Satellite-based Estimates of Sea Ice Volume Flux: Applications to the Fram Strait Region

Contents

1 Introduction
1.1 Aims
1.2 Structure
1.3 Publications

2 Fundamentals: The Arctic Climate System, Instruments and Data
2.1 The Arctic Climate System
2.1.1 The Arctic
2.1.2 The Arctic Ocean
2.1.3 Sea Ice
2.2 Instruments and Data
2.2.1 ICESat/GLAS
2.2.2 AMSR-E
2.2.3 QuikSCAT/SeaWinds
2.2.4 SARData
2.2.5 Polar Stereographic Projection and Study Region

3 Sea Ice Concentration
3.1 Introduction
3.2 ARTIST Sea Ice (ASI) Algorithm
3.2.1 Weather Filters
3.2.2 ASI Results
3.3 Tie-point Sensitivity Analysis
3.4 Error Estimation
3.5 Comparison to Ship Based Observations
3.6 AMSR-E Ice Concentration Algorithm Intercomparison
3.7 2007 Arctic Sea Ice Minimum and AMSR-E Time Series
3.8 Sea Ice Concentration Discussion
3.9 Sea Ice Concentration Summary

4 Sea Ice Drift
4.1 IFREMER AMSR-E 89 GHz Sea Ice Drift Product
4.2 SAR Sea Ice Drift and Comparison to Buoy Ice Drift
4.3 Validation of AMSR-E Ice Drift with SAR Ice Drift Data
4.4 Sea Ice Drift Summary

5 Sea Ice Thickness
5.1 Sea Ice Freeboard
5.1.1 Geoid
5.1.2 Lowest-Level Elevation Method
5.1.3 Validation
5.1.4 Gridded Freeboard
5.1.5 Outlook: Freeboard
5.2 Conversion of Freeboard to Ice Thickness
5.2.1 QuikSCAT Multi-Year Sea Ice Concentration
5.2.2 Snow Thickness and Density
5.2.3 Sea Ice Thickness Maps 2003-2007
5.2.4 Comparison to Ice Thickness From Helicopter-Borne EM- Sounding
5.3 Sea Ice Thickness Conclusion

6 Sea Ice Volume Flux: Determination and Physical Interpretation
6.1 Sea Ice Volume Flux Calculation
6.2 Sea Ice Volume Flux Discussion Ill
6.3 Divergence of Sea Ice Volume Flux
6.4 Fram Strait Sea Ice Volume Flux
6.4.1 Fram Strait Sea Ice Volume Flux for ICESat periods 2003-2007
6.4.2 Monthly Fram Strait Sea Ice Volume Flux Time Series 2003-2007
6.5 Error Evaluation and Comparison to Alternative Methods
6.5.1 Comparison to Volume Flux Obtained Using QuikSCAT Ice Drift
6.5.2 Sensitivity Study
6.5.3 Volume Flux from ULS and ICESat Ice Thickness Mea­surements
6.6 Comparison to Model Data
6.7 Comparison to Oceanographic Measurements

7 Conclusion
7.1 Summary
7.2 Relevance
7.3 Outlook

A Appendix
A.l Unusable and Missing Data
A.2 Freeboard - SAR Comparison
A.3 Additionally Used ICESat Ice Thickness Data
A.4 Ice Volume Flux Through Transects
A.5 1990-2007 Fram Strait Sea Ice Volume Flux Data
A.6 NAOSIM and MIT Ice Volume Flux

Acronyms

List of Figures

List of Tables

Bibliography

Acknowledgements

Abstract

The sea ice export out of the Arctic Ocean through Fram Strait into the Green­land Sea is the single largest source of freshwater in the Nordic Seas and therefore of special importance for the hydrological cycle of the North Atlantic. On its way south, the exported sea ice melts and thereby modifies the stratification of the ocean surface mixed layer, which in turn influences oceanic deep convection and water mass transformation processes in the Nordic Seas and thus impact global ocean thermohaline circulation. The lack of spatial sea ice thickness informa­tion has been one of the weaknesses for previous existing methods to determine the sea ice export. In this study a new method to obtain the sea ice volume flux exclusively from satellite measurements is presented. Previous estimates of the sea ice volume flux relayed on ice draft measurements of a single Upward Looking Sonar (ULS) in the Greenland Sea. The GLAS laser altimeter onboard the ICESat satellite launched in 2003 offers for the first time the opportunity to obtain the spatial sea ice thickness distribution up to 86°N latitude. In this study a method to determine the sea ice freeboard from ICESat altimeter data is developed and applied to nine ICESat measurement periods between 2003 and 2007. Assuming hydrostatic balance and by utilization of further satellite, in situ and climatological data these sea ice freeboard measurements are converted to sea ice thickness maps of the Fram Strait region. The satellite-based ice thick­ness estimates are combined with sea ice area and sea ice drift, as retrieved from AMSR-E microwave radiometer measurements at 89 GHz, to obtain the sea ice volume flux. The errors of the input quantities and the final sea ice volume flux are assessed. Using this method the spatial sea ice volume flux distribution is obtained from satellite observations for the first time. The Fram Strait sea ice volume flux is further investigated by calculating a monthly sea ice volume flux time series between January 2003 and April 2007. Summer months have to be disregarded due to missing sea ice drift data. The sea ice volume flux shows large inter-annual and -seasonal variability. A mean monthly Fram Strait sea ice vol­ume flux of (248 ± 90) km3/month with respective minimum and maximum values of 112km3/month (May 2003) and 484km3/month (December 2004) was found. These satellite-based sea ice volume flux estimates from the years 2003 to 2007 are compared to previous sea ice volume flux estimates obtained for the period 1990 to 1999 and can be used as extension of these previous time series. Finally, a comparison of sea ice volume flux estimates from this study with oceanographic salinity measurements shows good coincidence of summer melting events. A com­ parison to model results reveals large differences in the lateral distribution of the sea ice volume flux. The presented method does not just allow, as previously, to determine the sea ice export through Fram Strait but has the potential to investigate and better understand the dynamics of sea ice volume changes north and south of Fram Strait.

Zusammenfassung

Der Export von Meereis aus dem Arktischen Ozean durch die Framstraße in die Grönlandsee stellt die größte Quelle von Süßwasser im Europäischen Nordmeer dar und ist daher von zentraler Bedeutung für den Süßwasserhaushalt des Nordat­lantiks. Auf dem Weg nach Süden schmilzt das exportierte Meereis und bestimmt so maßgeblich die oberflächennahe Schichtung der Wassermassen, die wiederum die ozeanische Tiefenkonvektion im Europäischen Nordmeer und dadurch auch die globale thermohaline Zirkulation beeinflusst. Einer der bisherigen Schwach­punkte bei der Bestimmung dieses Eisexports ist das Fehlen flächendeckender Beobachtungen der Meereisdicke. In dieser Studie wird ein neues Verfahren vor­gestellt, den Meereisvolumenfluss alleinig aus Satellitenbeobachtungen abzulei­ten. Bisher beruhten Abschätzungen des Eisvolumenflusses in puncto Eisdicke auf den Eistiefgangsmessungen eines einzelnen Sonars in der Grönlandsee. Mit den seit 2003 gemessenen Daten des Laseraltimeters GLAS auf dem Satelliten ICESat ist es erstmalig möglich, die flächenhafte Eisdickenverteilung bis zu einer geographischen Breite von 86°N zu erfassen. In dieser Arbeit wurde ein Verfahren zur Bestimmung des Eisfreibords aus ICESat Laseraltimeterdaten entwickelt und auf neun ICESat-Messperioden zwischen 2003 und 2007 angewendet. Unter An­nahme hydrostatischen Gleichgewichts und mit Hilfe von weiteren Satelliten-, vor Ort gemessenen und klimatologischen Daten werden aus diesen Eisfreibordmes­sungen Eisdickenkarten der Framstraßenregion erstellt. Diese Meereisdickendaten werden mit Satellitenmessungen der Eisbedeckung und Eisdrift zum Meereisvolu­menfluss kombiniert. Für die Bestimmung der Eisbedeckung und Eisdrift werden jeweils AMSR-E Mikrowellenradiometermessungen bei 89 GHz verwendet. Die Fehler der Eingangsdaten und des Meereisvolumenflusses werden abgeschätzt. Mit dieser Methode kann erstmals die flächenhafte Verteilung des Meereisvolu­menflusses aus Satellitendaten beobachtet werden. Der Meereistransport durch die Framstraße wird mit Hilfe einer monatlichen Zeitreihe zwischen Januar 2003 und April 2007 ausführlicher untersucht. Hierbei werden die Sommermonate auf­grund fehlender Eisdriftmessungen nicht berücksichtigt. Der Eisvolumenfluss un­terliegt großer jährlicher und zwischenjährlicher Variabilität. Der mittlere monat­liche Meereisvolumenfluss durch die Framstraße betrug (248 ± 90) km3/Monat und erreichte minimale und maximale Werten von 112km3/Monat (Mai 2003) und 484km3/Monat (Dezember 2004). Der erhaltene Meereisvolumenfluss der Jahre 2003 bis 2007 wird mit früheren Meereisvolumenflussbeobachtungen ver­glichen und kann als Verlängerung dieser früheren Zeitserie verwendet werden.

Ein Vergleich der Volumenflussabschätzungen dieser Studie mit ozeanographi- schen Salzgehaltsmessungen zeigt eine gute Übereinstimmung der sommerlichen Eisschmelzperioden. Ein Vergleich mit Modellergebnissen läßt große Unterschiede in der räumlichen Verteilung des Volumenflusses erkennen. Die vorgestellte Me­thode erlaubt nicht nur, wie bisher, die Bestimmung des Meereisvolumenexports durch die Framstraße, sondern bietet auch die Möglichkeit, die Dynamik von Meereisvolumenänderungen nördlich und südlich der Framstraße zu untersuchen und besser zu verstehen.

Chapter 1 Introduction

Arctic sea ice: Where does it come from? Where does it go? The most fun­damental answers to these questions were already given by Fridtjof Nansen in 1896. His vessel Fram, which entered the Arctic pack ice in the Laptev Sea near the New Siberian Islands in 1893, left the ice again in August 1896 in the Fram Strait after three years of ice drift (Nansen, 1897). Since then we know that the main transport of sea ice out of the Arctic Ocean is taking place via Fram Strait and that the source regions for this ice are as far away as the East Siberian Sea on the opposite side of the Arctic Ocean. Nansen also anticipated the importance of sea ice for the Earth’s climate system when he described sea ice ocean interactions (Nansen, 1902). However, an accurate knowledge of sea ice dynamics and “where the ice goes” still remains an open question and is also the main topic of this work. Sea ice was realized to be one of the key components of the climate system and its interaction with the ocean and atmosphere has not only local but global relevance (ACIA, 2004, 2005). Thus here the variability of the Arctic sea ice mass exchange with the Greenland Sea and the possibilities of regularly monitoring it are in the focus. Anyhow, times have changed since Nansen’s Fram drift. The 2007 Tara ice drift following Nansen’s trace as part of the International Polar Year (IPY) about 110 years after the Fram drift, took only about 15 months for the same distance in a by extent significantly decreased sea ice cover (Gaseará et al, 2008). While the Arctic by exploitation of modern technique is not as hostile, dangerous and lonesome anymore as during Nansen’s time, still the number of in situ measurements taken there is below the world average. Therefore, observations from space are of special importance.

In this study a technique to derive the sea ice volume transported out of the Arctic Ocean through Fram Strait entirely from satellite measurements is de­scribed. It is a multi-sensor study, where different data products from different satellites are combined. For the observation of the sea ice thickness a new method was developed. This is of special importance as before sea ice thickness could only be measured by in situ campaigns and moorings. Finally a time series of the ice volume transport through Fram Strait for 2003 to 2007 is presented. Moni­toring anomalies in the Fram Strait sea ice volume flux is of special importance, as they can influence watermass transformation processes in the Greenland Sea and further downstream in the Atlantic Ocean. With the presented technique the lateral distribution of the sea ice volume flux can be directly observed, which was not possible with previous measurement techniques. The retrieval of sea ice volume is demonstrated for the Fram strait region but the used method can be easily adapted to other regions or applied globally.

1.1 Aims

The main aims and questions addressed in this study can be described as follows:

- Development of an exclusively satellite based method to monitor the sea ice volume flux.
- What is the amount and variability of the Fram Strait sea ice volume flux between 2003 and 2007? How does the Fram Strait ice volume transport change inter-annually and inter-seasonally?
- Calculation of a monthly Fram Strait sea ice volume flux time series between January 2003 and April 2007. How large is the amount and variability of the sea ice volume flux during these years in comparison to measurements during the 1990s? Can our estimates be used as an extension of the former time series9
- Combination of different satellite datasets to monitor the spatial distri­bution of the sea ice volume flux. For this purpose sea ice thickness es­timates obtained from ICESat laser altimetry are combined with sea ice area and drift measurements obtained from satellite microwave radiometry (AMSR-E) to retrieve the spatial distribution of the sea ice volume flux.
- Validation of the used sea ice concentration, drift, and thickness datasets to assure their quality for the sea ice volume flux retrieval. Error assessment of these quantities and the sea ice volume flux.
- How does the satellite based sea ice volume flux compare to oceanographic measurements? To get further insight in the sea ice - ocean interactions the sea ice volume flux observations will be compared with in situ ocean salinities measurements obtained from a mooring in the Greenland Sea.
- How does satellite based and modeled sea ice volume fluxes compare? Our sea ice volume flux observations will be compared with results from two coupled sea ice - ocean models.

illustration not visible in this excerpt

Figure 1.1: Schematic flow diagram of how the quantities involved (green boxes) have to be combined to get the sea ice volume flux. In the lower part of each box the belonging chapter is listed.

How long ICESat will continue to operate is unsure, as its designed lifetime of three years with a five-year goal is already exceeded. But plans for ICESat-II are underway and the radar altimeter satellite CryoSat-2 is scheduled for launch in 2009. It is anticipated that the presented sea ice volume flux retrieval method can be easily adapted to CryoSat-2 measurements. Thus, there is good hope that the time series can be continued in future and will help to understand climate relevant processes.

1.2 Structure

This work is organized as follows:

First, in Chapter 2 an introduction to the Arctic climate system and the main processes relevant for this study are given. In the second part of Chapter 2 the used data and sensors are described. The main quantities to retrieve the sea ice volume flux are the sea ice concentration (area), sea ice drift, and sea ice thickness. These quantities are described one after the other in Chapters 3, 4, and 5. The volume flux finally is described in Chapter 6. Figure 1.1 shows a flow diagram how the quantities have to be combined and in which chapter they are described.

The sea ice concentration and drift are derived from passive microwave AMSR- E data using existing methods. The focus for these two quantities therefore lies on the evaluation of the qualfiy of the datasets by comparing them with reference data. This is a prerequisite to estimate the uncertainty of the final ice volume flux data. For the ice thickness a new method was developed to obtain the ice freeboard from ICESat laser altimeter measurements and afterwards convert them to ice thicknesses using additionally QuikSCAT radar backscatter data for sea ice type discrimination. Finally, all three datasets have to be combined to derive the sea ice volume flux. The meridional Fram Strait sea ice volume transport is calculated and compared with model data and oceanographic salinity measurements in the Greenland Sea. Holfort and Meincke (2005) state that “the measurements of liquid freshwater flux are of minor value if not the information on freshwater fluxes with the ice are available in parallel”. Here a first step in that direction is made. Finally, a conclusion and outlook is given in Chapter 7.

1.3 Publications

Parts of this thesis were published in the following journals and book:

- A first version of sea ice volume flux retrieval method presented in Chap­ters 5 and 6 and first results were published in:

Spreen, G., S. Kern, D. Stammer, R. Forsberg, and J. Haarpaintner (2006), Satellite-based Estimates of Sea Ice Volume Flux through Fram Strait, Ann. Giacici., 44, 321-328.

Spreen, G., S. Kern, and D. Stammer (2006), Utilization of Multiple Satel­lite Sensors to Estimate Sea Ice Volume Flux through Fram Strait, in Arctic sea ice thickness: past, present & future, Climate Change and Natural Hazards Series 10, voi. EUR 22416, edited by P. Wadhams and G. Amanatidis, chap. 16, pp. 176-192, European Commission, Brussels.

- Sea ice volume flux estimates using sea ice drift data retrieved from Quik- SCAT instead of AMSR-E measurements (Chapter 6, Section 6.5.1) were presented in:

Haarpaintner, J. and G. Spreen (2007), Use of Enhanced-Resolution QuikSCAT/SeaWinds Data for Operational Ice Services and Climate Research: Sea Ice Edge, Type, Concentration, and Drift, IEEE Trans. Geosci. Remote Sens., 45(10), 3131-3137.

- Parts of the AMSR-E 89 GHz sea ice concentration retrieval and validation (Chapter 3) were published in:

Spreen, G., L. Kaleschke, and G. Heygster (2008), Sea ice remote sens­ing using AMSR-E 89-GHz channels, J. Geophys. Res., 113, C02S03, doi:10.1029/ 2005JC003384.

Chapter 2 Fundamentals: The Arctic Climate System, Instruments and Data

In the first part (Section 2.1) of this chapter an introduction to the main climate components of the Arctic climate system important for this study are given. In the second part (Section 2.2) the used satellite sensors and datasets will be introduced.

2.1 The Arctic Climate System

2.1.1 The Arctic

Different definitions for the Arctic geographical coverage exist. The Arctic region can be defined by the July 10°C isotherm of the air temperature at the surface. The thereby defined area covers the complete Arctic Ocean including all marginal seas and the Greenland, Bering and Labrador Sea. On land Greenland and parts of Iceland, Canada, Alaska, and Russia are covered. The 10°C isotherm in large parts lies near the Arctic Circle at 66°33;N latitude, which can be taken as an alternative border definition for the Arctic. Figure 2.1 on the following page shows a geographical overview including topography and bathymetry of the Arctic and surrounding areas including geographical names used throughout this study. The inset shows in detail the concrete study region around Fram Strait and the Greenland Sea. The majority of the northern hemisphere cryosphere is located in the Arctic with its most prominent features, the Greenland Ice Sheet and the Arctic sea ice cover.

The Arctic is a “hot-spot” of the global climate change occurred during the last hundred years, which means the Arctic is one of the most responsive regions to climate change. The Intergovernmental Panel on Climate Change (IPCC) reported in its 4th assessment (IPCC, 2007) that the Arctic surface temperature increase of about 1.5°C was twice as high during the last century (1906 to 2005) as the global surface temperature increase (Trenberth et al, 2007). For western Canada, Alaska, and Siberia an even higher warming of 2-3°C during the last

illustration not visible in this excerpt

Figure 2.1: Overview of the Arctic (left) and the study region around Frani Strait (right) including topography, bathymetry and names of geographical places used. In the Fram Strait map the 500 and 2500 m depth isolines are marked in black.

50 years (1954-2003) was reported (ACIA, 2004, 2005). The Arctic sea ice cover decreased by about 3% per decade since 1978. The Greenland Ice Sheet has been shrinking with a rate of about 50 to 100Gt/yr (equivalent to about 0.2 ± 0.1 mm/yr sea level rise) at least for 1993 to 2003. Before that date estimates are uncertain (all from Lemke et al, 2007). The land to sea distribution in the Arctic is completely different from the Antarctic. The mediterranean Arctic Ocean is completely surrounded by land masses. The ground is frozen all year around (permafrost) down to depths of several hundred meters to kilometers, but the surface is not permanently covered by ice or snow and thus exhibits the darker soil with larger absorption to the sun light during summer. Therefore, the winter to summer temperature differences on land are large. All this leads to the larger temperature increase on land compared to the ocean during the last decades.

2.1.2 The Arctic Ocean

The Arctic Ocean is a mediterranean sea with a depth of more than 4000 m over large parts of the Amundsen basin. Enclosed are several shallow marginal seas with depths below 500 m. These are the Barents, Kara, Laptev, East Siberian, Chukchi, Beaufort, and Lincoln Sea (see Figure 2.1). The Arctic Ocean has only one deep passage, the Eram Strait, where the majority of water mass exchange with the world oceans takes place. The Fram Strait is approximately 440 km wide and 3000 m deep. Other connections are the Bering Strait, the Barents Sea and the Canadian Archipelago.

The circulation in the Arctic Ocean is dominated by thermohaline forcing. This is in contrast to the major ocean basins Pacific, Atlantic, and Indian Ocean, where most currents are wind driven and only modified by thermohaline effects. The input of brine and fresh water due to freezing and melting of sea ice, respec­tively, is one of the major components of the thermohaline forcing. The influence of melting and freezing of sea ice on the ocean is enhanced by the fact that in the cold arctic regions changes of the salinity of seawater have a larger effect on the seawater density than they would have in warmer regions. Figure 2.2 on the following page shows the seawater density in dependence of its temper­ature and salinity. For example at — ГС a change of salinity from 32 to 34psu causes a density change of 1.6 kg/m3, while at 20°C for the same salinity change the density would only change by 1.5 kg/m3. This is only a slight difference but nevertheless important for the role of sea ice in enhancing or hampering ocean convection due to ice formation or melting, respectively. As can be seen from Fig­ure 2.2 on the next page for low temperatures the water density is getting almost independent of the temperature. Further cooling is not increasing the density anymore. Therefore, changes in salinity are the main driver for density changes at low temperatures. Figure 2.2 on the following page also shows the freezing temperature and line of maximum water density in dependence of temperature

illustration not visible in this excerpt

Figure 2.2: Seawater density p as a function of temperature and salinity. A value of 1000 kg/m3 has to be added to the annotations of the colored density isolines to get the full density value. The black dashed line is showing the sea water freezing point and in red the maximum density line is marked. For salinities above 24.7psu the maximum density lies below the freezing point. Calculation of densities are based on Fofonoff and Millard Jr. (1983) using surface pressure.

and salinity. For salinities below 24.7 psu the maximum density is laying above the freezing point. This is also the separation point between brackish water and true sea water. Sea ice can be formed more easily under brackish conditions, as water near the freezing point stays on top the denser bur warmer water. Brackish water in the Arctic only exists near the coast in the large river outflow regions in the Kara, Laptev and East Siberian Sea. In the rest of the Arctic first the upper water layer has to be completely cooled near the freezing point before ice formation can start (see also next Section 2.1.3).

Figure 2.3 on page 10 shows the main ocean currents in the Arctic together with the two main sea ice drift patterns, the Transpolar Drift and the Beaufort Gyre. Warmer Atlantic water (dark red arrows) enters the Arctic Ocean in the West Spitzbergen Current via Fram Strait. There it recirculates following the bathymetry in the main oceanic basins, the Nansen, Amundsen, Makarov and Canada basin. Another branch of Atlantic water enters the Arctic Ocean through the Barents Sea and Kara Sea and some Pacific waters enter via Bering Strait. In the Arctic Ocean the Atlantic water gets colder and partly sinks down (especially in ice growth regions due to additional brine input). Finally, it leaves the Arctic Ocean again via Fram Strait into the Greenland Sea. It is now colder and fresher than the surrounding waters and the main transport takes places in the upper 500 m along the shelf break and called East Greenland Current (EGC). A smaller part is also leaving via the Canadian Archipelago and then through Baffin Bay and Davis Strait. These cold waters flowing out of the Arctic are of crucial importance for the global thermohaline circulation. If the arctic water in the Greenland Sea is getting heavier by cooling from the Atmosphere and brine input from ice formation, it can sink down driven by convection and after passing the Denmark Strait overflow feed in the dense waters of the Atlantic Ocean. But on the other hand melting of sea ice can hamper this deep convection and thereby is one of the major processes which can alter the deep convection in the Greenland Sea.

The Arctic Ocean seems to be in a transition phase to a warmer state mainly due to a change of strength and location of the subpolar gyre (Polyakov et al, 2005; Polyakov, 2007). Especially, the Atlantic inflow in the upper layers is getting warmer. A part of the recent decrease of sea ice can be attributed to this ocean warming. But it is most likely contributing only to a small part to the sea ice shrinkage, as the sea ice is thermally insulated by the cold halocline layer to the warmer Atlantic waters. The cold halocline layer is formed by surface waters and waters from the large rivers flowing into the marginal seas of the Arctic Ocean. There the cold water is enriched with salt released from ice formation and thus is incorporated between the fresh surface layer and the even saltier but warmer Atlantic waters. The heat exchange between the Atlantic water and the sea ice thus is hampered by the cold halocline layer. The existence of this layer is one of the major differences between the Arctic and Antarctic sea ice to ocean interactions. On the other hand, also the upper ocean heat content is increasing during recent years due to enhanced summer short wave radiation in combination with larger open water areas during summer (Steele et al, 2008; Kay et al, 2008). This warming is causing in particular enhanced lateral melting of sea ice and can result in a positive feedback mechanism (“sea ice-albedo-feedback”, see below).

2.1.3 Sea Ice

A Short Introduction to Sea Ice Formation and Growth

Sea ice is formed by freezing of sea water. Its characteristics are due to the salt in the sea water, which influences the freezing process, and are significantly different from fresh water ice. The different forms of sea ice are defined in WMO (1989) by the World Meteorological Organisation (WMO). Table 2.1 on page 11 summarizes the most common sea ice types. Frazil ice is the first form of ice which forms when the temperature of the upper ocean drops below its freezing point, which is about —1.8 to —1.9°C. To achieve this, the upper mixed ocean layer has to be cooled, because in contrast to fresh water the maximum density

illustration not visible in this excerpt

Figure 2.3: Scheme of the main sea ice drift pattern (orange arrows) and cir­culation in the Arctic Ocean and Greenland Sea. Atlantic waters are shown in dark red (warm) and gray, Pacific inflow via Bering Strait is shown in blue. The typical extent of the summer and winter sea ice cover of recent years are shown as black dashed and dotted lines, respectively. The depth of 500m is marked as black isoline.

Table 2.1: Sea ice types for different stages of development after WMO (1989). Only the most common sea ice types are listed. of sea water with a salinity above 24.7 psu lies below the freezing point (see Figure 2.2 on page 8). Cooling from the atmosphere thus is causing convection in the upper water layer, which first has to be completely mixed and cooled before freezing can start. Frazil ice consists of small ice needles and plates suspended in the water. The immediate salt release to the ocean, when the first ice crystals are formed, lowers the freezing point of the surrounding water and increases the water density. Due to this, convection of the upper water layer including the frazil ice starts. Even under quiet ocean conditions this is hampering the growth of a solid, crystalline ice layer, like on lakes. If the freezing from on top continues and more and more frazil ice is produced, the ice needles and plates coagulate to a soupy layer called grease ice. Grease ice is damping the small scale surface waves and is reflecting less light, giving the sea surface a matt appearance. The next form of ice growth depends on the ocean swell. Under quiet conditions like in leads (opening of a fracture in the sea ice cover) a thin, closed ice cover called nilas forms. Nilas damps smallscale surface waves (centimeter scale) but is still elastic enough to follow longer scale waves. If stronger ocean swell is prevailing, small ice floes (30 cm to 3 m diameter) with raised rim, called pancake ice, form from grease ice or nilas. Under continued freezing conditions a solid ice cover forms and the sea ice can grow thermodynamically up to a thickness of up to 2 m during the first winter, but growth to a thickness of about 1 m during the first winter is more common in the Arctic. Sea ice which has not survived one summer melt is called first-year ice. After the first summer it is called residual first-year ice until 31 December of that year. The next year it is called second-year ice and all following years multi-year ice. These are the formal definitions by the WMO (WMO, 1989). For sea ice remote sensing beside young and new ice only two ice classes are commonly used: first-year ice and multi-year ice sometimes also

illustration not visible in this excerpt

Figure 2.4: Schematic diagram of first-year and multi-year ice (after Comiso (1983)). Typical values for some ice characteristic parameters are given. TQ is the emitted brightness temperature in the microwave spectrum, the optical path length of a laser altimeter is also marked. called perennial ice. All sea ice, which survived one summer melt is directly called multi-year ice (according to the WMO definition it should be called old ice). The flushing of the sea ice and drainage of brine out of the ice during summer melt is causing a change of the radiometric properties of the sea ice, which can be distinguished by remote sensing. Also in this study only the two classes first-year and multi-year ice are distinguished. More classes of thick ice can not be reliably separated by spaceborne radiometry and the thickness of new ice lies below the error margins of laser altimeter measurements from space.

During ice formation most salt from the sea water is directly released to the ocean, but some salt is incorporated as brine in pockets in between the ice struc­ture. The older the ice gets the less saline it is. During the aging of the ice more and more brine pockets get connected and thereby brine channels are building, through which the brine is draining out of the ice. This is caused mainly by gravity but also other mechanisms exist (Wadhams, 2000). After the first heavy brine release during the first days of ice growth the salinity of the ice is decreasing only slowly until the next summer. Then melt water from on top of the ice can flush through the brine channels and remove most of the brine. Thus multi-year ice is less saline than first-year and young ice, particularly in its top few ten cen­timeters. The strength properties of sea ice are controlled by the brine volume. Therefore, the ice after surviving one summer has a greater strength than before ( Wadhams, 2000). Figure 2.4 shows a scheme of typical first-year and multi-year ice together with their most important parameters.

During winter, snow can accumulate on sea ice. Thus the maximum in snow depth on Arctic sea ice occurs at the end of winter in May and June. During July and August the snow melts almost completely or transforms to slush on top the sea ice. The snow depth is not evenly distributed throughout the Arctic. Largest snow depths appear north of Greenland with a snow depth up to 45cm during winter. From there the snow depth gradually decreases towards the Eurasian marginal seas and the Chukchi Sea, where the maximum snow depth is about 25cm ( Warren et al., 1999). In general the Arctic Ocean is a cold desert with little snow fall compared to the Antarctic.

After an ice cover has been formed, it is moved around mainly by the drag exposed to the wind, especially on shorter time scales as hours and days. On longer time scales also the ocean currents below the ice are affecting the ice movement, as they are more steady then the atmospheric pressure. Inbetween these time scales lies the forcing by tides, which is larger in the marginal seas than in the central Arctic Ocean. In most areas the semidiurnal tides (М2 and S2) dominate the forcing but there are also places like the Yermak Plateau north west of Svalbard where the diurnal tides (Ki and Oi) play an important role (.Kowalik and Proshutinsky, 1994).

If these outer forces cause a convergent ice drift situation, the ice gets deformed. Under the outer pressure the ice gets piled up, parts of the floes get pushed upward (and downward) and thereby ridges are formed. This is the dynamic way of ice thickness increase. Thermodynamical sea ice growth general leads to ice not thicker than 2 m ( Wadhams, 2000). Therefore the thick old ice in the western part of the Arctic Ocean with thicknesses of 5 m and more is formed by a combination of ridgin of ice floes due to dynamic forcing and thermodynamical ice growth. Ridges strongly increase the drag coefficient of the ice and it is “sailing” more efficiently with the wind. Under divergent drift conditions cracks and leads in the ice are opened.

Continuous offshore blowing winds, e.g. katabatic winds along the coast, can open up large ice free areas, called polynyas. Alternative to these wind driven latent heat polynyas, polynyas can also be kept ice free by continued source of heat from the ocean below and then are called sensible heat polynyas. Sensible heat polynyas are less common. In the Northern Hemisphere the Kashevarov Bank polynya in the Sea of Okhotsk and the Whaler’s Bay polynya north of Svalbard are mainly sensible heat driven. Many polynyas like the North Water polynya in the northern Baffin Bay are maintained by a combination of both processes (Martin, 2001; Morales Maqueda et al, 2004). Under freezing conditions latent heat polynyas are areas of continuous ice growth, they are therefore sometimes called “ice factories”. Polynyas and leads are also important for the ocean to atmosphere heat flux. For low surface temperatures during winter the heat flux from the about —1.8°C warm ocean to the atmosphere in leads and polynyas can reach 1000W/m2, while already a 0.2 m thick ice cover reduces the heat flux by one dimension to less than 200W/m2 (Martin et al, 2004).

Sea Ice in the Climate System

Up to 7% of the worlds oceans are covered by sea ice ( Cavalieri and St. Germain, 1995). In winter the extent of Arctic sea ice can reach up to 16 · 106km2 and in summer it is reduced to about 7 · 106km2, but in recent years the maximum coverage was reduced to about 14 · 106km2 and the minimum coverage during summer to about 4 to 5 · 106km2. The amplitude of the seasonal cycle of the Antarctic ice cover is larger with 18 · 106 and 3 · 106 km2 for the respective winter and summer ice extent. The sea ice area makes up more than one quarter of the surface of the cryosphere and together with the snow on land has the highest variability in its extent.

In contrast to the surrounding water, sea ice has a high mean albedo of about 0.7 to 0.8. A new snow cover can even increase the albedo to more than 0.9, while melt ponds during summer and soot can reduce the ice albedo down to 0.2 (Gloersen et al, 1992; Hansen and Nazarenko, 2004). The mean summer albedo of the Arctic Ocean is about 0.5 (Laine, 2004). Water has an albedo of 0.04 to 0.15 and thus is absorbing most of the incoming shortwave radiation. The ratio between the sea ice cover and open water fraction is therefore ruling the radiation budget of the high latitude seas. Changes of this ratio cause the positive sea ice-albedo-feedback to come into account. For example if the sea ice cover decreases like during the last decades in the Arctic the amount of energy absorbed by the ocean is increasing. The upper ocean temperature consequently is increasing, which hampers sea ice formation, leading again to a reduction of the ice cover. The ice-albedo-feedback likely has supported the above-average strong warming of the Arctic compared to the mean global warming of the recent decades (ACIA, 2004, 2005). Satellite measurements of the global ice cover are therefore an important contribution to global climate monitoring.

It became evident from this, now about 35 year long time series of the satellite sea ice extent measurement that the sea ice area of the Arctic Ocean is currently shrinking (e.g. Comiso et al, 2008; Serreze et al, 2007; Stroeve et al, 2005; Cavalieri et al, 2003). The trend in sea ice extent reduction amounts to —3.7% per decade for the years 1978 to 2007, but increased to —10.1%/decade for 1996 to 2007 while during the period 1978 to 1996 the trend was only —2.2%/decade (Comiso et al, 2008). The reduction of the perennial ice cover (multi-year ice) is even stronger, as additional to the thermodynamical melting large fractions of old ice were transported out of the Arctic Ocean via Fram Strait (Nghiem et al, 2007; Comiso, 2002). The Arctic ice cover is expected to further decrease in response to accelerated Arctic climate warming (e.g. IPCC, 2007; Holland et al, 2006; Johannessen et al, 2004).

Our knowledge about the sea ice thickness in the Arctic is much poorer be­cause in situ data are very sparse. Therefore, satellite observations are very crit­ical for obtaining information about this quantity. Nevertheless, all available ice thickness measurements from submarines, drilling, and electromagnetic sound­ing reveal that the Arctic sea ice has thinned substantially since the late 1950s (Rothrock et al, 2008, 2003, 1999; Wadhams and Davis, 2000; Haas, 2004a,b; Tacker et al, 2001). Together with the shrinking ice area during the same time period, this means a significant reduction of the Arctic sea ice volume. Whether this reduction will continue towards an ice free Arctic Ocean in the future, or whether the downward trend can be attributed to a multi-decadal oscillation (e.g. Divine and Dick, 2006), which will reverse into an upward trend in the fu­ture, has to be carefully monitored. There is no evidence or explanation for a possible recovery of the sea ice cover in the near future and many climate models predict a summer ice free Arctic Ocean before the end of this century (IPCC, 2007; Zhang and Walsh, 2006). However, in most climate models the sea ice area decline during the recent decades is not well represented and thus the Arctic might be ice free during summer even earlier (Stroeve et al, 2007).

Sea ice also might play an important role in triggering the transition between glacials and inter-glacials (Stott et al, 2007; Gildor and Tziperman, 2003, 2001). Besides the increased direct warming of the upper ocean due to the sea ice-albedo- feedback, a retreat of the sea ice cover enhances the ocean to atmosphere CO2 flux and such leads to atmospheric CO2 rise. If the sea ice cover decreases, Ekman transport in the new open ocean areas increases, which leads to a decrease in stratification and by this to a ventilation of the ocean. This causes an enhanced CO2 flux to the atmosphere, which as result is warming (greenhouse effect). These positive feedbacks are likely important for the fast transition between glacials to inter-glacials.

In the context of this study the dynamic sea ice processes relevant for the climate are of greater importance. The complete global sea ice volume amounts to about 0.05 · 106km3. This is small in comparison to the ice volume stored in ice sheets and shelfs of about 33 · 106km3. But due to its high dynamic and variability in comparison to ice sheets the sea ice volume has despite its small amount strong climate relevance. In general, sea ice formation and melting take place at different locations of the ocean. Most of the sea ice in the Arctic is formed in the Eurasian marginal seas (East Siberian, Laptev, Kara, and Barents Sea; see Figure 2.1 on page 6) and is then transported elsewhere, e.g. through Fram Strait into the Greenland Sea, where it melts (see Figure 2.3 on page 10). This triggers two important climate relevant processes:

Transport of latent heat Heat is released to the ocean when sea ice forms and absorbed from the ocean again when the ice melts at a different location. Therefore, sea ice transport is an energy transport.

Transport of fresh water When sea ice forms salt is released to the ocean sur­face waters and when it melts fresh water is released, respectively. Sea ice transport therefore is a fresh water transport.

Especially the second point can impact the larger scale oceanic circulation. The input of fresh or dense water to the ocean (corresponding to melting and forming of sea ice) enhances or weakens ocean stratification, respectively. For example, a positive anomaly of sea ice fresh water export out of the Arctic Ocean through Fram Strait causes a freshening of the surface waters in the Greenland Sea and this hampers convective overturning and water mass formation there. This in turn can result in significant changes in the export of dense water from the Nordic Seas (Greenland, Iceland, and Norwegian Sea) into the Atlantic Ocean and then impact the global ocean thermohaline circulation (Dickson et al, 1988, 2007; Karstensen et al, 2005). Also several modelling studies, e.g. Komuro and Hasumi (2007); Stössel et al. (1998); Hasumi and Suginohara (1995), suggest that sea ice transport affect the global thermohaline circulation. The largest known fresh water anomaly event in the North Atlantic during the last century was the "Great Salinity Anomaly" from the late 1960s to the early 1980s (Dickson et al, 1988), which probably was due to an increased ice transport through Fram Strait. An overview of the various ways how changes in the mass balance of Arctic sea ice influence the global climate is given in Bamber et al. (2004).

On short timescales the main driver for the ice drift is the atmosphere. Under free drift conditions, i.e. an open ice cover or divergent drift, the geostrophic wind accounts for more than 70% of the ice drift velocity variance, but on long-term scales (several months) the ice movement can be attributed half to the wind and half to the mean ocean circulation ( Thorndike and Colony, 1982). As a rule-of- thumb the focal ice velocity is 2% of the surface wind speed and tilted about Fw = 30° to the right of the surface wind direction in the northern hemisphere. This relationship was already observed by Nansen during the Fram drift (Nansen, 1902). The turning angle between the geostrophic wind and the surface wind θα is of similar amount but opposite to the turning angle between the surface wind and the ice drift θω. Thus in absence of ocean currents sea ice drifts almost parallel to the geostrophic wind (θο = &w — θα æ 10° to 15° for wind speeds > 5 m/s; Wadhams, 2000; Hibler, III and Flato, 1992). Sea ice reacts rather quickly to changes of the focal wind. After wind forcing starts sea ice reaches a steady drift state after about an hour (Hibler, III and Flato, 1992). When the wind starts sea ice is not moving straight but in oscillating inertial loops due to the Coriolis force (the inertial period is 12 hours at the pole). Due to these reasons the sea ice cover is strongly influenced by cyclones passing by the ice edge, what frequently happens in the Greenland Sea (Brümmer et al, 2000; Zhang et al, 2004). These cyclones cause regions of convergent and divergent sea ice drift along the ice edge and to a lesser extent also in the solid ice cover (Brümmer et al, 2003, 2008).

In the Arctic the mean field of sea ice motion shows two main patterns: the Beaufort Gyre and the Transpolar Drift. A typical location and dimension of both are shown in Figure 2.3 on page 10. The stream of ice, which originates from the Laptev and East Siberian Sea in the Eurasian part of the Arctic, and then crosses the Arctic Ocean near the North Pole and ends at Fram Strait is called the Transpolar Drift Stream. In contrast, the Beaufort Gyre is an anticyclonic circulation of ice typically covering the Beaufort Sea and parts of the central Arctic Ocean. These two drift regimes are not completely separated. Ice from the Beaufort Gyre can flow along the North coast of Greenland and incorporates multi-year ice into the Transpolar Drift Stream (see Figure 2.3 on page 10). Also ice from the Kara and Barents Sea can contribute to the Transpolar Drift. Sea ice which is caught by the Beaufort Gyre can recirculate there for up to 7 to 10 years and gets 5 to 7 m thick due to thermodynamical growth and deformation. Ice originating from the Eurasian part of the Arctic and being transported by the Transpolar Drift Stream is not getting older than 5 years but in general not older than 3 years. Due to the large outflow of old ice from the Canadian part of the Arctic during the recent years today only 10% of the perennial ice in the Arctic is five or more years old (Maslanik et al, 2007b).

The location, extent and strength of the Beaufort Gyre and Transpolar Drift change due to the variability of long-term atmospheric pressure patterns. It can be distinguished between a cyclonic and anticyclonic wind drift regime. These can be connected with the phase of the North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) (Kwok, 2000; Rigor et ai., 2002; Martin and Martin, 2006). During NAO— phases there is a well defined high pressure cell over the Beaufort Sea (anticyclonic phase), while during NAO+ phases this high pressure system is weakend and not showing closed isobars anymore (cyclonic phase), which reduces the strength of the Beaufort Gyre and restricts its location to the Beaufort Sea nearer to the Canadian coast. The Transpolar Drift as consequence of NAO+ is bended more into the direction of the North Pole. The cyclonic and anticyclonic regimes also change inter-seasonal with a more anticyclonic regime dominating during winter and the cyclonic during summer. Which of the two regimes is dominating has also influence on the source region of the ice which is exported through Fram Strait. During a cyclonic phase (NAO+, AO+) more thick ice from north of Greenland and the Canadian Archipelago is transported through Fram Strait and thus enhances the ice volume transport during those years. During the anticyclonic phase the ice is recirculated in the Beaufort Gyre and thus is getting older and thick. During the cyclonic regime the total Arctic sea ice mass is therefore reduced while during the anticyclonic more sea ice mass can build up, if air temperatures allow thermodynamical growth. Between the late 1980s and mid 1990s a strong positive AO phase caused an enhanced outflow of thick old ice from the Arctic. Since then the AO is in a more neutral phase but in spite of that the extent and thickness of old ice have continued to decrease. This shows that AO and NAO can not explain all of the existing ice drift and export variability. Regional atmospheric circulations and internal Arctic processes are of great importance for the variability of both thickness and extent of the Arctic ice cover (Moslanik et al, 2007a,b; Overland and Wang, 2005).

To conclude, sea ice is an important climate player in several aspects and Lenton et al. (2008) identified the Arctic sea ice as one of nine potential tipping elements in the Earth’s climate system and the one which already may have passed a tipping point due to its recent decline.

Fram Strait Sea Ice Volume Flux

Two main processes can be identified for a change in Arctic sea ice mass: a change in the net amount of sea ice production and in the export of sea ice out of the Arctic Ocean. The first process depends on the length of the freezing period, snow accumulation, ice production in polynyas and surface air temperatures. The second process is largely determined by the sea ice export through Fram Strait into the Greenland Sea, since export through Fram Strait is by far the largest portion of the total Arctic sea ice export. Currently the net annual sea ice volume exported through Fram Strait into the Nordic Seas amounts to about 10% of the total sea ice mass of the Arctic Ocean and is the single largest source of freshwater in the Nordic Seas (Dickson et al, 2007; Serreze et al, 2006; Aagaard and Carmack, 1989). Only the liquid freshwater flux through Fram Strait is of the same order. As explained before, interannual perturbations in the sea ice transport through Fram Strait can modify the major water mass formation processes in the Greenland Sea and further downstream with consequences for the deep water formation and global ocean circulation. The amount of the Fram Strait sea ice volume flux is determined by the sea ice thickness at the northern entrance of the Fram Strait and mainly the wind forcing. It was shown by Pfirman et al. (2004) that sea ice export through Fram Strait can occur in surge­like events, where large portions of the old, thick ice leave the Arctic Ocean. Depending on the strength and location of the Beaufort Gyre and the Transpolar Drift, it takes several years until ice of similar thickness has formed again (see previous Section).

While in sea ice model studies the ice volume or ice mass flux is one of the quantities of the most interest (e.g. Koenigk et al., 2006, 2007), it is difficult to get this flux from observations. Sea ice area, motion, thickness and ice density have to be known to derive the sea ice volume flux, and it is not possible to obtain all these quantities with any one measurement technique. A multi sensor approach has therefore been chosen by other groups (e.g. Kwok et al., 2004a) and is also chosen for this study, with the goal to utilize only satellite measure­ments. Existing estimates of the sea ice volume flux through Fram Strait (1950s to 1990s) range from 1600km3/year to 5000km3/year and show high interannual variability ( Vinje, 2001; Vinje et al., 1998). During 1991-1999, averaged trans­ports amount to (2218 ± 497) km3/year, with individual annual values ranging from 1792km3 (1998/99) to 3364km3 (1994/95) (Kwok et al, 2004a). From the three for the volume flux determination needed parameters sea ice area, motion, and thickness the first two are available on a daily basis (area) or every other day (motion), based on all-weather and daylight independent spaceborne passive (e.g. Special Sensor Microwave/Imager (SSM/I)) and/or active (e.g. SeaWinds on QuikSCAT) microwave sensors since late 1978 (e.g. Agnew et al, 1997; Kwok et al., 1998; Cavalieri et al., 2003). In this study data from the Advanced Mi­crowave Scanning Radiometer for EOS (AMSR-E) are used to obtain sea ice concentration and motion. For comparison also QuikSCAT data are used for ice motion determination.

In contrast, knowledge about the sea ice thickness was limited in the past to a few, sparsely distributed measurements, obtained, e.g. by drilling, moored ULS ( Vinje et al, 1998), submarine-based sonar (e.g. Wadhams, 2000; Rothrock et al, 1999), and ground-based or air-borne electromagnetic thickness sounding (Haas, 2004b,a). Furthermore, all these thickness measurements are obtained on a quite different spatial scale than the ice area and motion measurements. Previous ice volume transport estimates through Fram Strait were obtained primarily using data from moored ULS by extrapolating local thickness estimates across the entire Fram Strait to obtain a complete cross-strait ice thickness profile (e.g. Vinje et al, 1998).

In 2003 Laxon et al. (2003) obtained the first satellite-based estimate of the Arc­tic sea ice thickness distribution from spaceborne radar altimetry, although severe limitations apply concerning the covered area, the minimum observable ice thick­ness, and the temporal resolution. Progress was obtained in ice thickness obser­vations after the launch of the Ice, Cloud, and land Elevation Satellite (ICESat) in 2003. ICESaťs Geoscience Laser Altimeter System (GLAS) is the first space­borne instrument, which at least comes close to the needed spatial and temporal resolutions needed to monitor the sea ice thickness globally. GLAS is measuring its height above the Earth’s surface, from which the Sea Surface Height (SSH) and the sea ice freeboard height can be inferred. Several only recently published studies (Kwok et al, 2004b, 2006, 2007; Spreen et al, 2006; Zwally et al, 2008) provide first estimates of the sea ice freeboard and thickness distribution obtained from ICESat data for the Arctic and Antarctic. Key problems for all these studies are i) inaccurate SSH estimates, ii) unknown snow depth and ice density, which are needed to convert freeboard to ice thickness, iii) contamination by clouds and due to this low data coverage and/or larger errors of the altimeter measurements. Therefore, best results are expected to be obtained in regions with a stationary sea ice cover, which permits averaging over long/large periods/areas. These con­ditions are not met in the Fram Strait/Greenland Sea: sea ice is known to drift several kilometres per day, divergence and convergence can continuously change surface roughness, and snow accumulation can be very variable. Therefore care­ful error estimates are crucial to determine the reliability and accuracy of the estimated quantities - like sea ice freeboard height and volume flux. However,

illustration not visible in this excerpt

Figure 2.5: Artist’s rendering of the ICESat satellite with GLAS transmitting a laser pulse towards the earth surface (courtesy NASA). as pointed out before, the sea ice volume flux through Fram Strait is the most important ice export of the Arctic Ocean and can have large influence on the ocean circulation and thus the climate system. Continual monitoring of Fram Strait sea ice volume flux is therefore of particular importance.

2.2 Instruments and Data

After the introduction to the different aspects of the Arctic climate system impor­tant for this study given in the last section, in this section a short introduction to the satellites sensors from which measurements arc utilized in this study is given. Namely these arc ICESat/GLAS (Section 2.2.1), AMSR-E (Section 2.2.2), QuikSCAT (Section 2.2.3), ASAR, and RADARSAT (both Section 2.2.4). The order of the satellite sections reflects their importance for this study. Also the used data products are shortly introduced. More detailed descriptions of the used algorithms and measurement principles are given in the accordant chapters where the data are applied. The last section of this chapter (Section 2.2.5) describes the study region and the map projection used consistently for all datasets.

2.2.1 ICESat/GLAS

The Ice, Cloud, and land Elevation Satellite (ICESat) is the first satellite mission which at least comes close to the needed spatial and temporal resolutions needed to monitor the sea ice thickness globally. The primary purpose of ICESat is to determine inter-annual and long-term changes in volume of the polar ice- sheets, mainly Antarctica and Greenland, and the influence of this changes on the global sea level. The possibility to determine also the sea ice freeboard, as it is done in this study, was already mentioned in the pre-launch studies but only as a secondary aspect, which feasibility was not guaranteed (Zwally et al., 2002). Another ICESat application are atmospheric measurements of cloud properties

illustration not visible in this excerpt

Table 2.2: The main parameters of the ICESat satellite and the GLAS instru­ment.

and aerosols height profiles. An artist view of ICESat in space is shown in Figure 2.5.

ICESat was launched on 12 January 2003 by the National Aeronautics and Space Administration (NASA) and carries one main instrument: the Geoscience Laser Altimeter System (GLAS). ICESat is operating in an orbit with 600km altitude and 94° inclination. This orbit configuration allows altimeter measure­ments up to 86° North and South. GLAS has three lasers, with only one operating at a time. Each laser produces an 1064 nm and 532 nm wavelength pulse. The 1064 nm pulses are mainly used for altimetry. The 532 nm data are used for the atmospheric products using the Light Detection And Ranging (LIDAR) princi­ple. In this study only 1064 nm altimetry data is used. The footprint size of the laser beam is about 64 m on the Earth surface. GLAS is transmitting 40 laser pulses per second (40 Hz), which results in a sampling distance of about 172 m on the Earth surface. Table 2.2 is summarizing the main ICESat and GLAS parameters.

The returned laser pulse is captured by a 1 m diameter telescope and the re­ceived power spectrum is digitized by 1 GHz sampler. These digitized pulses are referred to as laser waveforms and have a Gaussian shape for flat surfaces. The

illustration not visible in this excerpt

Table 2.3: Dates of all ICESat measurement periods and indication if the mea­sured data was used for this study.

waveform is now tracked for its maximum (or maxima in case of rough surfaces, different tracking algorithms are used for different surface types). From the trav­eling time At for the identified maximum and the speed of light c the distance Dlaser = cAt/2 between laser and Earth surface is calculated (see also Figure 5.1 on page 74).

To get from the measured distance Diaser to a surface elevation measurement the exact position of the laser and thus the position of the satellite in space has to be known. For ICESat this Precision Orbit Determination (POD) is done by a Global Positioning System (GPS) tracking system with a radial accuracy < 5 cm.

ICESat was designed to operate continuously for three to five years. Each laser had an expected lifetime of about two years to achieve this goal. Unfortunately the first laser already failed on 29 March 2003 after 37 days of operation. The degradation of the laser pump diodes was much faster than expected due to an improper material usage in manufacture. To obtain a reasonable long time series and to fullfill the mission lifetime the GLAS operating plan was modified. The planed 183 day repeat orbit was changed to a 91 day one and GLAS is operated now three times a year during an identical 33 day subcycle. Table 2.3 gives the names and dates for all ICESat measurement periods acquired so far. Thus ICESat now is operating longer than its nominal lifetime and how many more measurement periods can be obtained is unknown. Nevertheless, despite the time gaps the ICESat measurement time series now has achieved a length which allows first interpretations of the variability of the obtained geophysical parameters like

illustration not visible in this excerpt

Figure 2.6: The masks for the different, surface types used for ICESat data. Areas with different colors denote different surface classes or combinations of surface classes as described by the legend 011 the right.

sea ice thickness.

Further insight in the GLAS’s measurement principle and an overview of the ICESat mission are given in Schutz et al. (2005); Schutz (2002). and Zwally et al. (2002).

GLAS Data Product

For this study the “GLAS/ICESat L2 Sea Ice Altimetry Data” product (GLA13, Zwally et al. (2003)) in version 28 is used. It contains all ICESat measurements of potentially ice covered regions and GLAS range measurements calculated with a specially for sea ice adapted range offset. Each GLA13 dataset contains data of 14 ICESat orbits a 97 minutes, thus in total approximately 22.5 hours. Figure 2.6 shows the surface mask used for the ICESat data in the Arctic. The four surface types land, sea ice, ocean, and ice sheet and their combinations are defined. The GLA13 sea ice dataset contains all GLAS measurements falling inside any of the sea ice masks (light gray, turquoise, and purple-gray areas in Figure 2.6). For the sea ice freeboard and thickness calculations in this study only data from the “sea ice + ocean mask” (light gray area) are used.

The range D¡aser (see Figure 5.1 on page 74) for every shot is calculated after Schutz (2002) by

illustration not visible in this excerpt

with refRng as the reference range calculated from the laser pulse runtime, siRngOff the range offset to be added using the algorithm deemed appropriate for sea ice, dTrop the dry troposphere delay range correction, and wTrop the wet troposphere range correction. The troposphere corrections are calculated using surface pressure, temperature, and water vapor interpolated to the laser footprint from the National Centers for Environmental Prediction (NCEP) Global Analysis.

The main data product (beside nearly 90 other data fields with ancillary in­formation) contained in the GLA13 data product is the sea ice surface elevation E defined as

illustration not visible in this excerpt

The ellipsoid height hemp is the distance between the GLAS and a reference earth ellipsoid, which in the ICESat case is the same ellipsoid as used for the TOPEX/Poseidon radar altimeter mission (equatorial radius = 6378136.3 m and flattening = 1/298.257) in a mean-tide system. Additionally the tides erElv of the solid earth and ocElv of the ocean and the elevation of ocean tidal loading IdElv ( Yi et al, 1999) have to be subtracted. The ocean tides ocElv are calculated with the CSR 3.0 global ocean tide model (Rettadpur and Eanes, 1994; Eanes and Rettadpur, 1995). This is already done for the GLA13 data product and the elevation E as defined in equation 2.2 can be used directly.

The 1064 nm altimetry detector and receiver are able to record returned laser pulses between 0.05 - 13fJ energy without distortion. Return echos from flat ice or water surfaces under clear atmosphere conditions can have higher energies than 13fJ at the GLAS detector (Abshire et al, 2005). In these cases the de­tector is getting saturated, leading to distorted waveforms that are clipped and artificially wide. The standard Gaussian fit processing is getting biased towards longer ranges Diaser for such waveforms. Fricker et al. (2005) provide a method to recalculate the waveform energy for detector saturation cases and show that this correction significantly improve ICESaťs range measurements in the Boli­vian salt flat salar de Uyuni. This saturation correction is provided as additional data field in the GLA13, v28 dataset. By default the correction is not applied to the elevation measurements, as some slightly saturated cases may not be detected and the correction has a cut-off value of 1.5 m for very strong detector satura­tions. Nevertheless, as both of these cases are seldom and also in these cases the saturation correction is not deteriorating the final elevations, we decided to apply the saturation correction satElevCorr to the elevation measurements for this study (Ecorr = E + satElevCorr).

Beside the GLAS elevation measurements also the surface reflectivity at the 1064 nm laser frequency can give usefull information to discriminate different surface types and constrain the valid GLAS measurements. The uncorrected reflectivity [illustration not visible in this excerpt] is calculated as the ratio of the received energy [illustration not visible in this excerpt] after it has been scaled for range, and the transmitted laser energy [illustration not visible in this excerpt]·

illustration not visible in this excerpt

The uncorrected reflectivity [illustration not visible in this excerpt] is contained in the GLA13 dataset for ev­ery GLAS measurement. Beside the surface properties, e.g. specular or diffuse (Lambertian) reflection, also the atmosphere influences the ration [illustration not visible in this excerpt] Therefore [illustration not visible in this excerpt] not always represents the surface reflectivity but a mixture of surface and atmosphere scattering. To obtain the surface reflectivity R the at­mospheric effects have to be corrected for:

illustration not visible in this excerpt

where tc is the cloud (column) integrated optical depth, ta is the aerosol (column) integrated optical depth, and tm is the molecular optical depth. The reflectivity R corrected for atmospheric effects is contained in the GLA13 ICESat dataset for every 40th measurement, i.e. in distances of about 7 km. In this study for the comparison to SAR data (Section 5.1.3 on page 84) only the uncorrected reflectivity Runcorr is used as it is available for every GLAS measurement.

Error Budget

Under ideal conditions (cloud free, no detector saturation) ICESat elevation ac­curacy was found to be < 2 cm and precision < 3 cm over the world largest salt flat the salar de Uyuni in Bolivia. The accuracy of ICESat measurements can be heavily derogated by detector saturation caused by high pulse return energy, by forward scattering from clouds, and by higher noise ratios for declining transmit­ted laser power. These effects may cause biases of up to 1 m for single overpasses (.Fricker et al, 2005). Some of the bias caused by detector saturation can be corrected for but nevertheless, for a single ICESat measurement the error can exceed 1 m.

However, the mean error is much smaller. In Table 2.4 on the next page the different error contributions to the root mean square (RMS) error of 13.8 cm for a single ICESat elevation measurement according to Zwally et al. (2002) are listed. These are theoretically error estimates calculated before the actual ICESat launch. As mentioned above from field comparison the real error can expected to be smaller. Nevertheless, for the error calculations done in this study the conservative error value of 13.8cm for a single shot ICESat measurement is used.

Table 2.4: Error budget after Zwally et al. (2002) for one ICESat elevation measurement from a single GLAS shot. Last line gives the root mean square (RMS) of all errors, which is the expected single measurement error assuming a Gaussian error distribution.

illustration not visible in this excerpt

Our expected sea ice freeboard values he in the decimeter range. Thus, even if we would find a perfect ice freeboard algorithm, the reported ICESat error margins would cause large uncertainties for the freeboard estimate from a single ICESat measurement. Therefore, already now it becomes clear, that averaging over several ICESat measurements will be needed to reduce the mean error of the ice freeboard estimates.

2.2.2 AMSR-E

In this study Advanced Microwave Scanning Radiometer for EOS (AMSR-E) data is used to determine sea ice concentration (Chapter 3) and sea ice drift (Chapter 4). Both quantities are together with the ice thickness prerequisites to obtain the sea ice volume flux. The main advantage of observations in the microwave spectrum in comparison to the visual spectrum is the independence of daylight and clouds.

Figure 2.7: Artist’s view of NASA’s Aqua satellite. The red ellipse marks the AMSR-E microwave radiometer (top: 1.6 m diameter parabolic reflec­tor, bottom: feedhorn unit) (courtesy NASA).

illustration not visible in this excerpt

Table 2.5: Main characteristics of the AMSR-E radiometer on board NASA’s AQUA satellite (JAXA, 2005). For the footprint size and sampling interval dis­tances are given as sampling versus flight direction. For the temperature resolu­tion the worse one of the two channels (horizontal and vertical) is given.

illustration not visible in this excerpt

The AMSR-E sensor measures the Earth’s electromagnetic emission at six dif­ferent frequencies between 6.9 and 89 GHz at both horizontal and vertical polar­ization. AMSR-E is mounted on NASA’s AQUA satellite. An artist’s view of the Aqua satellite including the AMSR-E sensor is shown in Figure 2.7. Aqua is flying in a sun synchronous, near-polar orbit with an inclination of 98.2°. The orbit altitude is 705 km and the circulation period 99 minutes. AMSR-E is a conical scanning radiometer with 6 feedhorns and a parabolic reflector of 1.6 m diameter. The AMSR-E swath width is about 1450 km and thus daily complete coverage of the Earth surface north and south of ±55° is achieved. Details of the AMSR-E characteristics can be found in Table 2.5.

Both the sea ice concentration and drift datasets used for this study exploit brightness temperatures obtained from the vertically and horizontally polarized 89 GHz channels. With a footprint seaice of about 5 km these channels offer todays highest spatial resolution for spaceborne microwave radiometry. The lower frequency channels are only involved as weather filters to detect spurious ice in the open ocean, and for validation purposes. The AMSR-E 89 GHz swath is composed of measurements from two feedhorns, whose footprint locations on the Earth surface is shifted by about 5 km to each other. On 4 November 2004 the 89GHz feedhorn A (see Table 2.5) failed, what reduced the sampling resolution in flight direction from 5 to 10 km. However, for the polar regions, which are covered by several satellite overflights per day, daily brightness temperature maps with a resolution of about 5 km still can be constructed.

AMSR-E was developed by the Japan Aerospace Exploration Agency (JAXA). The National Snow and Ice Data Center (NSIDC) provides AMSR-E data in different processing levels. Level 3 daily gridded brightness temperatures on a 6.25 km polar stereographic grid (see Section 2.2.5) are used to calculate the sea ice concentration (Chapter 3). The sea ice drift is calculated from brightness temperatures on the same grid. Here brightness temperatures are first calculated by the method described in the “AMSR-E Data Users Handbook” (TAXA, 2005) from Level 1A swath raw observation counts, which are distributed from NSIDC within hours after acquisition. Afterwards all swath brightness temperatures of one day are interpolated onto a polar stereographic grid (see Section 2.2.5). From these grids sea ice drift is calculated at Institut français de recherche pour Vexploitation de la mer (IFREMER) (see Chapter 4).

2.2.3 QuikSCAT/SeaWinds

Data from the Quick Scatterometer Mission (QuikSCAT) satellite are used for two purposes in this study: (1) to obtain the multi-year sea ice fraction (Section 5.2.1 on page 96) and (2) as comparison dataset for the sea ice drift (Section 6.5.1 on page 131).

The SeaWinds instrument on QuikSCAT, launched in June 1999, is an active Ku-band dual-polarized scanning pencil-beam scatterometer. As SeaWinds is the only instrument on board QuikSCAT both will synonymously referred to as QuikSCAT from now on. QuikSCAT measures radar backscatter at 13.4 GHz in horizontal (HH) and vertical (W) polarization at incidence angles of 46°(1400 km swath-width) and 54°(1800km swath-width), respectively. The measurements have a footprint of 25 x 37 km, but sub-footprint range resolution is achiev­able and due to the rotating antenna each point on earth is covered twice for one overflight. All measurements of one day are averaged on a 25 km polar stereographic grid (see Section 2.2.5) for both polarizations covering the com­plete Arctic and Antarctic (Ezraty and Piollé, 2001). The gridded backscat­ter σο data for both VV and HH polarization are provided by Centre ERS dXrchivage et de Traitement (CERSAT) at IFREMER in Brest, France (http:// www.ifremer.fr/cersat/en/data/download/gridded/psiqscat.htm). From these daily backscatter maps the multi-year sea ice concentration is calculated with the algorithm described in Section 6.5.1 on page 131. The used QuikSCAT ice drift was calculated from enhanced-resolution (2.225 km) backscatter maps (.Haarpaintner, 2006) and the drift data were provided by Jörg Haarpaintner.

2.2.4 SAR Data

Due to its high spatial resolution Synthetic Aperture Radar (SAR) data is ideally suited for validation purposes of lower resolving satellite data. In Section 4.3 on page 68 sea ice drift obtained from SAR observations is used as reference for the AMSR-E sea ice drift. In Section 5.1.3 on page 84 ICESat sea ice freeboard heights are validated with SAR data. For both validation comparisons SAR data from two sensors are used: ASAR and RADARSAT. As all measurements in the microwave spectrum SAR observations are independent of daylight and clouds.

Envisat ASAR

The Envisat satellite operated by the European Space Agency (ESA) carries a variety of different sensors for environmental observations (in total nine instru­ments). One of those is the Advanced Synthetic Aperture Radar (ASAR). It is an imaging microwave radar operating at 5.331 GHz (C-band) in both vertically and horizontally polarization. The ASAR antenna has a size of 10 x 1.3m2. ASAR can acquire data in several different modes offering different spatial resolutions and swath widths. In this study only data from the Wide Swath Mode with a spatial resolution of 150 m and a swath width of 400 km is used. The radiometric resolution lies between 1.5 and 1.7dB. The incident angle varies between 15° and 45°.

RADARSAT

The second source used for SAR data is the RADARSAT satellite operated by the Canadian Space Agency (CSA). The onboard SAR is as the satellite called RADARSAT and is as ASAR measuring at 5.3GHz (C-band). In contrast to ASAR RADARSAT is only operating at horizontal polarization for both sending and receiving (HH). The antenna size is 15 x 1.5 m2. RADARSAT is operating in different acquisition modes. For this study only data from the ScanSAR mode with a spatial resolution of 100 m and a scene size of 500 x 500 km2 is used. The incident angle varies between 20° and 49°.

2.2.5 Polar Stereographic Projection and Study Region

All data are mapped onto a grid using the polar stereographic projection used by the NSIDC for several sea ice products (http://nsidc.org/data/grids/ ps_grid.html). As only difference for the sea ice thickness and volume flux datasets the World Geodetic System 1984 (WGS84) ellipsoid is used instead of the Hughes ellipsoid used by NSIDC (Snyder, 1987). For our grid size of 25km differences introduced by this are negligible. The polar stereographic projection is a conformal azimuthal map projection, thus preserving correct projected angles but not the area of all grid cells. The latitude of true scale is set to 70° N. With this projection the distortion in area at the North Pole is about 3% and therefore can be neglected. Different grid resolutions are used for the different datasets but either the grid resolution is 25km like for the final sea ice volume flux or is an integer factor of this 25km (e.g. 12.5 or 6.25 km) to guarantee easy data conversion. The region used for this study is covering the Fram Strait, Greenland Sea, and a part of the Arctic Ocean. The corner coordinates are upper-left: 89.54N/—135.00E, upper-right: 73.49N/45.00E, lower-left: 63.68N/-45.99E, and lower-right: 59.22 N/—13.17 E. The complete study region, for which all calculations are done, is for example shown in Figure 5.2 on page 77. Most Figures in this thesis only show the northern, most interesting part of that region (e.g. Figure 6.2 on page 112).

Chapter 3 Sea Ice Concentration

3.1 Introduction

As shown in Figure 1.1 on page 3, the sea ice concentration or respectively the sea ice area is the first quantity to be derived in order to obtain the sea ice volume flux. The sea ice concentration C is defined as the percentage of a given area covered with sea ice. In our case sea ice concentrations are calculated on a grid with 6.25 km x 6.25 km grid cell size. For every grid cell C defines the ice covered percentage of this area of 39.0625 km2. The rest of the area ([1 — C]39.0625 km2) consists of open water.

In this study we calculate sea ice concentration data from AMSR-E data using the ARTIST Sea Ice algorithm. It is an enhancement of the sea ice concentration algorithm described by Svendsen et al. (1987) for near 90 GHz satellite radiometer data. Within the framework of the Arctic Radiation and Turbulence Interaction STudy (ARTIST), this algorithm was evaluated for SSM/I 85GHz data and mod­ified to become the ARTIST Sea Ice (ASI) algorithm (Kaleschke et al., 2001). In Spreen (2004) the algorithm was adapted to AMSR-E data. Furthermore, the weather filters were refined, an automatic tie-point adaption scheme was intro­duced, and the ASI data were compared to an ice edge detection algorithm. In this study ASI ice concentrations are further validated to assure the quality of the data and thereby the usefulness for the sea ice volume flux retrieval. Parts of the results presented in this chapter are also published in Spreen et al. (2008), but the analysis here goes beyond that.

After a general description of the algorithm in Section 3.2, a tie-point sensi­tivity analysis is carried out (Section 3.3) and some error estimates are given (Section 3.4). Afterwards AMSR-E ASI data are compared to ship borne obser­vations (Section 3.5) and to two other AMSR-E sea ice concentration algorithms (Section 3.6). The chapter ends with a discussion of the minimum in sea ice coverage during summer 2007.

3.2 ARTIST Sea Ice (ASI) Algorithm

Sea ice concentration has been retrieved by passive microwave sensors since the launch of the Electrically Scanning Microwave Radiometer (ESMR) in December 1972. Since 1987 the SSM/I has been widely used for sea ice concentration determination. A restriction of these instruments is the coarse spatial resolution (approximately 50 km) of the data. In 1992 the 85GHz channels of SSM/I with a higher spatial resolution of about 15 km have become available.

In 2002 two new and similar microwave radiometers were launched. AMSR-E in May on the AQUA platform and AMSR in October on the MIDORI-II (for­merly ADEOS-II) satellite. Control over MIDORI-II was lost in October 2003. Therefore, only AMSR-E data is used in this study (see also Section 2.2.2 on page 26).

The main advantage of AMSR-E in comparison to SSM/I consists in its im­proved spatial resolution. For the 89 GHz channels the resolution is improved by factor of three relative to the SSM/I 85 GHz channels (SSM/I footprint size: 13 x 15 km2, AMSR-E footprint size: 4x6 km2). Thereby the elliptical footprint area is reduced from 153 km2 to 19 km2. The spatial resolution of ice concen­tration derived using the widespread NASA-Team and Bootstrap sea ice concen­tration algorithms is restricted to the resolution of the involved channels with the coarsest resolution, i.e. the 19 GHz channels. They have a footprint size of 43 x 69 km2 for SSM/I and 16 x 27 km2 for AMSR-E. Thus, the sea ice concentra­tions presented here represent an improvement in linear spatial resolution of more than a factor of three compared to non-89 GHz AMSR-E based sea ice concen­tration, and an improvement of more than ten times compared to the resolution of the SSM/I-based ice concentration based on the 37 and 19 GHz channels.

The ice concentration is calculated by the value of the brightness tempera­ture polarization difference P (hereinafter polarization difference or P) of the brightness temperatures Tq measured by the radiometer,

illustration not visible in this excerpt

with V for vertical and H for horizontal polarization. It is known from surface measurements that the polarization difference of the emissivity near 90 GHz is similar for most ice types and much smaller than for open water (Figure 3.1).

This is also valid for the polarization difference P, as the physical temperature is identical for horizontally and vertically polarized brightness temperatures and thus only emissivity differences influence P. For the influence of the atmosphere ac on the polarization difference we use

illustration not visible in this excerpt

with atmospheric opacity τ and surface polarization difference Ps. This approx­imation is valid for a horizontally stratified atmosphere under Arctic conditions

illustration not visible in this excerpt

Figure 3.1: Vertical (V) and horizontal (H) emissivity of sea ice and sea water measured at an incident angle of Θ = 50° at different frequencies. The vertical lines show the intersect with the AMSR-E frequencies at 19, 37, and 89 GHz. In winter the NORSEX Group (1983) measured first-year (green with stars), multi¬year ice (red with diamonds) and open sea water (blue with crosses) at 4.9, 10.4, 21, 37, and 94 GHz. In late summer Onstott et al. (1987) measured mixed first- year and multi-year ice (cyan with triangles) at 4.9, 10.4, 21, 35, and 94 GHz. As can be seen from this measurements at 89 GHz the emissivity differences A, В and C for the different ice types are similar and much smaller than the emissivity difference D of water.

with an effective temperature replacing the vertical atmospheric temperature profile and a diffusely reflecting surface viewed under an incidence angle of ap­proximately 50° (Svendsen et al., 1987).

The polarization difference in dependence of the ice concentration C can be written as

illustration not visible in this excerpt

where [illustration not visible in this excerpt] and [illustration not visible in this excerpt] are surface polarization differences for ice and water, re­spectively. The atmospheric influence [illustration not visible in this excerpt] in general is a function of the ice con­centration, as both the water vapor content and cloud liquid water decreases from open water to the inner ice pack due to reduced evaporation and cyclones mainly follow the ice edge (Svendsen et al, 1983, 1987; Brümmer et al, 2000). With equation 3.2 on the preceding page the polarization difference [illustration not visible in this excerpt] for the ice concentration C = 0 (open water) and atmospheric influence ao is given by

illustration not visible in this excerpt

and similarly for the ice concentration C = 1 (closed ice cover) by

illustration not visible in this excerpt

Taylor expansions of equation 3.2 on the previous page around C = 0 and C = 1 lead to

illustration not visible in this excerpt

if the derivatives of the atmospheric influence [illustration not visible in this excerpt] for C = 0 and [illustration not visible in this excerpt] for C = 1 are considered to be zero, assuming the variation of the atmospheric influence to be small for totally ice covered or open water areas. With equations 3.3 and 3.4 the dependence of the atmospheric influence in equations 3.5 and 3.6 can be substituted and the ice concentration is given by:

illustration not visible in this excerpt

For Arctic conditions [illustration not visible in this excerpt] is a typical value for sea ice signatures (Svendsen et al, 1987). To be able to retrieve all ice concentration values between 0% and 100% we need to interpolate between the solutions of equations 3.7 and 3.8. Assuming the atmospheric influence to be a smooth func­tion of the ice concentration C we select a third order polynomial for the sea ice concentration between open water and 100% ice cover:

illustration not visible in this excerpt

With equations 3.7 and 3.8 and their first derivatives the unknown coefficients di in equation 3.9 can be determined by solving the linear equation system:

illustration not visible in this excerpt

With the thereby found coefficients [illustration not visible in this excerpt] to [illustration not visible in this excerpt], equation 3.9 can be used to calculate the sea ice concentration if the tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt] for open water and 100% ice coverage are known. C is set equal to zero for [illustration not visible in this excerpt] and equal to one for [illustration not visible in this excerpt].

The correct choice of the tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt] is important for the retrieval of the sea ice concentration as they also include the mean atmospheric influence. According to the original Svendsen algorithm it was suggested to choose the max­ima and minima of the polarization difference of the accordant swath (data of one overflight) as tie points, forming a self-adjusting procedure for different at­mospheric conditions (Svendsen et al, 1987). However, it was found that due to changing atmospheric influence within one swath the maximum (minimum) polarization difference often is not the best representation for open water (100 percent ice cover) and is causing non-physical steps in the ice concentration when combining the swaths (Lomax et al, 1995; Kaleschke et al, 2001). Another study successfully used fixed, hand selected tie points for the Svendsen algorithm dur­ing the Arctic Ocean Section expedition between 24. July to 9. September 1994 (Lubin et al, 1997). This led to the approach of the ASI algorithm: It uses fixed tie-points that are found by comparing ice concentration of the Svendsen algorithm with well validated reference ice concentration from an independent source, such as an algorithm utilizing the by the atmosphere less influenced lower frequency channels. The tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt] are the two modifiable parameters of the ASI algorithm. They have to be well validated and can be adjusted to changing environmental conditions (e.g. different ice properties due to changing season). Additionally only the weather filter cut-offs for the open ocean can be adjusted (see Section 3.2.1). [illustration not visible in this excerpt] and [illustration not visible in this excerpt] determine the maximum and minimum polarization difference, respectively. The atmospheric influence on [illustration not visible in this excerpt] is small and all ice types even for different seasons have a similar polarization difference (Figure 3.1 on page 33). [illustration not visible in this excerpt] therefore has to be the best representation for all ice types in the dataset. The atmospheric influence on [illustration not visible in this excerpt] is larger as cloud liquid water and water vapor reduce the polarization difference above water. Addi­tionally the polarization difference is influenced by the wind driven roughening of the ocean. Thus the choice of [illustration not visible in this excerpt] also includes the general atmospheric in­fluence on the polarization difference. The tie-points [illustration not visible in this excerpt] have been chosen by correlation comparison with AMSR-E Bootstrap ice con­centration (Screen, 2004). They are used through the whole year and for both hemispheres to guarantee a consistent ice concentration from day to day. These tie-points lead to the specific version of equation 3.9:

illustration not visible in this excerpt

For regional studies adjusted tie-points may yield better results. For example, a different set of tie-points was used during Polarstern expedition ARK-XX/2 [illustration not visible in this excerpt] which visually represented the ice concentration around the ship better in agreement with the helicopter surveys. With the operational tie- [illustration not visible in this excerpt] К the ice concentration was slightly overestimated, as can be seen in Section 3.5 on page 41.

3.2.1 Weather Filters

One disadvantage of the 89 GHz channels is the pronounced influence of atmo­spheric cloud liquid water and water vapor on the brightness temperatures. Es­pecially cyclones over open water can reduce the polarization difference to values similarly small as those of sea ice. Therefore, effective filters are necessary to remove spurious ice concentration in open water areas. The weather filtering process consists of three steps. All of them use the lower frequency channels with lower spatial resolution. This in general does not lead to a lower resolution of the marginal ice zone, as the higher resolved ice edge always is covered by non zero ice concentration measurements of the lower frequency channels (see Kaleschke et al, 2001). It only may cause grid points along the ice edge to show too high ice concentrations due to missing weather filters.

The weather filtering steps are:

a)The first weather filter uses the gradient ratio (GR) of the 36.5 and 18.7 GHz channels (Gloersen and Cavalieri, 1986), which is positive for water but near zero or negative for ice. This ratio mainly filters high cloud liquid water cases:

illustration not visible in this excerpt

Fourteen scatter plots GR(36.5/18.7) vs. the 18.7 GHz polarization ratio distributed over all seasons and both hemispheres were analyzed to find an optimal threshold which does not filter out too many low ice concentrations but cuts off all spurious ice (Spreen, 2004):

illustration not visible in this excerpt

This threshold keeps all ice concentrations above 15%, which is in general defined as the ice edge contour line (Gloersen et al, 1992; Cavalieri and St. Germain, 1995). For conditions with small atmospheric influence also ice concentration below 15% can be observed.

b) The gradient ratio GR(23.8/18.7) is used to also exclude high water vapor cases above open water (Cavalieri et al, 1995). By again analyzing scatter plots analogue to a) a second threshold was found (Spreen, 2004):

illustration not visible in this excerpt

After applying this filter almost all spurious ice cases in the open ocean are eliminated.

illustration not visible in this excerpt

Figure 3.2: Sea ice concentra­tion on 26 February 26 2003 in the Fram Strait region obtained from AMSR-E 89 GHz data (ASI algo­rithm). Grid spacing is 6.25 km. The corresponding sea ice drift is shown in Figure 4.1 on page 60. The black box marks a region of a dissolving ice edge.

c) Finally, all ASI ice concentrations with corresponding “Bootstrap” ice con­centrations (Comiso et al, 2003) equal zero are set to zero:

illustration not visible in this excerpt

After applying these filters only very few extreme weather events may still cause spurious ice in the open ocean, which than also would appear in lower frequency ice concentration algorithms as is assured by weather filter c). But, as mentioned above and as demonstrated in Section 3.4 on page 40, for low to medium high ice concentrations the atmospheric influence can cause an overestimation of the ice concentration.

3.2.2 ASI Results

An exemplary sea ice concentration map showing the complete Arctic on a 6.25 km polar stereographic grid and using the tie-points [illustration not visible in this excerpt] is shown in Figure 3.9 on page 51. A color table which mimics the human visual impression of ice is used to visualize the ice concentration for non-scientific users. Figure 3.2 shows an example for our study region, the Fram Strait, on 26 February 2003. The same date is also used as an example for the ice drift in Chapter 4 (Figure 4.1 on page 60). Clearly, this ice concentration map shows the variable ice conditions typical for the Fram Strait during winter, with regions of a very compact but also quite open ice cover along the ice edge. Moreover, the fine spatial resolution of 6.25 km allows discrimination of smaller scale features such as polynyas along the coast or downstream of huge multi-year ice floes, as well as the disintegration of the ice pack into ice patches and fingers in the Marginal Ice Zone (MIZ) (see black box).

illustration not visible in this excerpt

Figure 3.3: Comparison of ice concentration on 23 February 2005 in the Sea of Okhotsk. The left image shows the Bootstrap ice concentration on a 12.5 km grid which matches the spatial resolution of the data. The middle image shows the ASI ice concentrations on a 3.125 km grid. The color code gives the ice concentration between 0 and 100%, missing data is marked gray and land is shown in brown. The red ellipse marks a region of open water which is clearly visible in the ASI ice concentrations and the MODIS false color image of that day (right image; bands 7,2,1 (2155 nm, 876 nm, 670 nm)); image courtesy of MODIS Rapid Response Project at NASA/GSFC) but is not visible in the Bootstrap data due to the lower spatial resolution. In the MODIS image (right) dark areas are open water, while bright and blue colors represent clouds, ice and land. White lines mark land borders.

An example of the accomplished improvements in the spatial resolution in comparison to more traditional algorithms using the 19 and 37 GHz channels is demonstrated in a section of the Sea of Okhotsk (Figure 3.3), where a re­gion of open water evolved along the south-easterly end of Sakhalin. This re­gion of open water can be clearly identified in the Moderate Resolution Imaging Spectroradiometer (MODIS) false color image (Figure 3.3 right) of that day. It is correctly reproduced as open water in the ASI AMSR-E ice concentration map (middle), but the Bootstrap AMSR-E map only shows a region of reduced, non­zero ice concentration (left). The coarse resolution of the 18.7 (~ 20.1km) and 36.5 GHz (~ 10.6 km) channels used by the Bootstrap algorithm and all other low frequency algorithms smears out the open water.

In the following sections the error of the ASI algorithm is evaluated (Sec­tions 3.3 and 3.4) and the ASI algorithm results are compared to in-situ ship data (Section 3.5) and to two other ice concentration algorithms (Section 3.6).

3.3 Tie-point Sensitivity Analysis

The ideal tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt] may vary first with each overflight due to chang­ing direct atmospheric influence (equations 3.3 and 3.4), second on the scale of weeks due to changing radiative properties of the surfaces caused by indirect atmospheric influence (temperature, rain and snow) (e.g. Voss et al, 2003) and third with the seasons (Figure 7 in Spreen et al. (2008)). E.g. the fixed tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt] = 11.7 К used here and found by comparison with AMSR-E Bootstrap ice concentration differ from the adaptive tie-points in Section 4.2 in Spreen et al. (2008) and the ones used during R/V Polarstern campaign ARK- XX/2. For all these reasons a difference between the true/ideal and the used tie-points is likely.

To estimate the influence of small errors in the tie-points on the sea ice concen­tration results, a sensitivity analysis has been carried out. The constant tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt], which are also used in the operational AMSR-E ASI data product, were chosen as reference. The sea ice concentration C in equa­tion 3.9 on page 34 is a function of the polarization difference P and the tie-points [illustration not visible in this excerpt] and [illustration not visible in this excerpt], as they are needed to determine the coefficients di in equation 3.9: C[illustration not visible in this excerpt]. These three variables P, [illustration not visible in this excerpt] and [illustration not visible in this excerpt] were varied separately by a value of ΔΡ between —4 and +4 К from their reference values. Then the value of the difference AC between the varied and the reference ice concentration was calculated. For example for [illustration not visible in this excerpt] follows

illustration not visible in this excerpt

The two dashed curves in Figure 3.4 on the following page show example results for ΔΡ = IK (black) and ΔΡ = 4K (red), respectively. For [illustration not visible in this excerpt] and P, AC is calculated accordingly. Additionally [illustration not visible in this excerpt] and [illustration not visible in this excerpt] were varied simultaneously by ΔΡ = [—4... 4]. Some example results are shown in Figure 3.4, where AC for ΔΡ = 1 К and ΔΡ = 4 К is plotted against the reference ice concentration C.

In all these analyzes the difference AC never exceeds ±15% and varies linearly with ΔΡ. The error of P is given by the radiometric resolution of approximately 1 К of the sensor (Table 2.5 on page 27), the deviation of [illustration not visible in this excerpt] and [illustration not visible in this excerpt] from the true value is unknown. However, the seasonal variation of the tie-points in Section 4.3. in Spreen et al. (2008) indicate that the error in [illustration not visible in this excerpt] is of the order of 2 K, leading to an error in C of about 6% at ice concentration near 100%. But for [illustration not visible in this excerpt] the deviation may exceed even 4 К and the error near 0% ice concentration therefore may be larger than 15%. These results will be confirmed in the next section.

illustration not visible in this excerpt

Figure 3.4: Plot of sea ice concentration differences AC between original ice concentration (tie-points [illustration not visible in this excerpt] = 47 К and Ρχ = 11.7К) and ice concentration where the tie-points were altered by 1 К (black curves) and 4 К (red curves). Dashed: differences for variation of the open water tie-point [illustration not visible in this excerpt] by 1 К and 4 K; solid: variation of the ice tie-point [illustration not visible in this excerpt] by —1 К and —4K; dotted: variation of P by 1 К and 4K, respectively

3.4 Error Estimation

The tie-points [illustration not visible in this excerpt], [illustration not visible in this excerpt] depend on the near-surface polarizations [illustration not visible in this excerpt] and [illustration not visible in this excerpt], respectively, and on the atmospheric opacity r (equations 3.1, 3.2). In order to estimate the errors introduced into the ASI results by these quantities, results from the ship campaigns NORSEX and MIZEX (Svendsen et al, 1987), when all required quantities were measured simultaneously, are used:

illustration not visible in this excerpt

The optimal tie-points under these circumstances are found as [illustration not visible in this excerpt] = 46 К and [illustration not visible in this excerpt] = 7.4 К by using equation 3.1 on page 32. They are kept constant and the standard deviation of the ice concentration ас in dependence of C is calculated from equation 3.2 on page 33 assuming r to decrease linearly between [illustration not visible in this excerpt] and [illustration not visible in this excerpt]

The standard deviation of P is given as:

illustration not visible in this excerpt

With equation ( 3.9 on page 34) follows for the standard deviation of C:

illustration not visible in this excerpt

As can be seen in Figure 3.5 on the next page, [illustration not visible in this excerpt] decreases from 25% for C = 0% to 5.7% for C = 100%. Above C = 65%, [illustration not visible in this excerpt] is smaller than 10%. This gives an impression about the error introduced through average day by day and regional variations of the atmospheric opacity and the surface polarization difference if reliable tie-points are used.

Another error is introduced by the measuring accuracy of the AMSR-E ra­diometer of about 1 К at 89 GHz (see Temperature Resolution in Table 2.5 on page 27). Additional calculations show that its influence on the ASI ice concen­tration is below 3.7% (Spreen, 2004).

The assumed accuracy of the lower frequency algorithms is approximately 7%, but also cases with discrepancies up to 30% have been observed (Cavalieri et al, 2006; Comiso et al, 1997, (for SSM/I)). For high ice concentration values the ASI algorithm fits well into this range. For low ice concentration the algorithm may significantly overestimate in cases of high cloud liquid water content, especially when cyclones cross the ice edge. On the other hand, the 89 GHz channels are less affected by ice types, refrozen meltponds and snow layering, however they are sensitive to the density and grain size of the snow on top of the sea ice ( Tonboe et al, 2006b).

3.5 Comparison to Ship Based Observations

During R/V Polarstern cruise ARK-XIX/1 (28 February to 24 April 2003), the already mentioned cruise ARK-XX/2 (16 July to 29 August 2004), and Polarstern cruise ARK-XXII/2 (28 July to 7 October 2007) sea ice conditions around the ves­sel were routinely observed from the bridge by the scientists on board by visual

illustration not visible in this excerpt

Figure 3.5: The expected standard deviation [illustration not visible in this excerpt] (у-axis) in dependence of the ASI ice concentration C (x-axis) using fixed tie-points and standard deviations of [illustration not visible in this excerpt] and [illustration not visible in this excerpt] obtained during field measurements. The red curve shows the total expected standard deviation of C the other, not solid curves, show the error con­tributions of the atmosphere (black dashed, [illustration not visible in this excerpt]), and of the surface polarization differences of open water (green dash-dotted, [illustration not visible in this excerpt]) and sea ice (blue dashed, [illustration not visible in this excerpt])·

surveillance. The winter/spring cruise ARK-XIX/1 started in the Storfjorden and Barents Sea and continued along the west coast of Svalbard up to 82° N in the Fram Strait. Sea ice observations were conducted between 2003-03-06, 09:00 UTC and 2003-04-21, 11:00 UTC. The summer cruise ARK-XX/2 started in Long}'earbyen and went through the Greenland Sea through Fram Strait up to 85° N. Sea ice observations were conducted between 2004-07-24, 15:00 UTC and 2004-08-18, 13:00 UTC. The summer to fall cruise ARK-XXII/2 started in Tromsp, Polarstern steamed through the Barents Sea passing East of Svalbard up to 84.5° N. From there the cruise continued to the East covering almost the com­plete part of the Eurasian and Russian Arctic Ocean. The northern most point was 88.4° N and the eastern most 135° W. Sea ice observations were conducted between 2007-08-01, 17:00 UTC and 2007-09-25, 12:20 UTC in the Laptev Sea. Plots of the cruise track of the three expeditions are shown on the right side of Figure 3.6.

One of the several observed quantities is the total sea ice concentration, which is shown as gray lines in Figure 3.6 for ARK-XIX (top), ARK-XX (middle), and

illustration not visible in this excerpt

Figure 3.6: Comparison of ice concentrations observed from R/V Polarstern with those obtained from AMSR-E data using three different algorithms. From top to bot­tom: expedition ARK-XIX/1 (March/April 2003), ARK-XX/2 (July/August 2004), and ARK-XXII/2 (August/September 2007). Gray line: visual Polarstern ice concen­trations. The differences between these and the ASI, NASA-Team 2, and Bootstrap algorithm ice concentration are shown in black, green, and red, respectively. X-axes give data point numbers (bottom) and the corresponding dates (top). Right side: respective cruise plots.

ARK-XXII (bottom), respectively. Details can be found in Lieser (2005) and Hendricks et al. (2008), the datasets including photos of every observation for ARK-XIX and ARK-XX are available through Lieser et al. (2005) and Lieser and Haas (2005). As the observations were conducted by up to 16 different persons, errors may be introduced due to different subjective estimates of the ice concentration around the ship. The ice concentration estimates represent the area visible from the vessels bridge. If the observations had been done following the ASPeCt sea ice observation protocol ( Worby, 1999), like it was done for cruise ARK-XXII, the observed area should have been limited to 1 km radius. However, the observed area depends on the overall visibility (fog, haze etc.) and thus is often considerably smaller than the AMSR-E 89 GHz footprint and certainly smaller than the 36.5 GHz and 18.7 GHz footprints. Still these are valuable in-situ data for validation of sea ice concentration algorithms.

These in-situ observations are compared to three different AMSR-E sea ice concentration data sets: (1) ASI ice concentrations on a 6.25 km grid using the tie-points [illustration not visible in this excerpt] = 47 К and [illustration not visible in this excerpt] = 11.7 K, (2) NASA-Team 2 ice concentrations on a 12.5 km grid (Markus and Cavalieri, 2000), which is the standard AMSR-E ice concentration data available from NSIDC (Cavalieri and Comiso, 2004), and (3) ice concentrations from the Basic Bootstrap Algorithm (BBA) (Comiso et al, 1997) on a 12.5 km grid, which are provided as differences to NASA-Team 2 con­centrations in the NSIDC data set, too. The differences between these three algorithms and the Polarstern data are shown in Figure 3.6 on the previous page. All three AMSR-E data sets are highly correlated with the Polarstern ice concen­trations. The mean difference lays between —4% and 12% with standard devia­tions of 14% to 19%. The statistical values of the comparisons are summarized in Table 3.1.

[...]