Progress in Physical Geography, Vol. 19, No. 2, pp. 216-242, 1995.

HYPERTEMPORAL ANALYSIS OF REMOTELY SENSED SEA ICE DATAFOR CLIMATE CHANGE STUDIES

Joseph M. PiwowarEllsworth F. LeDrew

Department of Geography and Earth Observations LaboratoryInstitute for Space and Terrestrial Science

University of WaterlooWaterloo, ON N2L 3G1

Canada

tel.: (519)885-1211 ext. 6563fax: (519)888-6768

internet: piwowar@wateol.uwaterloo.ca

ABSTRACT

Climatologists have been concerned for some time about the possibleclimatic ramifications of increased atmospheric concentrations o fanthropogenically produced greenhouse gases. Many general circulat ionmodel experiments show amplified warming in the polar regions as t h estrongest response to enhanced atmospheric greenhouse gasconcentrations. In light of this, there has been speculation that a spatiallycoherent pattern of high-latitude temperature trends could be an earlyindicator of climatic change. The sensitivity of sea ice to the t empera tu reof the overlying air implies that observed trends in Arctic ice condi t ionsmay also indicate general climatic changes. For example, changes in t h eglobal average temperature might be detectable by observing variations i nsea ice extent. Extent, then, is one common climatic indicator obtainablefrom sea ice. Other indicators include ice type, concentration, thickness,and dynamics. These are shown to have distinct regional and spatialdistributions within the Arctic.

The advent of satellite remote sensing has offered intriguing opportuni t iesto investigate, in good spatial and temporal detail, even the most r e m o t eor rarely visited areas of the earth's surface. The large extent of sea ice i npolar regions, the difficulty and expense of access, and the normallyadverse weather conditions make satellite data almost indispensable f o rstudies requiring global or large-scale ice cover characteristics in theseregions. Remote sensing in the microwave portion of the electromagneticspectrum is particularly relevant for polar applications because

microwaves are capable of penetrating the atmosphere under virtually allconditions (of particular significance during the frequent and extendedperiods of cloudiness in the Arctic) and microwave remote sensingsystems are not dependent on the sun as a source of illumination ( a nimportant consideration during the long polar night). In part icular ,analyses of passive microwave imagery can provide us with dailyinformation on sea ice extent, type, concentration, dynamics, and me l tonset . An historical record of Arctic imagery from orbiting passivemicrowave sensors starting from 1973 provides us with an excellent d a t asource for climate change studies.

The development of analysis tools to support large area monitoring isintegral to advancing global change research. Traditionally, maps haveshown phenomena distributed over space at some point in time. Explicitrepresentation of time in cartographic displays has either taken the f o r mof change maps or been expressed by a series of static maps. The criticalneed is to create techniques which highlight the space-time relationshipsin the data rather than simply displaying voluminous quantities of da ta .In particular, hypertemporal image analysis techniques are required t ohelp find anticipated trends and to discover unexpected t empora lrelationships. Direct hypertemporal classification, principal componen t sanalysis, and spatial time series analysis are identified as three pr imarytechniques for enhancing change in temporal image sequences. There is aneed for the development of new tools for spatial-temporal modelling,however.

1. INTRODUCTION

The earth's polar regions play a critical role in global change scenarios.Polar climates exist through a delicate balance between the l imitedamount of radiant energy received from solar radiation and generalatmospheric and oceanic circulation, and the high proportion of that h e a twhich is reflected back into space or otherwise tied up with in changes o fphase of surface moisture. The net result is a permanent presence of iceand snow. In areal extent, sea ice accounts for nearly two-thirds of t h eearth's ice cover. In the north polar region, it spans an area ranging f r o ma summer minimum of 8.5 x 106 km 2 to almost twice that extent i nwinter .

Climatologists have been concerned for some time about the possibleclimatic ramifications of increased atmospheric concentrations o fanthropogenically produced pollutants. At the 1987 WorldMeteorological Organization / United Nations Environmental Programme /

International Council of Scientific Unions workshop on climate changeheld in Villach, Austria a scenario was developed which suggests that d u eto progressive accumulation of greenhouse gases (mainly carbon dioxideand methane) in the atmosphere, the global average surface t empera tu rewould rise between 1.5°C and 4.5°C above the mean 1980s decadaltemperature by about the year 2030. In high northern latitudes, t h eaverage rise in winter temperature is expected to be more than twice t h eglobal average. Thus, if the median Villach scenario of a globaltemperature rise of 3°C by 2030 were taken, one could expect, on t h ebasis of current knowledge and modelling, that the polar regions would b eabout 6-7°C warmer than at present (Roots, 1989).

Although the predictions of the Villach scenario are substantiated b yother working groups, such as the International Panel on Climate Change(Houghton, 1990) and the Canadian Climate Centre (Daley, 1989), not allscientists are in agreement (Mysak, 1990; Kukla, 1993). Washington a n dMeehl (1989) report on a study where an atmospheric model was coupledto a four layer ocean model to investigate the response of the cl imatesystem to increases of greenhouse gases. Their results show the o c e a nsurface warming only near 30-50°N latitude and a cooling of the n o r t h e r nNorth Atlantic resulting in greater ice cover in the Greenland Sea.

A common hypothesis is that the most direct effect of global warming o nthe climate of the Arctic would be manifested in changes of the ice cover(Hare, 1982). Recent studies on the climatic role of sea ice do show t h eestablishment of a bi-directional feedback mechanism between the iceand the adjacent atmospheric and oceanic systems (Walsh and Johnson,1979a; Parkinson, 1989; LeDrew et al., 1991). That is, sea ice has asignificant influence on the atmosphere and adjacent oceans and in turn issignificantly influenced by the atmosphere and oceans. There is now littledoubt about the intimate connection between sea ice and climate a n dsince the atmosphere, oceans, and ice cover are all components of asingle thermodynamic system, a change in any one part necessarily resul tsin compensating changes in the others (Maykut and Untersteiner, 1971 ;LeDrew, 1992a).

Our research in this area is built upon the assumption that, due to the bi-directional feedbacks operating in the polar basin, observed trends i nArctic sea ice conditions may be indicative of global climatic changes.What is not clear, however, is what sort of changes we should b eexpecting to find in the sea ice cover and whether the ice is a leading o rtrailing indicator of global change.

Recent advances in two related technologies may hold the key to helpingus detect and monitor changes in global climates. First, we note t h a trepetitive, continuous, and global imaging of the earth's surface beganearly in the 1970s with the launch of the Nimbus, NOAA, and Landsatseries of remote sensing satellites. Thus, the length of the historicalrecord of global remote sensing data is quickly approaching the thir ty-year mark usually deemed necessary for the establishment of climaticnorms. Secondly, new techniques are being developed to analyze thesedata. It is not sufficient to simply apply existing time series and t r e n danalyses to long sequences of remote sensing imagery because these havenot been developed to capture the spatial relationships inherent in t h edata. Research has begun into a new type of image analysis technique,which we call hypertemporal image analysis, to work with these data t oextract temporal signals in a spatially coherent manner. It is the couplingof these new hypertemporal image analysis techniques with t h eincreasingly long remote sensing record which holds the promise o fsignificantly advancing our understanding of the changes now happeningin the global climate system.

Although much of the discussion about hypertemporal image analysis o fthe historical remote sensing archive is generic to almost any globallocation, we focus our attention on the north polar region because of o u rinterests in the sea ice-climate linkages. In the balance of this paper t hen ,we summarize the properties of the Arctic sea ice cover which a r esignificant for climate change studies in light of the data available in t h eremote sensing record. We then explore the potential of hyper tempora limage analysis techniques for extracting these parameters.

2. ARCTIC SEA ICE AND CLIMATE CHANGE

The climatic role which sea ice plays is generally expected to be based o nchanges in its surface albedo, insulative properties, thermal capaci tance,and brine capacitance (Walsh, 1983; Anderson, 1987; Parkinson, 1 9 8 9 ) .Table 1 lists the physical properties of sea ice which can be used t oestimate each of these climatic properties on a global (or at leasthemispheric) basis. Any analysis of climate change and variability usingsea ice as a proxy indicator must ultimately reference one or more o fthese physical properties. In this section we assess the informat ionpotential of sea ice extent, type, concentration, thickness, mo t iondynamics, and seasonality for climate monitoring.

2.1. SEA ICE EXTENT

The polar regions tend to amplify climatic variations occurring elsewhere,thus relatively minor climatic changes are expected to produce largevariations in the areal extent of the ice cover. Since the existence of alarge thermal gradient at the ice-ocean boundary significantly affects t h elarge-scale transfer of energy between the atmosphere and oceans ,continued monitoring of the extent of the floating ice pack is crucial t oour understanding of the present climatic state.

Areal extent is the most easily, hence most frequently, mapped pa rame te rof the Arctic sea ice. Arctic sea ice has an average seasonal cycle rangingfrom a summer minimum of 8.5 x 106 km 2 in September to a wintermaximum of 15 x 106 km2 in March (Figure 1). The historical record as awhole does not show any definitive indications of consistent growth o rdecay of the north polar ice extent, although there is considerableinterannual, regional, and seasonal variation (Gloersen and Campbell,1988; Parkinson and Cavalieri, 1989). In a seasonal analysis, however,Chapman and Walsh (1992) have found statistically significant decreasesof the summer ice extent and three new summer minima between 1 9 7 6and 1990. This is in contrast to the 1960s where five of the largestsummer extent values are found.

2.2. SEA ICE CONCENTRATION/OPEN WATER AREA

Although it is generally very compact, sea ice never completely covers t h eentire ocean surface because of atmospheric and oceanic forcing.Expansion and deformation within the ice pack can cause the formation o fopen water areas known as "leads" (long, narrow cracks ranging f r o mmetres to kilometres in width) and "polynyas" (more expansive regions o fopen water up to several thousands of square kilometres in size). Sea iceconcentration - the percentage of a given area of ocean covered by ice - isan often used measure of the amount of open water within the icemargins. Even at high ice concentrations, an open lead 10-20 m acrosscan have a dramatic impact on the air-ice-ocean climate system. Ice ac t sas a thermal regulator, controlling the exchange of energy between t h eocean and the atmosphere. Whenever the relatively warm ocean isexposed directly to the cold winds blowing over the ice surface, the verystrong temperature contrasts establish a massive energy flux. The o p e nwater areas thus act as "thermal volcanoes" (LeDrew, 1992a).

Figure 2 shows that even at the time of maximum ice extent there a r elarge portions of the polar ice pack which have ice concentrations of less

than 100%. An annual examination of similar images may p r o d u c eevidence of climate change through significant differences in iceconcentrations both regionally and/or seasonally.

2.3. SEA ICE TYPE

Sea ice is a complex material formed by the freezing of sea water. Manyof the geophysical properties of sea ice are largely determined by its age.Of significance to climate research is the metamorphosis which occurs i nsea ice as it survives the first summer melt season. Before this t ransi t ion,so called "first-year" ice tends to be relatively thin, attaining a maximumthickness of about 2 m (Weeks, 1976). Ice which has survived at leastone summer melt period - termed "multiyear" ice - is substantiallyaltered from first-year ice, because water from melt ponds percola testhrough the ice causing desalination of the surface and underlying layers(Parkinson et al., 1987b). The low brine content of multiyear ice impar t sa particularly high strength to an already thickening ice sheet.

Sea ice is, therefore, generally classed by age: thinner, weaker first-yearice; and thicker, stronger multiyear ice. The maximum age of Arctic s e aice is estimated to be nineteen years old, with only about 2% of the a r e aoccupied by this group (Vowinckel and Orvig, 1970). Changes in t h erelative distributions of first-year and multiyear ice across the Arctic a r epotential indicators of climatic variability although no significant findingshave been reported to date.

2.4. SEA ICE THICKNESS

Pack ice reaches its final equilibrium thickness of 2-4 m in about fiveyears when the amount of ice lost during summer melt is made up dur ingthe following winter (Vowinckel and Orvig, 1970; Maykut a n dUntersteiner, 1971). Although local bathymetry, tides, and cu r ren t saffect the growth of landfast ice, the relative thickness of sea ice in aparticular year reflects average winter and early spring temperatures a swell as the magnitude of mid-summer solar radiation (Jacobs and Newell,1979; Ledley, 1990). Thus, sustained decreases in the average thicknessof sea ice on a regional scale may well be indicative of a warmer climate.

2.5. SEA ICE DYNAMICS

Much of the Arctic sea ice exists in a geographically cons t ra inedenvironment. The Arctic Basin holds a small ocean which is su r roundedby two large continents (Figure 2). The Arctic Ocean has res t r ic tedinteraction with the world's ocean system, the principal access po in tbeing through Fram Strait, a deep trench between Greenland and Svalbard.The connection of the Arctic Basin with the Pacific Ocean through t h eBering Strait is of lesser consequence because its shallow bench inhibitsthe passage of deep ocean currents (Clarke and Jones, 1989).

The general drift pattern of the ice in the Arctic Ocean follows t w odominant features: the Beaufort Sea Gyre and the Transpolar DriftStream. The oldest ice in the Arctic circulates in a generally c losedclockwise drift stream in the Beaufort Sea/Arctic Ocean and is control ledby the Arctic High atmospheric pressure system. The Transpolar DriftStream, on the other hand, stretches from the East Siberian Sea across t h eNorth Pole to the northeast of Greenland where it exits through FramStrait. The Transpolar Drift Stream is the major carrier of ice out of t h eArctic Basin. Mean annual net drift rates vary from 0.4 to 4.8 k m / d a ywith monthly averages observed as high as 10.7 km/day (Weeks, 1976).

Recent investigations of ice motion using satellite data have revealed t h a tthe mean drift rate of the ice pack is quite variable and can, at times b ereversed. In particular, the predominantly clockwise circulation of t h eBeaufort Gyre has been seen to reverse for periods lasting at least 30 daysin late summer (Serreze et al., 1989). LeDrew et al. (1991) show th isphenomenon to be triggered and maintained by persistent cyclonicactivity during that time of year. It is clear a reversal of this nature m u s thave a direct impact on the climate system. LeDrew et al. (1991) suggestthat there is some evidence of ice divergence during the reversal episodesexposing large amounts of relatively warm ocean to the cold a tmosphere .Hypertemporal remotely sensed imagery can play a key role in moni tor ingthe motion of the ice pack and perhaps reveal other anomalous dr i f tphenomena. In addition, tracking of ice masses across image sequencesmay help verify the relative ages of sea ice in different regions of t h eArctic.

2.6. SEASONALITY

In the Northern Hemisphere, the seasonal cycle of ice advance and r e t r ea tis approximately symmetrical in all regions, although the periods of m o s trapid growth and decay do vary regionally (Comiso and Zwally, 1984). In

the late autumn, sea ice morphology continues to change until the surfaceair temperature drops below about -10°C in November or December, a twhich point the ice stabilizes and remains so until the air t empera tu rerises back above that level in the spring. Comiso (1986) and Chapmanand Walsh (1992) did not find any significant trends of winter ice ex tentin the Arctic between 1961 and 1990.

The variations in sea ice are particularly significant during the me l tseason, when rapid changes in ice extent and surface conditions occur. Inan examination of the onset of melt across the Arctic Basin in 1979 a n d1980, Anderson (1987) found the melt signature is observable first in t h eChukchi Sea and the Kara and Barents Seas. As melting progresses, t h emelt signature moves westward from the Chukchi Sea and eastward f r o mthe Kara and Barents Seas to the Laptev Sea region. Anderson (1987) a lsonotes that the initial location and date of the melt signal varies with year .This leads him to conclude that monitoring the occurrence of me l tsignatures could be used as an indicator of climate variability in t h eArctic's seasonal ice zones.

The significant processes of the summer Arctic ocean are the loss of t h esnow cover, the opening up of extensive areas of ocean, and t h eformation of melt ponds on the existing ice surfaces. In the context o fclimatological monitoring, the effects these changes have on the surfacealbedo and the interannual variation of albedo are of prime impor tance(Carsey, 1985). Global warming scenarios generally project a more r ap idretreat of Arctic sea ice in summer than in winter. Chapman and Walsh(1992) report finding statistically significant decreases of the summer iceextent of Arctic ice and three new summer minima in the past f if teenyears. When they compare this decrease with the concomitant fact t h a tthe strongest atmospheric warming has occurred over those land a reas(northwestern North America, northern Asia) and seasons (late winter ,spring) when the albedo-temperature feedback is potentially s t rongest ,they conclude that northern land areas as well as the summer sea iceregime should be closely monitored for greenhouse signals (Chapman a n dWalsh, 1992).

Crane (1978) observed an "early" and "late" mode of ice retreat in t h espring seasons from 1964-1974. He also noted a similar pattern in t h eautumn period of ice advance. When he related this pattern to t h egeneral synoptic circulation over the area he found that for the years o fearly ice advance, there is an increased frequency of northerly a n dwesterly air flow, bringing lower temperatures and an influx of mult iyearice into the Davis Strait-Labrador Sea area (Crane, 1978). Ice-atmosphere

feedbacks are thought to operate in both directions during the ice-growthregime (Barry, 1983). Not only do the cooler air temperatures precipi ta teice formation, but the strongest correlations between ice anomalies a n dsubsequent atmospheric fluctuations are found in the autumn m o n t h s(Walsh and Johnson, 1979a). This bi-directional coupling has a tendencyto increase the variability within the ice pack and to mask any potent ia lclimatic signals. Thus, the autumn is a less desirable season f o rmonitoring the sea ice than the spring or summer.

2.7. SEA ICE AS A CLIMATIC INDICATOR

Table 2 summarizes what we currently know about the possible climaticinfluences of sea ice extent, type, concentration, thickness, mo t iondynamics as well as an estimation of their ideal observationalrequirements. Analyses of these parameters are compounded by the fac tthat the oceanic, atmospheric, and geographic composition of the Arcticcombine to create distinct regional variations in sea ice. For example,there is only a 20% seasonal difference in sea ice extent in the Greenlandand Norwegian Seas while Baffin and Hudson Bays experience summer -winter variations of over 200% (Comiso and Zwally, 1984). This m e a n scertain regions may exhibit climatic signals with more clarity; regionsexhibiting high interannual variability would not be good candidates f o rclimate change studies since a particularly large change from average s e aice conditions would have to occur before a climate change cou ldconfidently be said to have taken place (Zwally et al., 1983b; Parkinson,1991). High variability is the noise in which a climatic signal must b efound .

3. REMOTE SENSING OF SEA ICE

Remote sensing has a particularly prominent role in the monitoring o fclimate change; it is the only source of data with which we can view t h eentire planet and monitor the change in the nature of the surface of t h eplanet through time, in a consistent, integrated, synoptic and numericalmanner (Davis et al., 1991; LeDrew, 1992b). Global observations of t h eearth by satellite sensors are required simply because no other app roachpermits measurements to be made at the required spatial and t empora lresolutions. This is particularly relevant in Arctic research where t h elarge extent of the sea ice, the difficulty and expense of access, and t h enormally adverse weather conditions make satellite data indispensable f o rstudies requiring global or large-scale ice cover characteristics in theseregions (Comiso and Sullivan, 1986; CGCP, 1989).

Many of the changes in the global environment may be monitored on aworldwide basis with remote sensing satellite systems. This moni tor ingcan be direct, through such observations as surface temperature increase,or indirect, through the observations of parameters thought to br ingabout changes in climate or to respond to the changing climate, such a schanges in polar ice extent (CGCP, 1989). Little quantitative informat ionexisted on seasonal or interannual changes of the polar oceans' sea icecover until repetitive and synoptic satellite observations began. Theability of earth observing satellites to collect data from the polar regionsat a variety of scales for extended periods of time have been of significantvalue in our understanding of ocean-ice-atmosphere climate interact ionprocesses. These data provide information on current climatic condi t ionsin polar regions as well as information useful in studies of the t ime-dependent behaviour of sea ice. In short, satellite data have he lpedscientists see the ocean and ice systems as an integral part of a complexglobal climate system (Johnson, 1989).

Of particular relevance to remote sensing in polar environments are t h erestrictions imposed by the limited times of direct sunlight and a c learatmosphere. Further, for any variable to be useful in climate changestudies, it must be routinely measured over a period of at least 1 0 - 2 0years. In the following review of remote sensing of sea ice, optical ,thermal and microwave sensors are evaluated against these criteria i naddition to assessing their potential to monitor the specific sea iceparameters highlighted above.

3.1. OPTICAL AND THERMAL SENSORS

Data from the AVHRR sensor carried by the NOAA satellite series havebeen the most widely used in climate change studies in general, and haveparticular relevance for polar remote sensing. These data have hightemporal resolution (at least two images can be acquired daily for anylocation north of 60° latitude), medium spatial resolution (1.1 k m :detailed enough for monitoring individual ice floes while minimizingunnecessary data volume), thermal band imaging capabilities (which allowthe collection of data through the long polar night), and inexpensiveacquisition costs. Results from ice mapping experiments show that iceconcentrations determined from optical imagery differ from thosederived from passive microwave data by ±15-20%, probably due to t h epresence of thin new ice not visible in the optical imagery and the abilityof the microwave sensors to measure partial concentrations within t h epack more accurately (Dey, 1981; Comiso and Zwally, 1982; Alfultis a n dMartin, 1987; Burns and Viehoff, 1991; Emery et al., 1991).

Remote sensing at optical and thermal wavelengths is limited for longterm climatic monitoring of the northern cryosphere, however, becauseof the inability to collect data during the long polar night of winter (wi ththe exception of bands located at thermal infrared wavelengths) a n dthrough an optically opaque atmosphere (Arctic cloudiness averages 7 0 -80% from June through August). In addition, since these optical sensorswere not originally designed around polar applications, their response isfrequently saturated by bright ice and snow returns, inhibiting detai ledsurface discrimination. Few Arctic images from these platforms havebeen archived and some of the early data tapes which did exist are n o wdeteriorating, further reducing the availability of historical data for use i nlong term climate monitoring studies (Massom, 1991).

3.2. MICROWAVE SENSORS

The planetary and atmospheric restrictions imposed on optical a n dthermal sensors in the polar regions are removed when radiation isdetected in the microwave portion of the electromagnetic spectrum (i.e.,at wavelengths between 1 mm and 1 m). Microwave sensors de t ec tthermal microwave radiation reflected and emitted from the ea r th ' ssurface. Additional radiative contributions or interference introduced b yclouds, other atmospheric effects, and extraterrestrial sources exist b u tare relatively minor in the polar regions (Parkinson et al., 1987b). Twoadvantageous features characterize microwave energy from a po la rremote sensing standpoint: (i) microwaves are capable of penetrating t h eatmosphere under virtually all conditions (of particular significanceduring the frequent and extended periods of cloudiness in the Arctic); a n d(ii) microwave remote sensing systems are not dependent on the sun as asource of illumination (an important consideration during the po la rnight) (Lillesand and Kiefer, 1987; Johnson, 1989). Remote sensorsoperating in the microwave portion of the spectrum are typically dividedinto two types: those which provide their own source of illumination,called "active" systems; and those which only sense naturally occurr ingradiation, called "passive" systems.

So called "active" microwave sensors derive their designation from t h efact that, unlike most other remote sensing systems, they send out the i rown energy to illuminate the surface for imaging. The main activemicrowave system in use today is called the synthetic aperture radar, o rSAR. SAR is arguably the most powerful remote sensor available f o rcryospheric research due to its high resolution and all-weather, all-seasonimaging capabilities (Massom, 1991). It is not without its limitations,

however. While SAR data are quite powerful for the discrimination o fvarious ice types under ideal conditions (i.e., in the middle of winter), t h eenergy-matter interactions in a melting snow/ice surface are still poor lyunderstood (Barber et al., 1992a). In addition, spaceborne SARs ( f o rexample ERS-1, JERS-1, RADARSAT, and EOS) have (or will have) relativelynarrow swath widths (in the range of 100 km) which are unsuitable f o rhemispheric climate analyses. Continuous, repetitive SAR remote sensingof the polar regions has only just begun with the launch of the ERS-1satellite in 1991. While the future of active microwave remote sensinglooks bright (at least three other spaceborne SARs are to be launchedbefore the end of the century), the lack of an historical image archivefrom which climatic variations can be derived is a limiting factor to t h epresent use of SAR data for climate change studies.

Another type of microwave remote sensing system which exhibits t h efavourable characteristics of all-weather and day-night imagingcapabilities, and for which a reasonable historical archive of Arctic d a t aexists, is that of passive microwave. As its name suggests, a passivemicrowave radiometer detects microwavelength radiation incident u p o nits antenna without transmitting any illuminating signals by itself. Thismicrowave energy typically originates as naturally emitted radiation f r o mthe earth's surface and to a lesser extent, the atmosphere.

Three passive microwave sensors have been placed into earth orbit s ince1972: the Electrically Scanning Microwave Radiometer (ESMR)(operational lifespan: December 1972 - October 1976, with some d a t agaps), the Scanning Multichannel Microwave Radiometer (SMMR)(operational lifespan: October 1978 - August 1987), and the SpecialSensor Microwave / Imager (SSM/I) (operational lifespan: June 1987 -present). With varying degrees of success, data from all three can be u s e dto derive ice extent, type, concentration, motion dynamics, and the onse tof melt.

The ESMR was the first instrument to provide the research communi tywith the repeated, synoptic overviews of the polar ice masses necessaryfor monitoring ice extents. It was only a single-channel system, however,which inhibits the derivation of ice types solely from the ESMR data a lone(Parkinson et al., 1987b). Ice concentrations calculated from ESMRimagery are estimated to have an accuracy range of 15%-25% (Parkinson,1989; Massom; 1991).

The use of single frequency data from the ESMR sensor creates ser iousambiguities in data interpretation, particularly in areas of he terogeneous

ice cover such as the extensive marginal ice zones of the Arctic. Anattempt was made to address these problems by designing the secondgeneration of polar orbiting passive microwave sensors, SMMR, to haveten channels. The availability of passive microwave data at a variety o ffrequencies and dual polarizations permit more accurate determination o fsea ice characteristics.

Based on the knowledge gained from the ESMR and SMMR systems, t h ethird generation of polar orbiting passive microwave radiometers, SSM/I,operates at frequencies which are further optimized for the measuremen tof sea ice. Error analyses of ice parameters derived from SSM/I imageryusing an algorithm developed by researchers at NASA show g o o dagreement (less than 5% difference in autumn through spring; near 10%in summer) with supplementary data sources (principally optical r e m o t esensing imagery). The analysis technique, however, has been found t ounderestimate ice concentration in areas of close pack ice a n doverestimate ice concentrations in areas of open pack ice (Bjerkelund e tal., 1990; Cavalieri et al., 1991; Steffen and Schweiger, 1991). Analternate algorithm has been developed jointly by the CanadianAtmospheric Environment Service and the Institute for Space a n dTerrestrial Science ("AES/ISTS algorithm") which incorporates wea therand sea state corrections to aid in the retrieval of ice parameters (LeDrewet al., 1992c). In a comparative analysis between the two techniques,Bjerkelund et al. (1990) found the AES/ISTS algorithm provided m o r eaccurate estimates of ice concentration, type, and extent than the NASATeam algorithm.

Spaceborne passive microwave sensors are not without limitation,however. Because of the long integration time required at the sensor t oreceive enough microwave energy to create an acceptable signal, passivemicrowave images have very low spatial resolutions - commonly in t h erange 30 km to 70 km. This inhibits their application to all b u themispheric-scale analyses. In the context of climate change this mightnot be such a bad restriction, however. Since passive microwave imagestend to be reasonably small it is possible to include many more t empora linstances in a long term study than it would be if the images were larger.

3.3. PASSIVE MICROWAVE DATA FOR SEA ICE MONITORING

In section 2, six sea ice climatic indicator parameters were evaluated. InTable 3, the potential of passive microwave remote sensing to providedata for the measurement of these parameters is assessed. The onlyparameter not directly retrievable from multichannel passive microwave

imagery is ice thickness, although broad generalizations can be infer redfrom ice type (older ice tends to be thicker). Past and present passivemicrowave sensors cannot meet the ideal spatial resolutions proposed f o reach parameter in Table 3. The sensors' coarse spatial resolut ionsproduces a noticeable "land contamination" effect where the sensor ' sfield of view contains enough land (which generally has brightnesstemperatures closer to those of sea ice than to those of open water) t ogive anomolously high ice concentration calculations (Parkinson, 1 9 9 1 ) .This can have a particularly large effect in the Canadian Archipelago,which may limit the usefulness of satellite passive microwave data in th isregion. The spatial averaging inherent to passive microwave imagery m a ywell be useful, however, for reducing the effects of localized anomalieswhich could obscure any true climatic signals on broader regional a n dhemispheric scales.

A finer discrimination of ice types than first-year/multiyear is a desirableparameter not yet retrievable from passive microwave observations. Inaddition, the extraction of accurate ice information during seasonaltransitions is very limited. Research should be initiated to developmeasurement techniques for microwave observations of snow cover ,surface albedo, ice production, and atmospheric water during the sea iceseasonal cycle (Carsey, 1984).

These limitations do not negate the usefulness of passive microwaveimagery, however, their potential implications must be acknowledged i nany long-term climatic studies.

4. HYPERTEMPORAL IMAGE ANALYSIS

Thus far in this report, we have reviewed the need for acquiring accura tehistorical information on Arctic sea ice for climate change studies a n dhave established that remote sensing imagery from passive microwavesensors is the only data source available to provide us with th isinformation. We now focus our attention on the tools required to ex t rac tthe change information from the remote sensing data.

Time is the fourth dimension; it is unlike the first three (spat ial)dimensions in that it is asymmetrical (it flows in only one direction) a n dis difficult to envision, much less comprehend (we see the effects o ftime's passage, but we do not perceive it directly) (Saab a n dHaythornthwaite, 1990). Because global change is directly related t otime, the ability to perform multitemporal analyses is vital to globalchange research. Thus, multitemporal functionality should be cons idered

a fundamental enhancement to all spatial analysis systems (Stutheit ,1 9 9 1 ) .

Although satellite remote sensing has long been advocated for moni tor ingsurface processes through time, there has been little progress to date i nspatiotemporal analysis of multitemporal imagery largely because of alack of repetitive and inexpensive data and because of a lack of adequa teprocessing procedures (Davis et al., 1991). The basic premise i ntemporal monitoring using remote sensing data is that changes in surfacefeatures must result in changes in sensed radiance values and that t h echanges in radiance due to surface changes must be large with respect t oradiance changes caused by other factors (e.g., atmospheric condi t ionsand sun angle) (Singh, 1989). Most temporal analyses reported to d a t eexamine the changes between an image pair; between time t and time t+1.We distinguish such two or three period comparisons, or "changedetections", from studies of temporal series containing tens or evenhundreds of images, which is the meaning we ascribe to "hyper temporalanalysis".

Our intent in this section is to examine the existing status o fhypertemporal image analysis and to suggest new approaches to analyzingtemporal image sequences which may have potential applications i nclimate change studies. In the discussion that follows, we briefly reviewthe status of methods for change detection before critiquing potent ia lhypertemporal analysis techniques.

4.1. CHANGE DETECTION

Two excellent reviews of methods for extracting changes from image pa i rscan be found in Singh (1989) and Eastman and McKendry (1991), and a r esummarized in Table 4. Although most of these techniques may be o flimited use for long-term climatic studies, they may be useful in p r e -processing temporal data for subsequent analysis by other methods. Forexample, image differencing has been shown to effectively remove anygeographically fixed variations in sea ice extent and when the data to b eanalyzed are arranged as monthly averages (Cahalan and Chiu, 1 9 8 6 ) .Differencing can also be used to remove the seasonal cycle from a t imeseries (Hipel and McLeod, 1992). Both effects can be advantageous f o risolating regional and seasonal patterns in long spatial-temporal d a t aseries.

4.2. VISUALIZATION

The human mind is unequalled in its powers of spatial association a n dpattern recognition. Computers have unprecedented powers o fmanipulating large masses of digital data. Computers, then, a r eeffectively used as display tools for the presentation of unique views o fdata for the mind to analyze. This is the fundamental premise behind allaspects of image processing, but most importantly that of the descriptiveand exploratory techniques known as computer visualization.

Underlying the concept of visualization is the idea that an observer c a nbuild a mental model, the visual attributes of which represent d a t aattributes in a definable manner. Computer graphics has an impor tan trole in scientific visualization: it is not in generating elaborate, realisticimages of ideas; rather it is by creating abstractions of the data. Theabstraction process, if successful, helps to distinguish pattern from noise.Choosing the appropriate graphics display representation can be the keyto critical and comprehensive application of the data, thus benefitingsubsequent analysis, processing, or decision making. The choice of t h erepresentation ultimately depends on the research context (MacEachrenand Ganter, 1990; Wills et al., 1990; Robertson, 1991).

The most basic method of presenting the temporal elements of ahypertemporal data set is to display each time slice as an independentview, ideally juxtaposed so that the viewer can simultaneously c o m p a r ethe patterns at all times in the series. This is a technique we call "stat icimage sequencing". The focus here is on the status of the m a p p e dphenomena at each discrete instant; the search for any patterns of changeis left up to the viewer (Monmonier, 1990). This approach is a c o m m o nmethod for displaying hypertemporal data because of its computat ionalsimplicity. The ice atlases compiled by Zwally et al. (1983a), Parkinson e tal. (1987b), LeDrew et al. (1992c), and Gloersen et al. (1993) all provideexcellent examples of static image sequencing. The technique benefi tsfrom the arrangement of as many temporal data slices as possible in asingle view (e.g., on a single page). The human mind has superior pa t t e r nrecognition powers within a single image, however, its capabilities f o rimage comparisons is somewhat more limited and reduced further stillwhen the image arrangement does not promote instantaneous visualassociations (Calkins, 1984).

The next logical step beyond visual examination of static image sequencesis the visualization of dynamic image sequences, or animation. In a nanimation, temporally sequenced views of the data are displayed in qu icksuccession so as to give the impression of motion. Animation is based o nthe principle that the eye-brain mechanism retains, momentarily, images

of objects it has seen after the objects have been removed. Therefore, ifthe eye is shown a series of static views of objects at a rapid rate, with t h eobjects changing positions only slightly from frame to frame, the illusionof fluid life-like motion is created by the brain (Campbell and Egbert,1990). The technique can provide new insights into the data to b eanalyzed. Gloersen (1990) constructed a time-lapse study of a sea icefeature of the Greenland Sea known as the "Odden" from 2-day iceconcentration estimates over 9 years from SMMR image data. Theanimation allowed the extraction of previously unknown informat ionabout the formation, growth and movement of this ice feature. In a n o t h e rstudy, Assel and Ratkos (1991) created an interactive animation for t h eseasonal ice cycle on the Great Lakes. Spatial and temporal variations o fthe ice characteristics quickly emerged and lead to a review of potent ia lforcing mechanisms.

4.3. CLASSIFICATION

If multispectral classification is based on the premise that image pixelscan be grouped according to unique spectral characteristics, then it wouldappear that an image should also be classifiable according to un iquetemporal characteristics. Two requirements of mult ispectralclassification are: (i) a good understanding of the relationship be tweenthe spectral data and the surface features; and (ii) the choice o fappropriate statistical tools to discriminate relevant categories from t h edata (Marceau, 1989). With corresponding changes from "spectral" t o"temporal" this becomes an appropriate creed for hyper tempora lclassification. Much work has already been done in relating passivemicrowave brightness temperatures to ice characteristics and scientistsnow have a reasonable understanding of some of the more generalrelationships. In this section then, some appropriate statistical tools a r ereviewed. We focus on unsupervised classification methods because theyprovide a more accurate description of the natural patterns in the d a t aand because of the scarcity of adequate training data for supervisedtechniques.

The de facto standard approach to multispectral image segmentation isthe maximum likelihood technique where each pixel is assigned to t h eclass in which it has the highest statistical probability of being a m e mb e r .For hypertemporal applications, this method assumes a Gaussian t empora ldistribution of ice parameters. This assumption may hold for Arctic s e aice data where the seasonal cycle of growth and decay is approximatelysymmetrical (Comiso and Zwally, 1984).

There are other approaches to multispectral image classification,including some nonparametric techniques which could also haveapplicability in the hypertemporal domain. For example, Key et al. ( 1 9 8 9 )and Maslanik et al. (1990) describe the use of neural networks and SMMRimagery to examine "second order" ice information (e.g., the onset a n dduration of melt, ice pack metamorphosis, and timing of new ice growth)and relate these to changes in ice condition over an annual cycle. Theyconclude that the neural network approach to classification showspromise for sea ice applications because of its flexible pattern recognit ionskills, fuzzy tolerances, and ability to use non-numeric information.

When reviewing classification methods for hypertemporal imagery,perhaps we can glean insight from the attempts which have been made t oanalyze hyperspectral data (remote sensing images acquired by imagingspectrometers potentially including over 200 spectral channels per pixel).The classification of such high spectral resolution data is clearly achallenging task due to the large number of channels involved. Analysismethods are required to extract the most information from each channelwhile minimizing computation time. Two possible classificationmethodologies include full spectrum classification and feature selectionfollowed by classification. Full spectrum classification combinestechniques for a parametric representation of the spectrum derived f o reach pixel with similarity measures used for classification purposes, whilethe second approach uses techniques and transformations for d a t areduction prior to classification (Staenz and Goodenough, 1 9 9 0 ) .Techniques for full spectrum classification do exist, but they are n o tgenerally found in standard image analysis software packages (Mazer e tal., 1988; Cetin and Levandowski, 1991; Edwards et al., 1991).

The main criticism of full spectrum classifiers is that their computa t iontime increases exponentially with an increase in the number of channels .It is evident, however, that not every time slice or spectral band conta insunique information, in fact pixel values are often highly cor re la tedbetween channels. Thus, a more efficient approach may be to employ adata reduction strategy before classification. Existing methods f o rchannel reduction include band averaging (Staenz and Goodenough,1990), band-moment analysis (Rundquist and Di, 1989), statisticalchannel separability measures (e.g., Sheffield, 1985; Labovitz, 1986; Gongand Howarth, 1990; Mausel et al., 1990), and principal componen t sanalysis (Chavez and Kwarteng, 1989).

4.4. PRINCIPAL COMPONENTS ANALYSIS

Principal Components Analysis (PCA) is a commonly used technique i nremote sensing image analysis. Traditionally it has been applied for imageenhancement and channel reduction, however, it has also been effectivelyused in change detection studies (Table 5). An appealing aspect of PCA i nthe present context is that the technique is also widely accepted b yclimatologists (who frequently refer to it as eigenvector analysis o rempirical orthogonal function analysis) as an effective data analysis t oo l(e.g., Stidd, 1967; LeDrew, 1976; Kelly et al., 1982). This overlap be tweendisciplines has not been overlooked by scientists looking for t empora lchanges in both polar and non-polar environments.

In a principal components analysis, the measurement space is rotated i nsuch a way as to maximize the variance between the variables. In r e m o t esensing, these variables can be multispectral channels a n d / o rhypertemporal images. The objectives in applying PCA are to unders t andhow the variables are interrelated and to assess the level of statisticalredundancy present. In hypertemporal studies, the technique offers auseful exploratory tool to analyze the interrelationships between regularlysampled hypertemporal remote sensing data sets (Townshend et al.,1 9 8 5 ) .

When images of different dates are compared with a view to moni tor ingchange, there will be a high correlation between them: those parts of t h escene which show an absence of correlation are of interest as theyrepresent areas of change (Byrne and Crapper, 1980). By the very n a t u r eof the method, the first component will always contain the maximumamount of variation derived from the set of variables. Thus, the f irs tcomponent is commonly referenced as "integrated brightness"; it providesa good index of long-term variations since it is not affected by strong, b u tlocalized, extremes which can influence large-scale averages and obscu reunderlying trends (Kelly et al., 1982). Experiments using monthly d a t aover an annual cycle have found that the second component tracks t h eseasonal evolution quite well (Townshend et al., 1985; Tateishi a n dKajiwara, 1990). The higher components represent increasingly localvariations and anomalies; they are often the ones selected for m o r edetailed analysis (Walsh and Johnson, 1979b; Crane, 1983).

It is evident that PCA can be a powerful tool for the monitoring of po la rclimate with sequential remote sensing observations. It allows for t h eabstraction of change from a time series as a whole, not just a segment o fi t .

4.5. TREND SURFACE ANALYSIS

In climate change analyses we are interested in determining how much o fa variable's response over time is real (i.e., a climatic signal) and h o wmuch of it is local variation (i.e., climatic noise). If a regression line ( o rcurve) is fit to a time series plot of sea ice extent, the parameters of t h eregression equation would give us some indication of a climatic signal,while the residuals could be interpreted as a measure of local climaticvariability. This technique has been used extensively in climate resea rchto extract trends from time series plots (e.g., Chapman and Walsh, 1992).

Image regression is a technique which has been applied to image pairs t odifferentiate between actual change and predicted change (Singh, 1989 ;Eastman and McKendry, 1991). The procedure is initiated by consideringpixel values from the image at time t to be samples of the independentvariable and pixels from the time t+1 image as samples of the dependen tvariable. A regression line is then fit to the data points. The c o m p u t e dslope and intercept values are applied on a per-pixel basis to the f irs timage to produce a predicted image for time t+1. The actual a n dpredicted time t+1 images are then differenced or ratioed to isolate t h echanged areas. The image regression technique is limited to pairwisecomparisons, however.

Trend surface analysis extends this concept into two dimensions. Insteadof fitting a regression line to a time series plot, a regression ( " t rend")surface is fit to a spatial-temporal data set (an image in our case). Idealtrend surface images provide a smooth characterization of their i npu tdata. When a trend surface has time as one of its independent variables,it can be thought of as an isochronic or time contour map (Monmonier ,1990). Some studies have found residual images to be more useful f o rpattern exploration than the trend image (e.g., Chorley and Haggett,1965). Trend surface analyses have not been widely applied in cl imatechange studies although they have the potential for revealing significanttrends and regionalizations.

4.6. TIME SERIES ANALYSIS

In time series analysis, the deterministic components to a series (e.g.,seasonality and trends) are identified and removed in a process called"pre-whitening". The residual stochastic component in the series is t h e nmodelled using autoregressive and/or moving average functions (Hipeland McLeod, 1992). Although time series analyses have not b e e ngenerally used in studies of Arctic sea ice or incorporated into image

analysis programs, they have been extensively applied by climatologistsand have a certain appeal to the problem at hand. This attraction c o m e snot only from the modelling capabilities of time series equations, but a lsofrom their demonstrated proficiency at generating forecasts beyond t h emeasured data space.

In the 1970s there was a flurry of activity to extend the two-dimensionalnature of conventional time series analysis techniques into t h r e edimensions, i.e., spatial time series (Bennett, 1975; Martin and Oeppen,1975; Bennett and Haining, 1976; Bennett, 1979; Pfeifer and Deutsch,1980). A spatial time series is a multivariate model which attempts t osimultaneously describe and forecast a collection of time seriesrepresenting spatially diverse regions, taking into account the spatialautocorrelation inherent between neighbouring regions (Pfeifer a n dDeutsch, 1980; Adamowski et al., 1985). Unfortunately not m u c hprogress has been made in developing spatial time series since then ,largely because of the complexity of the model and the difficulty o fparameter estimation (Curry, 1970; Parkes and Thrift, 1980).

5. DISCUSSION

Mysak (1990) proposed six working questions which should b econsidered when assessing the suitability of sea ice (or any o t h e rgeophysical feature) as a proxy indicator of climate change. Based on t h ereview presented here, we are now in a position to suggest some answers.

1 . Through what mechanisms can sea ice be used as an indicator o fclimate change?

• Sea ice has intimate bi-directional feedback mechanisms with t h eearth's atmospheric and oceanic circulation systems. The climaticrole that sea ice plays is generally expected to be based on i tssurface albedo, insulative properties, thermal capacitance, and b r inecapaci tance.

2 . What specific geophysical parameters are required to characterize t h ebehaviour of these indicators?

• The potential sea ice indicators for climate change identified i nsection 2 are extent, type, concentration, thickness, dynamics, a n dseasonality. These were shown to have distinct regionaldistr ibutions.

3 . With what accuracy and precision do these indicators need to b emeasured? Over what spatial and temporal scales?

• From the analyses in section 2 and Barber et al. (1992b), we find t h efollowing ideals:

IndicatorIdeal Spatial

RequirementsIdeal Temporal

Requirementsice extent >1 km 3 daysice type > 1 km 3 daysice concentration > 1 km 3 daysice thickness 10 km 3 daysice dynamics 10 km 1 dayonset of melt 10 km 3 days

4 . What data sources are currently available for the measurement of theseindicators?

• An evaluation of remote sensing systems indicated that passivemicrowave imagery is the only data source with the necessaryhistorical record and all weather/day-night sensing capabilities.

5 . Where are the discrepancies between what is required and what isavailable?

• Of all the parameters listed in Question 2, the only one passivemicrowave imagers cannot measure ice thickness. Past and p resen tmicrowave sensors do not meet the required spatial resolutions o fany indicator, however, this may not be critical in hemispher icanalyses.

6 . What areas of product development/research are needed to use sea iceas an indicator of climate change?

• If effective use is to be made of the historical passive microwavedata record, new hypertemporal image analysis techniques must b edeveloped.

Table 6 lists the potential hypertemporal image analysis funct ionsapplicable to examining sea ice parameters on remote sensing imagery. Itis clear that there are many options for the analysis of hyper tempora limage sequences. How do we proceed when faced with analyzing ahypertemporal data set? Which analysis methods have the potential t ohelp us extract meaningful information from these data? We propose ahierarchy of analysis functions in Table 7 which may provide s o m eguidance. The primary techniques were selected based on the i r

demonstrated capabilities of scene processing, information extraction a n dprocess description. Primary functions are likely to be the most useful f o rgeneral explorations of the temporal nature of the data. Whereinteresting patterns of change are found the secondary techniques may b ehelpful in explaining their nature. Re-application of the primary me t h o d susing a different parameter set or to different subsets of the data m a yalso be useful. The tertiary functions can be considered to be the "workerroutines". They have limited applicability, but are frequently useful whenapplied to pre-processing or post-processing the data.

During the course of an entire hypertemporal analysis exercise, it i sprobable that techniques from each category will be applied in a varietyof sequences until the analyst is satisfied with the meanings a n ddefinitions which have been derived from the imagery. Much depends o nthe nature of both the data and the problem. The key is in effectivecombination of the computer's data processing capabilities with t h ehuman mind's pattern recognition strengths.

6. CONCLUSIONS

In this review, we have shown the critical role that polar sea ice plays i nthe world's climatic systems; it is clear that the Arctic is a region o fsensitive indicators of climatic change. An analysis of potential r e m o t esensing techniques indicated that sea ice extent, type, concentra t ion,dynamics, and melt onset have been monitored with sufficient (a l thoughnot ideal) spatial and temporal resolution by orbital passive microwavesensors to make these variables potential climatic indicators. The passivemicrowave data record has revealed large interannual variability in e a c hof these parameters and few indications of long term trends. Examinationof these images also show, in addition to the large temporal variability o fthese data, that they exhibit strong regionalisms and seasonalities. Thisinformation can be used to focus attention in those seasons and placeswhen changes may be most apparent.

Hypertemporal image analysis of remote sensing data is still in its infancy,largely because of a lack of repetitive and inexpensive data, and adaptableprocessing algorithms. We are on the brink of temporal data explosion a sa longer historical record is accumulated for existing sensors, as r e m o t esensing scientists begin to explore the utility of "less conventional"sensors for which the data are relatively inexpensive (e.g., AVHRR a n dpassive microwave), and as new, high data-rate instruments are b rough ton-line towards the end of this century (e.g., EOS and RADARSAT).

Existing image processing tools need to be adapted, and new techniquesdeveloped, to handle these data.

Application of the hypertemporal image processing techniques reviewedhere is not necessarily restricted to studies of sea ice. Indeed, the adventof satellite remote sensing has presented us with intriguing opportuni t iesto investigate, in good spatial and temporal detail, even the most r e m o t eor rarely visited areas of the earth's surface. However, most existingclimate change studies which are derived from remotely sensed data havebeen temporal analyses based on aggregation of the spatial informat ionfrom each image, generally into a single parameter (e.g., ice extent). Withthe exception of a few recent efforts, not many published results a r ebased on detailed spatial analyses of the type which are possible t h roughimage processing methods. This is partly attributable to the inability o fstandard image analysis software to work in both a spatial and a t empora lmode. We have examined the issue of hypertemporal image analysis a n dconclude that many of the techniques used in multispectral imageprocessing, such as classification and principal components analysis, a r ealso applicable in the time domain. We call for their increased use in t r u espatial-temporal analyses and for the development of new imageprocessing algorithms specifically for modelling and forecasting change.

ACKNOWLEDGMENTS

We would like to thank Dave Barber, Doug Dudycha, Barry Goodison, PhilHowarth, and Ed Jernigan for their valuable comments on an ear l ierversion of this paper.

REFERENCES

Adamowski, Kazimierz, Fadil B. Mohamed, Nicholas R. Dalezios and LouisG. Birta, 1985. "Space-Time ARIMA Modelling of Precipitation TimeSeries", in Multivariate Analysis of Hydrologic Processes, (Shen, HsiehWen, et al., eds.). Fort Collins, Colorado: Engineering Research Center,Colorado State University. pp. 217-227.

Alfultis, Michael A. and Seelye Martin, 1987. "Satellite Passive MicrowaveStudies of the Sea of Okhotsk Ice Cover and its Relation to OceanicProcesses, 1978-1982", J. Geophysical Research, Vol. 92, No. C12, pp.13013-13028 .

Anderson, Mark R., 1987. "The Onset of Spring Melt in First-Year IceRegions of the Arctic as Determined from Scanning Multichannel

Microwave Radiometer Data for 1979 and 1980", J. Geophysical Research,Vol. 92, No. C12, pp. 13153-13163.

Assel, R.A. and J.M. Ratkos, 1991. "Animation of the Normal Ice Cycle ofthe Laurentian Great Lakes of North America", Proceedings, SeventhInternational Conference on Interactive Information and ProcessingSystems for Meteorology, Hydrology, and Oceanography, January 1991,New Orleans LA., pp. 331-335.

Barber, D.G., E.F. LeDrew, D.G. Flett, M. Shokr and J. Falkingham, 1992a ."Seasonal and Diurnal Variations in SAR Signatures of Landfast Sea Ice",IEEE Transactions on Geoscience and Remote Sensing, Vol. 30, No. 3, p p .638-642 .

Barber, David G., Michael J. Manore, Thomas A. Agnew, Harold Welch,Eric D. Soulis and Ellsworth F. LeDrew, 1992b. "Science Issues Relating toMarine Aspects of the Cryosphere: Implications for Remote Sensing", Can.J. Remote Sensing, Vol. 18, No. 1, pp. 46-55.

Barry, R.G., 1983. "Arctic Ocean Ice and Climate: Perspectives on aCentury of Polar Research", Annals of the Association of AmericanGeographers , Vol. 73, No. 4, pp. 485-501.

Bennett, R.J., 1975 "The Representation and Identification of Spatio-Temporal Systems: An Example of Population Diffusion in North-WestEngland", Transactions, Institute of British Geographers, Vol. 66, pp. 73-9 4 .

Bennett, R.J., 1979. Spatial Time Series. London: Pion. 674 pp.

Bennett, R.J. and R.P. Haining, 1976. Space-Time Models: An Introductionto Concepts. London: University College Department of Geography,Occasional Paper No. 28. 35 pp.

Bjerkelund, C.A., R.O. Ramseier and I.G. Rubinstein, 1990. "Validation ofthe SSM/I and AES/York Algorithms for Sea Ice Parameters", in Ackley, S.F.and W.F. Weeks (eds.), Sea Ice Properties and Processes, Proceedings:W.F. Weeks Sea Ice Symposium, San Francisco CA, December 1988. U.S.Army Cold Regions Research and Engineering Lab, Monograph M90-1, pp.206-208 .

Brown, Ross D. and Phil Cote, 1992. "Interannual Variability of LandfastIce Thickness in the Canadian High Arctic, 1950-89", Arctic, Vol. 45, No.3, pp. 273-284.

Burns, Barbara A. and Thomas Viehoff, 1991. "Merging AVHRR and SSM/IData: An Example for Ice Concentration Analyses in the Weddell Sea",Proceedings, 14th Canadian Symposium on Remote Sensing, CalgaryAlberta, May 1991. Ottawa: Canadian Remote Sensing Society, pp. 371-3 7 4 .

Byrne, G.F. and P.F. Crapper, 1980. "Land Cover Change Detection byPrincipal Components Analysis of Multitemporal MSS Data: The Presenceof Clouds", Proceedings, 14th International Symposium on Remote Sensingof Environment, April 1980, San Jose Costa Rica, pp. 1175-1382.

Byrne, G.F., P.F. Crapper and K.K. Mayo, 1980. "Monitoring Land-CoverChange by Principal Component Analysis of Multitemporal LandsatImagery", Remote Sensing of Environment, Vol. 10, pp. 175-184.

Cahalan, Robert F. and Long S. Chiu, 1986. "Large-Scale Short-Period SeaIce Atmosphere Interaction", J. Geophysical Research, Vol. 91, No. C9, pp.10709-10717 .

Calkins, Hugh W., 1984. "Space-Time Data Display Techniques",Proceedings, First International Symposium on Spatial Data Handling,Zürich, August 1984. Zürich: Univsersität Zürich-Irchel, Vol. 2, pp. 324-3 3 1 .

Campbell, Craig S. and Stephen L. Egbert, 1990. "Animated Cartography:Thirty Years of Scratching the Surface", Cartographica , Vol. 27, No. 2, pp.24-46 .

Carsey, F.D., 1984. "Prospects for Describing and Monitoring From Spacethe Elements of the Seasonal Cycle of Sea Ice", Annals of Glaciology, Vol.5, pp. 37-42.

Carsey, F.D., 1985. "Summer Arctic Sea Ice Character from SatelliteMicrowave Data", J. Geophysical Research, Vol. 90, No. C3, pp. 5015-5 0 3 4 .

Cavalieri, D.J. and C.L. Parkinson, 1987. "On the Relationship BetweenAtmospheric Circulation and the Fluctuations in the Sea Ice Extents of theBering and Okhotsk Seas", J. Geophysical Research, Vol. 92, No. C7, pp.7141-7162 .

Cavalieri, D.J., J.P. Crawford, M.R. Drinkwater, D.T. Eppler, L.D. Farmer,R.R. Jentz and C.C. Wackerman, 1991. "Aircraft Active and PassiveMicrowave Validation of Sea Ice Concentration from the Defense

Meteorological Satellite Program Special Sensor Microwave Imager", J .Geophysical Research, Vol. 96, No. C12, pp. 21989-22008.

Cetin, Haluk and Donald W. Levandowski, 1991. "Interactive Classificationand Mapping of Multi-Dimensional Remotely Sensed Data Using n-Dimensional Probability Density Functions (nPDF)", PhotogrammetricEngineering & Remote Sensing, Vol. 57, No. 12, pp. 1579-1587.

CGCP (Canadian Global Change Program), 1989. Contribution of SatelliteObservations to the Canadian Global Change Program, Ottawa: The RoyalSociety of Canada and the Canada Centre for Remote Sensing, CanadianGlobal Change Program Report No. 3, 48 pp.

Chapman, William L. and John E. Walsh, 1992. "Recent Variations of SeaIce and Air Temperature in High Latitudes", Bull. American MeteorologicalSociety , December 1992, in press.

Chavez, Pat S., Jr. and Andrew Yaw Kwarteng, 1989 "Extracting SpectralContrast in Landsat Thematic Mapper Image Data Using Selective PrincipalComponents Analysis", Photogrammetric Engineering & Remote Sensing,Vol. 55, No. 3, pp. 339-348.

Chorley, R.J. and P. Haggett, 1965. "Trend-Surface Mapping inGeographical Research", in B.J.L. Berry, Spatial Analysis . EnglewoodCliffs: Prentice Hall. pp. 195-217.

Clarke, R. Allyn and E. Peter Jones, 1989. "The Arctic Ocean: Its Role inthe Global Climate Engine", Proceedings, Symposium on the Arctic andGlobal Change, Ottawa Canada, October 25-27, 1989. Washington D.C.:The Climate Institute. pp. 42-57.

Comiso, J.C., 1986. "Characteristics of Arctic Winter Sea Ice FromSatellite Multispectral Microwave Observations", J. Geophysical Research,Vol. 91, No. C1, pp. 975-994.

Comiso, J.C. and C.W. Sullivan, 1986. "Satellite Microwave and In SituObservations of the Weddell Sea Ice Cover and its Marginal Ice Zone",Journal of Geophysical Research, Vol. 91, No. C8, pp. 9663-9681.

Comiso, J.C. and H.J. Zwally, 1982. "Antarctic Sea Ice ConcentrationsInferred from Nimbus 5 ESMR and Landsat Imagery", J. GeophysicalResearch , Vol. 87, No. C8, pp. 5836-5844.

Comiso, J.C. and H.J. Zwally, 1984. "Concentration Gradients andGrowth/Decay Characteristics of the Seasonal Sea Ice Cover", J .Geophysical Research, Vol. 89, No. C5, pp. 8081-8103.

Crane, R.G., 1978. "Seasonal Variations in Sea Ice Extent in the DavisStrait-Labrador Sea Area and Relationships with Synoptic ScaleAtmospheric Circulation", Arctic , Vol. 31, No. 4, pp. 434-447.

Crane, R.G., 1983. "Atmosphere - Sea Ice Interactions in theBeaufort/Chukchi Sea and in the European Sector of the Arctic", J .Geophysical Research, Vol. 88, No. C7, pp. 4505-4523.

Crane, Robert G. and Mark R. Anderson, 1989. "Spring Melt Patterns inthe Kara/Barents Sea: 1984", GeoJournal , Vol. 18, No. 1, pp. 25-33.

Crane, R.G., R.G. Barry and H.J. Zwally, 1982. "Analysis of Atmosphere-Sea Ice Interactions in the Arctic Basin using ESMR Microwave Data", Int.J. Remote Sensing, Vol. 3, No. 3, pp. 259-276.

Curry, Leslie, 1970. "Applicability of Space-Time Moving AverageForecasting", Regional Forecasting (Chisholm, et al., eds.), Hamden CT:Archon, 470 pp.

Daley, R., 1989 "General Circulation Models", Proceedings, Symposium onthe Arctic and Global Change, Ottawa Canada, October 25-27, 1989.Washington D.C.: The Climate Institute. pp. 32-33.

Davis, Frank W., Dale A. Quattrochi, Merrill K. Ridd, Nina S-N Lam,Stephen J. Walsh, Joel C. Michaelsen, Janet Franklin, Douglas A. Stow,Chris J. Johannsen and Carol A. Johnston, 1991. "Environmental AnalysisUsing Integrated GIS and Remotely Sensed Data: Some Research Needsand Priorities", Photogrammetric Engineering & Remote Sensing, Vol. 57,No. 6, pp. 689-697.

Dey, B., 1981. "Monitoring Winter Sea Ice Dynamics in the CanadianArctic with NOAA-TIR Images", J. Geophysical Research, Vol. 86, No. C4,pp. 3223-3235.

Eastman, J. Ronald and Jean E. McKendry, 1991. Change and Time SeriesAnalysis, Volume 1 of Explorations in Geographic Information SystemsTechnology, Geneva Switzerland: United Nations Institute for Trainingand Research, 85 p.

Edwards, Geoffrey, Annick Jaton and Keith P.B. Thomson, 1991."Manipulating Hyperspectral Data", Proceedings, 14th Canadian

Symposium on Remote Sensing, Calgary Alberta, May 1991. Ottawa:Canadian Remote Sensing Society, pp. 356-359.

Emery, W.J., M. Radebaugh, C.W. Fowler, D. Cavalieri and K. Steffen, 1991."A Comparison of Sea Ice Parameters Computed from Advanced Very HighResolution Radiometer and Landsat Satellite Imagery and From AirbornePassive Microwave Radiometry", J. Geophysical Research, Vol. 96, No. C12,pp. 22075-22085.

Eyton, J. Ronald, 1991. "Rate-of-Change Maps", Cartography andGeographic Information Systems, Vol. 18, No. 2, pp. 87-103.

Fung, Tung and Ellsworth LeDrew, 1987. "Application of PrincipalComponents Analysis to Change Detection", Photogrammetric Engineeringand Remote Sensing, Vol. 53, No. 12, pp. 1649-1658.

Gloersen, P. 1990. "A Time-Lapse Motion Study of the 'Odden' During1978-1987 as Observed with the Nimbus-7 Scanning MultichannelMicrowave Radiometer", in Ackley, S.F. and W.F. Weeks (eds.), Sea IceProperties and Processes, Proceedings: W.F. Weeks Sea Ice Symposium,San Francisco CA, December 1988. U.S. Army Cold Regions Research andEngineering Lab, Monograph M90-1, pp. 177-179.

Gloersen, Per and William J. Campbell, 1988. "Variations in the Arctic,Antarctic, and Global Sea Ice Covers During 1978-1987 as Observed withthe Nimbus-7 Scanning Multichannel Microwave Radiometer", J .Geophysical Research, Vol. 93, No. C9, pp. 10666-10674.

Gloersen, Per, William J. Campbell, Donald J. Cavalieri, Josefino C. Comiso,Claire L. Parkinson and H. Jay Zwally, 1993. Arctic and Antarctic Sea Ice,1978-1987: Satellite Passive-Microwave Observations and Analysis,Washington DC: National Aeronautics and Space Administration, NASA SP-511, 290 pp.

Gong, Peng and Philip J. Howarth, 1990. "A Comparison of Two Feature-Selection Procedures for Supervised Land-Cover Classification",Proceedings, 13th Canadian Symposium on Remote Sensing, Fredericton,N.B., 16-19 July, pp. 190-202.

Hare, F. Kenneth, 1982. "Confusion About Climate: What is Happening?",Canadian Geographic, Vol. 101, No. 6, pp. 48-53.

Hipel, Keith W. and A. Ian McLeod, 1992. Time Series Modelling of WaterResources and Environmental Systems, Amsterdam: Elsevier.

Houghton, John T., 1990. "Scientific Assessment of Climate Change:Summary of the IPCC Working Group I Report", Climate Change: Science,Impacts and Policy, Proceedings of the Second World Climate Conference,29 October 1990. Cambridge: Cambridge University Press, pp. 23-45.

Howarth, Philip J. and Gregory M. Wickware, 1981. "Procedures forChange Detection Using Landsat Digital Data", Int. J. Remote Sensing, Vol.2, No. 3, pp. 277-291.

Ikeda, M., 1990a. "Feedback Mechanism Among Decadal Oscillations inNorthern Hemisphere Atmospheric Circulation, Sea Ice, and OceanCirculation", Annals of Glaciology, Vol. 14, pp. 120-123.

Ikeda, M., 1990b. "Decadal Oscillations of the Air-Ice-Ocean System inthe Northern Hemisphere", Atmosphere-Ocean , Vol. 28, No. 1, pp. 106-1 3 9 .

Ingebritsen, S.E. and R.J.P. Lyon, 1985. "Principal Components Analysis ofMultitemporal Image Pairs", Int. J. Remote Sensing, Vol. 6, No. 5, pp. 687-6 9 6 .

Jacobs, John D. and John P. Newell, 1979. "Recent-Year-to-Year Variationsin Seasonal Temperatures and Sea Ice Conditions in the Eastern CanadianArctic", Arctic, Vol. 32, No. 4, pp. 345-354.

Johnson, Douglas D., 1989. Remote Sensing of Polar Regions: A Review ofits Development and Application to Atmosphere-Ice-Ocean InteractionProcesses. Earth Observations Laboratory Technical Report ISTS-EOL-TR89-002, Department of Geography, University of Waterloo, 72 pp.

Kelly, P.M., P.D. Jones, C.B. Sear, B.S.G. Cherry and R.K. Tavakol, 1982."Variations in Surface Air Temperatures: Part 2. Arctic Regions, 1881-1980", Monthly Weather Review, Vol. 110, pp. 71-83.

Kelly, P.M., C.M. Goodess and B.S.G. Cherry, 1987. "The Interpretation ofthe Icelandic Sea Ice Record", J. Geophysical Research, Vol. 92, No. C10,pp. 10835-10843.

Key, J., J. A. Maslanik, and A. J. Schweiger, 1989. "Classification ofMerged AVHRR and SMMR Arctic Data with Neural Networks",Photogrammetric Engineering & Remote Sensing, Vol. 55, No. 9, pp. 1331-1 3 3 8 .

Kukla, G., 1993. "Glacial Start and Global Warming: What to Watch",Proceedings, Snow Watch '92: Detection Strategies for Snow and Ice,

Niagara-on-the-Lake Canada, March 1993. Waterloo Canada: EarthObservations Laboratory, Institute for Space and Terrestrial Science, ISTS-EOL-TR93-005, pp. 93-110.

Labovitz, M.L., 1986 "Issues Arising from Sampling Designs and BandSelection in Discriminating Ground Reference Attributes Using RemotelySensed Data", Photogrammetric Engineering & Remote Sensing, Vol. 52,No. 2, pp. 201

Ledley, Tamara Shapiro, 1990. "The Impact of Milankovitch SolarRadiation Variations on Sea Ice and Air Temperature in a Coupled Energy-Balance Climate-Sea-Ice Model", Annals of Glaciology, Vol. 14, pp. 144-1 4 7 .

LeDrew, Ellsworth Frank, 1976. Physical Mechanisms Responsible for theMajor Synoptic Systems in the Eastern Canadian Arctic in the Winter andSummer of 1973. National Center for Atmospheric Research, University ofColorado, NCAR Cooperative Thesis No. 38, 205 pp.

LeDrew, Ellsworth F., Douglas Johnson and James A. Maslanik, 1991. "AnExamination of Atmospheric Mechanisms that may be Responsible for theAnnual Reversal of the Beaufort Sea Ice Field", Int. J. Climatology, Vol. 11,pp. 841-859

LeDrew, Ellsworth F., 1992a. "Polar Regions, Influence on ClimateVariability and Change", in Encyclopedia of Earth System Science, Vol. 3(Nierenberg, William A., ed.), Academic Press.

LeDrew, Ellsworth, 1992b. "The Role of Remote Sensing in the Study ofAtmosphere-Cryosphere Interactions in the Polar Basin", CanadianGeographer , Vol. 36, No. 4, pp. 337-350.

Lemke, P. E.W. Trinkl and K. Hasselmann, 1980. "Stochastic DynamicAnalysis of Polar Ice Variability", J. Physical Oceanography, Vol. 10, pp.2100-2120 .

Lillesand, Thomas M. and Ralph W. Kiefer, 1987. Remote Sensing andImage Interpretation (2nd ed.). New York: Wiley, 721 pp.

Lodwick, G.D., 1979. "Measuring Ecological Changes in MultitemporalLandsat Data Using Principal Components", Proceedings, 13thInternational Symposium on Remote Sensing of Environment, April 1979,Ann Arbor MI, pp. 1131-1141.

MacEachren, Alan M. and John H. Ganter, 1990. "A Pattern IdentificationApproach to Cartographic Visualization", Cartographica , Vol. 27, No. 2,pp. 64-81.

Marceau, Danielle, 1989. A Review of Image Classification Procedureswith Special Emphasis on the Grey-Level Cooccurrence Matrix Method forTexture Analysis, ISTS-EOL-TR89-007, 59 pp.

Martin, R.L. and J.E. Oeppen, 1975. "The Identification of RegionalForecasting Models Using Space-Time Correlation Functions",Transactions, Institute of British Geographers, Vol. 66, pp. 95-118.

Maslanik, James, Jeff Key and Axel Schweiger, 1990. "Neural NetworkIdentification of Sea-Ice Seasons in Passive Microwave Data", Proceedings,IGARSS'90, Washington D.C., 20 May 1990. New York: IEEE, pp. 1281-1 2 8 4 .

Massom, Robert A., 1991. Satellite Remote Sensing of Polar Regions.London: Belhaven, 307 pp.

Mausel, P.W., W.J. Kramber, and J.K. Lee, 1990. "Optimum Band SelectionFor Supervised Classification of Multispectral Data", PhotogrammetricEngineering & Remote Sensing, Vol. 56, No. 1, pp. 55 - 60.

Maykut, Gary A. and Norbert Untersteiner, 1971. "Some Results from aTime-Dependent Thermodynamic Model of Sea Ice", J. GeophysicalResearch , Vol. 76, No. 6, pp. 1550-1575.

Mazer, Alan S., Miki Martin, Meemong Lee and Jerry S. Solomon, 1988."Image Processing Software for Imaging Spectrometry Data Analysis",Remote Sensing of Environment, Vol. 24, pp. 201-210.

Monmonier, Mark, 1990. "Strategies for the Visualization of GeographicTime-Series Data", Cartographica , Vol. 27, No. 1, pp. 30 - 45.

Mysak, L., 1990. "Current and Future Trends in Arctic Climate Research:Can Changes of the Arctic Sea-Ice be Used as an Early Indicator of GlobalWarming?", personal notes from a Canadian Ice Working Group workshop,McGill University, November 1990.

Niebauer, H.J., 1983. "Multiyear Sea Ice Variability in the Eastern BeringSea: An Update", J. Geophysical Research, Vol. 88, No. C5, pp. 2733-2 7 4 2 .

Orheim, Olav and Baerbel Lucchitta, 1990. "Investigating Climate Changeby Digital Analysis of Blue Ice Extent on Satellite Images of Antarctica",Annals of Glaciology, Vol. 14, pp. 211-215.

Parkes, Don and Nigel Thrift, 1980. Times, Spaces, and Places. Chichester:Wiley. 527 pp.

Parkinson, Claire L., 1989. "On the Value of Long-Term Satellite PassiveMicrowave Data Sets for Sea Ice/Climate Studies", GeoJournal , Vol. 18,No. 1, pp. 9-20.

Parkinson, Claire L., 1991. "Interannual Variability of the SpatialDistribution of Sea Ice in the North Polar Region", J. Geophysical Research,Vol. 96, No. C3, pp. 4791-4801.

Parkinson, C.L., J.C. Comiso, H.J. Zwally, D.J. Cavalieri, P. Gloersen andW.J. Campbell, 1987b. Arctic Sea Ice, 1973-1976: Satellite PassiveMicrowave Observations, Washington DC: National Aeronautics and SpaceAdministration, NASA SP-489, 296 pp.

Parkinson, Claire L. and Donald J. Cavalieri, 1989. "Arctic Sea Ice 1973-1987: Seasonal, Regional, and Interannual Variability", J. GeophysicalResearch , Vol. 94, No. C10, pp. 14499-14523.

Pfeifer, Phillip E. and Stuart Jay Deutsch, 1980. "A Three-Stage IterativeProcedure for Space-Time Modelling", Technometrics , Vol. 22, No. 1, pp.35-47 .

Robertson, Philip K., 1991. "A Methodology for Choosing DataRepresentations", IEEE Trans. Computer Graphics & Applications, May1991, pp. 56-67.

Roots, Fred, 1989. "Environmental Issues Related to Climate Change inNorthern High Latitudes", Proceedings, Symposium on the Arctic andGlobal Change, Ottawa Canada, October 25-27, 1989. Washington D.C.:The Climate Institute. pp. 6-31.

Ropelewski, Chester F., 1983. "Spatial and Temporal Variations inAntarctic Sea-Ice (1973-82)", J. Climate and Applied Meteorology, Vol.22, pp. 470-473.

Rothrock, D.A., Donald R. Thomas and Alan S. Thorndike, 1988."Principal Component Analysis of Satellite Passive Microwave Data OverSea Ice", J. Geophysical Research, Vol. 93, No. C3, pp. 2321-2332.

Rundquist, Donald C. and Liping Di, 1989. "Band-Moment Analysis ofImaging-Spectrometer Data", Photogrammetric Engineering & RemoteSensing , Vol. 55, No. 2, pp. 203-208.

Saab, David A. and Thomas W. Haythornthwaite, 1990. "Application ofSpatio-Temporal Database Models to Street Network Files", Proceedings:GIS for the 1990s; Ottawa Canada. Ottawa: Canadian Institute ofSurveying and Mapping. pp. 1072-1085.

Serreze, Mark C., Roger G. Barry and Alfred S. McLaren, 1989. "SeasonalVariations in Sea Ice Motion and Effects on Sea Ice Concentration in theCanada Basin", J. Geophysical Research, Vol. 94, No. C8, pp. 10955-10970.

Sheffield, Charles, 1985 "Selecting Band Combinations from MultispectralData", Photogrammetric Engineering & Remote Sensing, Vol. 51, No.12,AP6, pp. 681

Singh, Ashbindu, 1989. "Digital Change Detection Techniques UsingRemotely-Sensed Data", Int. J. Remote Sensing, Vol. 10, No. 6, pp. 989-1 0 0 3 .

Staenz, K. and D.G. Goodenough, 1990. "Airborne Imaging SpectrometerData Analysis Applied to an Agricultural Data Set", Global andEnvironmental Monitoring Techniques and Impacts, Proceedings: ISPRSCommission VII Mid-Term Symposium: September 1990, Victoria Canada.International Society for Photogrammetry and Remote Sensing, pp. 552-5 5 9 .

Steffen, Konrad and Axel Schweiger, 1991. "NASA Team Algorithm for SeaIce Concentration Retrieval from Defense Meteorological Satellite ProgramSpecial Sensor Microwave Imager: Comparison with Landsat SatelliteImagery", J. Geophysical Research, Vol. 96, No. C12, pp. 21971-21987.

Stidd, Charles K., 1967. "The Use of Eigenvectors for Climatic Estimates",J. Applied Meteorology, Vol. 6, pp. 255-264.

Stutheit, Juliann, 1991. "Temporal GIS Investigates Global Change", GISWorld , Vol. 4, No. 9 (Dec. 1991), pp. 68-72.

Tateishi, Ryutaro and Koji Kajiwara, 1990. "Global Land Cover Monitoringby NOAA GVI Data", Global and Environmental Monitoring Techniques andImpacts, Proceedings: ISPRS Commission VII Mid-Term Symposium,September 1990, Victoria Canada. International Society forPhotogrammetry and Remote Sensing, pp. 34-41.

Thomas, D.R. and D.A. Rothrock, 1989. "Blending Sequential ScanningMultichannel Microwave Radiometer and Buoy Data Into a Sea Ice Model",J. Geophysical Research, Vol. 94, No. C8, pp. 10907-10920.

Townshend, John R.G., Thomas E. Goff and Compton J. Tucker, 1985."Multitemporal Dimensionality of Images of Normalized DifferenceVegetation Index at Continental Scales", IEEE Transactions on Geoscienceand Remote Sensing, Vol. GE-23, No. 6, pp. 888-895.

Vowinckel, E. and S. Orvig, 1970. "The Climate of the North Polar Basin",in Orvig, S. (ed.) Climates of the Polar Regions. Amsterdam: Elsevier, pp.129-252 .

Walsh, John E., 1983. "The Role of Sea Ice in Climatic Variability:Theories and Evidence", Atmosphere-Ocean , Vol. 21, No. 3, pp. 229-242.

Walsh, John E. and Claudia M. Johnson, 1979a. "Interannual AtmosphericVariability and Associated Fluctuations in Arctic Sea Ice Extent", J .Geophysical Research, Vol. 84, No. C11, pp. 6915-6928.

Walsh, John E. and Claudia M. Johnson, 1979b. "An Analysis of Arctic SeaIce Fluctuations, 1953-77", J. Physical Oceanography, Vol. 9, pp. 580-591.

Washington, W.M. and G.A. Meehl, 1989. "Climate Sensitivity Due toIncreased CO2: Experiments with a Coupled Atmosphere and OceanGeneral Circulation Model", Climate Dyn., Vol. 4, pp. 1-38.

Weeks, W.F., 1976. "Sea Ice Conditions in the Arctic", AIDJEX Bulletin, No.34, Dec. 1976, pp. 173-205.

Wills, G., R. Bradley, J. Haslett and A. Unwin, 1990. "StatisticalExploration of Spatial Data", Proceedings, 4th International Symposium onSpatial Data Handling, Zurich Switzerland, July 23-27, pp. 491-500.

Zwally, H.J., J.C. Comiso, C.L. Parkinson, W.J. Campbell, F.D. Carsey and P.Gloersen, 1983a. Antarctic Sea Ice, 1973-1976: Passive MicrowaveObservations, Washington DC: National Aeronautics and SpaceAdministration, NASA SP-459, 206 pp.

Zwally, H. Jay, C.L. Parkinson, J.C. Comiso, 1983b. "Variability ofAntarctic Sea Ice and Changes in Carbon Dioxide", Science, Vol. 220, No.4601, pp. 1005-1012.

Figure 1: Minimum and Maximum Arctic Ice ExtentsAverages of February and September SMMR Ice Concen t r a t i onsbetween 1978-1987.

Figure 2: Principal Arctic Ice Drift Patterns

Table 1: Climatic Influence of Sea Ice

Climatic Property Physical Properties Climatic Influence

surface albedo extent, type,concentration

sea ice increases the surface albedo from 0.10-0.15over open water to 0.4-0.6 in the case of snow-free iceor 0.6-0.9 in the case of snow-covered ice

insulative properties concentration,thickness, extent,motion dynamics

sea ice inhibits exchanges of heat, mass andmomentum between the atmosphere and ocean

thermal capacitance type (age), seasonality,thickness

sea ice acts as a thermal reservoir which delays theseasonal temperature cycle

brine capacitance type (age), motiondynamics, thickness

sea ice alters the ocean's salinity distribution, bothvertically by the expulsion of brine during freezing andhorizontally by the large-scale transport of low-salinityice

Table 2: Sea Ice as a Climatic Indicator

Sea IceParameter

Observations Potential*Indicators

Ideal1Spatial

Requirements

Ideal1Temporal

Requirementsextent • has been shown to

influence weather and tobe influenced by weather

• extreme variability;oscillations on a 10-12year period noted; noconsistent trends found

+ Baffin Bay2,3

+ Labrador Sea2,4

+ Greenland Sea5,6

– Hudson Bay6

– Bering Sea6,7,8

– Sea of Okhotsk6,7,9

– Kara Sea3,6

>1 km 3 days

type • composition of multiyearice pack is a potentialrepository for accumulatedclimatic changes

• it has not been studied fortrends or variability

> 1 km 3 days

concentration • areas of low concentrationare significant areas foratmosphere-oceaninteraction

• high variability in themarginal ice zone; notrends found

+ Baffin Bay6

+ Bering Sea6

+ Labrador Sea6

+ Sea of Okhotsk6

+ Kara Sea6,10

> 1 km 3 days

thickness • relative thickness of seaice reflects average winterand early springtemperatures as well asthe magnitude of mid-summer solar radiation

– shorefast ice11 10 km 3 days

dynamics • motion changes can leadto changes in iceconcentration; long-termmotion is a potentialrepository for accumulatedclimatic changes

• high variability; sometimesreverses; no trends found

10 km 1 day

sesonality:melt onset

• closely related to winterand spring airtemperatures; long-termvariation is a potentialindicator of climaticchanges

• high variability; no longterm trend studies made

+ Chukchi Sea12

+ Kara Sea12,13

+ Barents Sea12,13

– Labrador Sea6,12

10 km 3 days

*potential regions for which consistent deviation from an established norm for that ice parameter might indicate achange of climate; '+': likely candidate; '-': region to avoid1Barber et al., 1992b 2Chapman and Walsh, 1992 3Barry, 19934Ikeda, 1990a,b 5Kelly et al., 1987 6Parkinson, 19917Cavalieri and Parkinson, 1987 8Niebauer, 1983 9Parkinson, 198910Crane and Anderson, 1989 11Brown and Cote, 1992 12Anderson, 1987

13Kelly et al., 1982

Table 3: Significant Sea Ice Parameters Retrievable from PassiveMicrowave Data

A √ indicates the parameter is retrievable from the data or t h edata meet the specified requirement; an × indicates the p a r a m e t e ris not retrievable or requirements are not met. Spatial a n dtemporal requirements are derived from Barber et al., 1992b.

Sea Ice Parameter Ideal Spatial Requirements Ideal Temporal Requirements√ ice extent × >1 km √ 3 days√ ice type × > 1 km √ 3 days√ ice concentration × > 1 km √ 3 days× ice thickness × 10 km √ 3 days√ ice dynamics × 10 km √ 1 day√ seasonality × 10 km √ 3 days

Table 4: Change Detection TechniquesAdapted from Howarth and Wickware (1981), Lillesand and K i e f e r(1987), Singh (1989), Eastman and McKendry (1991), and Ey ton(1991).

Technique Synopsis Commentscolour mapping temporal composite display images are

created by mapping a different timeslice into a unique colour

simple; can simultaneously view up to3 time slices; limited to 3 time periods

colour transformations transform standard red-green-bluedisplay structure into another system,e.g., intensity-hue-saturation

produces an effective and controlledvisual presentation which may be moreclosely matched to the human visualsystem; limited to 3 time slices

image differencing subtract time 1 image from time 2image

simple; indicates absolute changemagnitude, not significance

image ratioing divide time 1 image by time 2 image simple; allows relative imagedifferences to be examined

image regression a linear regression is fit between allpixels at time 1 with those at time 2; a'predicted' image is created byprojecting time 1 pixels through theregression; the projected and time 2images are differenced and examinedfor change

accounts for temporal differences inimage means and variances due touneven atmospheric or illuminationeffects

acceleration mapping calculated as the first derivative of thedifference image

provides information on the rate ofchange between images

change vector analysis 2 images are created: the first showingthe Euclidean distance between pixels("change vector"); the other showingthe angle of this change vector

the change vector image gives anindication of the degree of change;the direction image can be used toexamine the nature of the change

backgroundsubtraction

low-pass filtered image is subtractedfrom original data

removes slowly varying ("no-change")background

post-classificationanalysis

comparison of independentlyproduced classifications;crossclassification maps the logicalAND of all possible class combinationsin the form of a confusion matrix

statistical measures of change areeasily derived from thecrossclassification matrix; difficult toproduce comparable classificationsbetween dates

principal componentsanalysis

measurement space is rotated tomaximize the inter-channel variance

allows for the abstraction of changefrom a time series as a whole; can beused to reduce the number ofchannels input into other techniques

hypertemporalclassification

all data from both dates are classifiedconcurrently

classification can be complex andproduce too many classes

Table 5: Selected Global Change Studies Based on PrincipalComponents Analysis

of Remote Sensing Imagery

Temperate Applications

Variable(s) Synopsis Referencenatural vegetation components from a sequence of 7 Landsat images

show seasonal variationsLodwick, 1979

different components generated from multispectraland multitemporal imagery reveal local and globalchanges

Byrne et al., 1980

PCA of monthly NDVI images within an annual cyclehighlights seasonal and regional patterns

Townshend et al.,1985

PCA of monthly GVI images within an annual cyclehighlights seasonal, regional, and other patterns

Tateishi & Kajiwara,1990

geology & naturalvegetation

higher order components provide information onalbedo differences and vegetative changes

Ingebritsen & Lyon,1985

agricultural & urbanland cover

standardized components provide more accurateinformation for change detection

Fung & LeDrew, 1987

Polar Applications

Variable(s) Synopsis Referenceinterannual ice extent dominant hemispheric and regional variations are

foundWalsh & Johnson,1979a

regional trends are examined Lemke et al., 1980the leading spatial component was used to divide theAntarctic region into regions where the magnitudes ofthis component were approximately the same

Ropelewski, 1983

PCA of TM and MLA Antarctic images show a decreasein blue ice area was observed over a 2 year time span

Orheim & Lucchitta,1990

atmosphere - iceinteractions

statistically significant correlations between thedominant modes of atmospheric and sea ice variabilityare found

Walsh & Johnson,1979b

air temperature, rather than atmospheric pressure,shows the strongest relationship to ice conditions

Crane et al., 1982;Crane, 1983

seasonal variations insea ice type

using principal components in an ice classificationalgorithm would not improve its accuracy; the first 2components generated from the 10 SMMR channelsare an efficient data set

Rothrock et al., 1988;Thomas & Rothrock,1989

"extended" PCA is used to categorize the spatial andtemporal patterns of surface change that occur in theseasonal sea ice zone during the spring/early summer

Crane & Anderson,1989