund an effort to combine this new technology and data into an improved seaice forecasting system.

This system, aptly named PIPS 3.0 ( is presently being developed through a joint effort among the U.S. Naval Postgraduate School,various other academic institutions and the Naval Research Laboratory. The PIPS3.0 ice model will contain an improved calculation of ice growth/decay based onthe use of a multi-level icethickness formulation. Ice edge forecasts will be improvedby using higher horizontal resolution. Ice motion and ice edge location will beimproved by the assimilation of satellite derived ice drift data. In addition,higher resolution and more realistic ice rheology will improve the PIPS abilityto predict areas of lead formation and lead orientation.

These improvements to the PIPS 2.0 are being tested in an incrementalfashion. The first tests were designed to examine the effects of higherresolution on an ice model similar to that of PIPS 2.0 coupled to an improvedocean model with improved bathymetry. At the same time, techniques for theassimilation of satellite derived ice drift into the ice model were developed andtested. Results from these tests are described below.

High Resolution Coupled Model Tests Similar to the PIPS 2.0, thecoupled ice-ocean model used in these experiments extends from the NorthPacific (at ~30°N) through the Arctic Ocean and the Nordic Seas into the NorthAtlantic (to ~45°N; Figure 6) and uses a rotated coordinate system. The ice andocean model horizontal grid is configured at 1
⁄12° (~9 km) with 45 vertical levels in the ocean. In the verticaldirection, the upper 100 meters of the ocean are resolved by eleven layers, andthe upper 500 meters of the ocean by nineteen layers. This resolution allows detailedrepresentation of bathymetry over the vast Arctic shelves and in marginal seasas well as many local bathymetric features important to large-scale oceancirculation (e.g. St. Anna Trough, Canadian Archipelago, or Bering Strait). Thebathymetry data used to define the grid shown in Figure 6 consists of the ETOPO5database (National Geophysical Data Center, 1988), NRL charts and Canadian Hydrographic Centercharts for latitudes south of 64°N. North of 64°N, the
2.5 km resolution International Bathymetric Chart of the Arctic Ocean (IBCAO) database (Jakobsson et al., 2000) isused. These data represent a considerable improvement over previous dataavailable for this area used in PIPS 2.0. This high resolution, couplediceocean model is an extension of previous modeling research using a horizontalresolution of 1⁄6° and 30 vertical levels (Maslowski et al., 2000, 2002; Zhanget al., 1999, The coupled model adapts the Los Alamos National Laboratory (LANL) global Parallel Ocean Program (POP) ocean model with afree surface (Maltrud et al., 1998; Dukowicz and Smith, 1994). The sea icemodel at present is similar to PIPS 2.0 with a viscous-plastic rheology, thezero-layer approximation of heat conductionthrough ice and a surface energy budgetcalculated following Parkinson and Washington (1979). The main differencebetween this ice model and the PIPS 2.0 ice
model is its ability to utilize modern parallel computer architectures.

A 27-year spin-up integration of the coupled iceocean model has beenrun, forced with a daily-averaged annual cycle of climatological atmosphericfields derived from the European Centre for Medium-range Weather Forecasts(ECMWF) 1979–1993 reanalysis.

Forcing fields include: 10-meter winds, surface pressure, surface airtemperature and dew point, and longwave and short-wave radiation. During thespinup, the ocean surface temperature and salinity are restored with a 30-daytime scale to monthly mean climatology derived from the Universityof Washington’s Polar Science Center HydrographicClimatology (PHC; Steele et al., 2001).

Following the 27-year integration, an additional 12-year run has beencompleted using the repeated 1979 ECMWF annual cycle for the first 6 years and 1979–1981inter-annual fields for the last 6 years. This approach has been used to forcethe sea ice and ocean states towards conditions of the late 1970s and early 1980s.These conditions may then be used to initializethe new version of the PIPS 2.0ice model and spin it up to the present day for forecast use.

Understanding the ice thickness distribution and its variability isimportant in both short-term operational forecasts and longer-term climatestudies.

Model-derived ice thickness fields are useful in obtaining alarge-scale, but detailed picture of ice distribution and, more importantly, anunderstanding of its temporal variability.

In Figure 7, the 3-year mean ice thickness distribution is presentedfrom the first three years of the 1979–1981 output. The modeled distribution ofice thickness compares reasonably well to that known from observations (Bourkeand Garrett, 1987; Bourke and McClaren, 1992). In agreement with data, thethickest ice (> 6.0 m) in the Arctic is found along the Canadian Archipelagoand northern coast of Greenland. Farther northnear the pole, the ice thickness decreases to values of 3.0–3.5 m. Thissimulation also shows relatively thick ice (> 3.5 m) on the East Siberianshelf. Since few observations are available for that region and that time, itis difficult to quantitatively verify these results. In the 1990s, measurementsof ice thickness were made in the central Arctic Ocean from U.S. Navy submarines and at a number of keypointlocations (e.g. the Fram Strait) using upwardlooking sonar. Model results will be verified against these recent observationsonce the model output for that time period becomes available. Based on thecomparisons made thus far, these model results look promising in terms offorecasting ice thickness and ice edge variability. Coupled ice-ocean models,such as those used by PIPS 2.0 and PIPS 3.0, are most often applied in academicstudies of global climate issues such as climate change. In these studies themodels are used for decadal or longer simulations and focus on long term changesin the overall pattern of ice thickness and ice extent. This application of thecoupled model requires access to powerful computers and large quantities of computertime. However, coupled ice-ocean models used for forecasting face a differentset of requirements. As forecasts are concerned with sea ice variability onmuch shorter time scales, the models must be designed to produce the mostaccurate daily changes in ice concentration, ice edge location and ice motion.

These models are required to go through rigorous validation studies toprove their capability to produce accurate short term variability. Dataassimilation plays a major role in the accuracy of these forecasts. Once operational,continuous quality control and evaluation of the products may be used to upgradethe system and improve forecast accuracy. In addition, forecast  systems are limited by the amount of computer resourcesavailable for each forecast as they compete with other forecast models eachday. Therefore the combined model/assimilation system must be designed to “fit”within these limitations. This places restrictions on the “size” or gridresolution of the models as well as the complexity of the modelparameterizations and the data assimilation techniques. Each of these issuesmust be taken into account when developing a new forecast systemPIPS 3.0 IceMotion Data Assimilation As stated earlier, a popular strategy to reduce the initialerror in numerical forecast models is to use information from real-timeobservations to correct the previous forecast field through data assimilation.Data assimilation analysis makes use of a predictor-corrector strategy, wherethe model predicted ice state (known as the background field) is corrected bywhatever observations are available.

The specific process of correction fundamentally determines the natureof the assimilation. Current ice state observations tend to be either scatteredthrough space or infrequent in time.

They may also contain noise that can introduce non-physical structureinto the model. The direct replacement of model values with the observed values(known as insertion) can have undesirable consequences. It is therefore,preferable to employ some sort of self-consistent procedure that reduces theerror and distributes the observational knowledge widely over the domain.Statistical assimilation methods meet these criteria.

One of the first investigations of the assimilation of ice motion datainto a stand-alone ice model (similar to the original PIPS 1.0) employs theoptimal interpolation methodology (Meier et al., 2000). Optimal interpolation isthe simplest type of statistical assimilation. Statistical assimilation meansthat the background field is corrected to “minimize the error statistics” ofthe assimilated field. This can be done, only if the error statistics of the observationsand the model are known. The technique is “optimal” since the error variance isminimized.

Individual ice velocity vectors of the background field are correctedby applying a linear combination of the nearby observed ice motions. Thecorrection coefficient for each observed value results from solving a linearsystem dependent, among other things, on the ratio of the model to data errorcovariances. When these weights are applied to the corresponding observations, theresulting correction minimizes the error variance of the corrected solution.This formula illustrates that it is not sufficient to be concerned with theobserved ice motions alone. The error statistics for the model and the data arejust as important. They instruct the algorithm as to which observations to use,and what relative
weight they must have, in order for the correction to minimize theerror. Two types of ice motion observations are used. The first, which providesthe data to be assimilated, is not
directly observable. It is derived from the SSM/I passive microwavebrightness temperature imagery through a technique known as feature tracking.

Individual swaths of the SSM/I satellite are combined or “composited”into a daily gridded field. The displacement of features common to the fieldson consecutive days can provide an estimate of the ice velocity.

The derived ice motion observations used here were obtained from thePolar Remote Sensing Group at the NASA Jet Propulsion Laboratory (Kwok et al.,1998).

They were chosen for the assimilation analysis because they are similarto the real-time ice motion products that will be available from FNMOC. Icemotion observations are available daily from the beginning of October throughthe end of May. The data are not available for the summer months due to thesusceptibility of the passive microwave signal to excessive error from moisturein the form of either heavy summer cloud cover or the formation of melt pondsin the ice.

The second set of observations comes from IABP drifting ice buoys(Rigor and Heiberg, 1997). In the operational configuration of the model, thebuoy motions will be included as observations to be assimilated. This will beespecially important in the summer months when SSM/I derived motions are notavailable.

The buoys were withheld in this study to be used for validationpurposes, and to compute the error covariances. For the period of the study,there are roughly 30 buoys, irregularly distributed throughout the Arctic, atany time. Each buoy transmits its locationtwice per day. The ice buoy velocity can then be calculated from the spatialdisplacement of the buoy location.

As a test of concept, this proposed PIPS 3.0 ice motion assimilationmethod was instituted in a previously tested high resolution fully coupledseaice/ocean model (Zhang et al., 1999). The coupled seaice/ocean model employsa reduced domain consisting of the Arctic basin, with a closed Bering Strait,and extending south into the Atlantic to about50°N. The spatial resolution of the model is roughly 18 km: half the resolutionproposed for the PIPS 3.0 model. The model can be run in either control (no assimilation)or assimilation mode.  Two model runswere generated for the period from January 1992 through December 1994. The ECMWFreanalysis product provided the atmosphericforcing in both configurations. Inthe control run, no assimilation takes place. In the assimilative run, the modeldigests the derived SSM/I ice motion product for the months of October throughMay of each year.

During the summer months, no assimilation takes place. The drifting icebuoy data are withheld as validation. Figure 8 shows selected trajectories ofthe IABP drifting buoys, the control run and the assimilative run.

Trajectories from the control run and the assimilative run are producedby integrating forward in time from the initial buoy location using a standardfourth-order Runge-Kutta method and the respective daily instantaneous velocityfield. The trajectories displayed represent the range of ice motion behaviorseen during the spring of 1992, and the period covering the fall of 1992 throughthe spring of 1993. The subset of trajectories displayed in the figure waschosen for its coverage ofthe figure domain. The trajectories for the controlrun and the assimilative run have the same starting locations, and the samestarting and ending dates as the buoys.

Consider the trajectories numbered 1–10. Both the control run (green)and the assimilative run (black) generated motions are consistently slower thanthe buoy motion (red). The pattern of buoy twists and turns is stronglyreflected in the assimilative run trajectories for about a half of the cases.The error, for the assimilative run trajectories, is largely due to the differencein speed. The error in the control run is manifested in terms of an evengreater speed differential as well as large errors in the direction of themotion. In nine out of the ten cases, the assimilative run produces atrajectory that more closely follows the buoy trajectory then does the controlrun’s trajectory. The assimilation process generally increases the ice speed,and preserves the observed direction of the ice motion. On average, thetrajectories with assimilation contain half the error of the trajectorieswithout assimilation.

The remaining two trajectories (11 and 12) illustrate a differentsituation. The marginal ice zone (MIZ)in the Greenland Seais extremely dynamic, commonly exhibiting the largest ice speeds of anywhere inthe domain. In addition, south of 80°N there are very few SSM/I derivedobservations that can be assimilated, and those that exist tend to have largeerror bounds. As a result, trajectories south of 80°N can have quite large errors.For trajectory 12, which starts east of Spitsbergen,the buoy moves clockwise, encircling the island. In contrast, the trajectoriesrepresenting the control run and the assimilative run speed south into the Norwegian Sea. The similarity between these two trajectoriesis due to the lack of SSM/I derived ice motions to assimilate. The differencein the path of the two trajectories reflects the difference between the ice statesfor that location in the two model runs. Since the current ice state for agiven location is a function of its history, each model can have ice withdifferent concentration and thickness characteristics. The different ice characteristicscan result in different reactions to the same forcing.

Trajectory 11 shows the buoy flowing south into the Fram Straitand along eastern Greenland. Initially, thethree trajectories are indistinguishable. As they approach Fram Strait,the trajectory from the control run begins to diverge from the path of theother two.

From the starting location to approximately 79°N, the assimilative runproduces a trajectory that coincides well with the buoy. Past this point, wherethere are no longer any SSM/I derived motions for the model to assimilate, thethree trajectories rapidly diverge.

Eventually, both the control run and the assimilative run producetrajectories that travel up to twice as far as the buoy, with the control run’strajectory traveling the farthest. Neither of the model runs, with or without dataassimilation, realistically represents the buoy motion south of the Fram Strait.These two cases illus-trate the need for the assimilation of the buoy motions inregions where the SSM/I derived motions are not available. In addition, thebuoy tracks can provide badly needed data during the summer season, when SSM/Iderived products are not available. Further details on the assimilative modelcan be found in Stark (2001).


The U.S.Navy has been running sea ice forecast systems operationally since the late1980s. These models have been gradually upgraded through the years to higherresolution, data assimilative models that span more of the sea ice coveredoceans in the Northern Hemisphere.

The models have provided forecasts of ice motion, ice thickness and icecoverage over regions that are physically difficult to observe due to hostileenvironmental conditions. Advancements in satellite technology provide theseforecast systems with continuously improving data for assimilation, thusenabling improvements to the daily forecasts.

 Advancements  in computer technology provide these forecastsystems larger and faster computers to generate improved forecasts.

The most recent upgrade to the Navy’s ice prediction capability is thedevelopment of the next generation forecast system, PIPS 3.0. Improvements tothis new forecast system include, higher horizontal resolution, a moresophisticated ocean model, improved data assimilation and perhaps mostimportant, an improved sea ice model. The ongoing upgrade of the sea ice modelwill include a Lagrangian formulation for calculating a multi-category icethickness distribution, a snow layer, a brine pocket parameterization,non-linear profiles of temperature and salinity (Bitz and Lipscomb, 1999), anda Coulombic yield curve for the viscous-plastic rheology (Hibler, 2000). Theseimprovements are geared towards providing better forecasts of ice edge, icemotion, ice thickness and regions of lead formation and lead orientation. ThePIPS 3.0 is presently going through its final development and will begin itsadaptation for operational use next year with a scheduled transition intooperational use in late 2002 or early 2003.


The authors appreciate the helpful comments from Dr. Mary AliceRennick, Mr. R. Michael Clancy and Dr.Michael Steele. The authors also thankMr. Ignatius Rigor for providing Figure 5a. This work has been funded throughthe Office of Naval Research’s Navy Ocean Modeling and Prediction Program(program element 602435), the Office of Naval Research’s High Latitude DynamicsProgram (program element 61153) and the Naval Space and Warfare Systems Command(program element 603207N). This paper, NRL contribution NRL/JA/7320/01/0019, isapproved for public release, distribution unlimited.


Bitz, C.M. and W.H. Lipscomb, 1999: An energyconserving thermodynamic model of sea ice. J. Geophys. Res.,104(C7),15669–15677.
Bourke, R.H. and R.P. Garrett, 1987: Sea ice thicknessdistribution in the Arctic Ocean. Cold Reg.Sci. Technol.,13, 259–280.
Bourke, R.H. and A.S. McClaren, 1992: Contour mapping ofArctic basin ice draft and roughness parameters. J.Geophys. Res., 97(C11),17715–17728.
Cavalieri, D.J., P. Gloersen, C.L. Parkinson, J.C.Comiso and H.J. Zwally, 1997: Observed Hemispheric asymmetry in global sea icechanges.  Science, 278,1104–1106.
Cheng, A. and R.H. Preller, 1996: The development ofan ice-ocean coupled model in the Northern Hemisphere.  NRL/FR/7322-95-9627, Naval Research Laboratory,Stennis Space Center, MS, 61 pp.
Cox, M., 1984: Aprimitive equation, 3-dimensionalmodel of the ocean.  Geophysical FluidDynamics Laboratory Ocean Group Technical Report, Princeton, NJ,1141 pp.
Dukowicz, J.K. and R.D. Smith, 1994: Implicitfree-surface method for the Bryan-Cox-Semtner ocean model.  J.Geophys. Res., 99(C4), 7991–8014.
Gow, A.J. and W.B.Tucker III, 1990: Sea ice in thepolar regions. In: Polar Oceanography Part A: Physical Science.
W.O. Smith Jr., Ed., Academic Press, NY, 47–122.
Hibler, W.D. III, 1979: A dynamic thermodynamic seaice model. J. Phys. Oceanogr., 9, 815–864.
Hibler, W.D. III, 1980: Modeling a variable thicknesssea ice cover. Monthly Weather Review, 108, 1943–1973.
Hibler, W.D. III, 2000: On modeling the anisotropicfailure and flow of flawed sea ice.  J.Geophys. Res., 105(C7), 17105–17120.
Hollinger, R.J., R. Lo, G. Poe, R. Savage and J.Pierce, 1991:Special Sensor Microwave/Imager calibration/Validation—FinalReport, Volume II. Naval Research Laboratory, Washington, DC,20 pp.
Jakobsson, M., N.Z. Cherkis, J. Woodward, R. Macnaband B. Coakley, 2000: A new grid of Arctic Bathymetry: A significant resourcefor scientists and mapmakers. EOS Transactions, American Geophysical Union, 81(9), 89, 93, 96.
Johannessen, O.M., M. Miles and E. Bjorgo, 1995: The Arctic’s shrinking sea ice. Nature, 376, 126–127.
Kwok, R., A. Schweiger, D. A. Rothrock, S. Pang and C.Kottmeier, 1998: Sea ice motion from satellite passive microwave imageryassessed with ERS SAR and buoy motions. J. Geophys. Res., 103(C4), 8191–8214.
Levitus, S., 1982: Climatological Atlas of the World Ocean.NOAA Prof. Pap., 13, 173 pp.
Maltrud, M.E., R.D. Smith, A.J. Semtner and R.C.Malone, 1998: Global eddy resolving ocean simulations driven by 1985–1995atmospheric winds.  J. Geophys. Res., 103(C13),30825–30853.
Maslowski, W., B. Newton, P. Schlosser, A. SemtnerandD. Martinson, 2000: Modeling recent climate variability in the Arctic Ocean. Geophys. Res. Lett., 27(22), 3743–3746.
Maslowski, W., D.C. Marble, W. Walczowski and A. Semtner,2002: On large scale shifts in Arctic Ocean
and sea-ice conditions during 1979-98.  Annals of Glaciology, in press.
Meier, W.N., J.A. Maslanik and C.W. Fowler, 2000:Error analysis of remotely sensed ice motion within an Arctic sea icemodel.  J. Geophys. Res., 105(C2), 3339–3356.
Naval Oceanographic Office (NAVOCEANO), 1997: DatabaseDescription for Digital Bathymetric Data Base— Variable Resolution (DBDB-V,)Version 1.0. Report, Naval Oceanographic Office, Stennis Space Center, MS.
National Geophysical Data Center, 1988:Digital relief of the Surface of the Earth. Data Announcement 88-NGG-02, NOAANational Geophysical Data Center Boulder CO.
Parkinson, C.L., D. Cavalieri, P. Gloersen, H.J.Zwally and J.C. Comiso, 1999: Arctic sea ice extents, areas and trends,1978–1996. J. Geophys. Res., 104, 20837–20856.
Parkinson, C.L. and W.M. Washington, 1979: Alarge-scale numerical model of sea ice. J. Geophys. Res., 84, 311–337.
Preller, R.H. and P.G. Posey, 1995: Validation testreport for a Navy sea ice forecast system: The Polar Ice Prediction System 2.0.NRL/FR/7322-95-9634, 31 pp.
Rigor, I.G. and A. Heiberg, 1997: International ArcticBuoy Program data report, 1 January 1995 √ 31 December 1995. Applied PhysicsLaboratory, University of Washington, Seattle, Washington.
Sarmiento, J.L. and K. Bryan, 1982: An ocean transportmodel for the North Atlantic.  J. Geophys. Res., 87, 395–408.
Stark D.R., 2001: Response of a high resolutioncoupled ice/ocean model to the assimilation of ice motion fields derived frommicrowave satellite imagery. Proceedings of the American Meteorology Society 6thConference on Polar Meteorology and Oceanography, May14–18.
Steele, M., R. Morley and W. Ermold, 2001: Polar Science Center Hydrographic Climatology (PHC) Aglobal ocean hydrography with a high quality Arctic Ocean.J. Clim., 14,2079–2087.
Van Woert, M.L., W.N. Meier, C.Z Zou, J.A. Beesley andP.D. Hovey, 2001: An assessment of the Polar Ice Prediction System iceconcentration fields during May 2000. Canadian Journal of Remote Sensing, inpress.
Zhang, Y., W. Maslowski and A.J. Semtner, 1999: Impactof mesoscale ocean currents on sea ice in high-resolution Arctic ice and oceansimulations,  J. Geophys. Res.,104(C8),18409–18429.

No comments:

Post a Comment