high-resolution mapping and spatial variability of soil

High-resolution mapping and

spatial variability of soil organic carbon

storage in permafrost environments

Matthias Benjamin Siewert

Department of Physical Geography

Stockholm University

Stockholm 2016

As for me, I am tormented with an everlasting itch for things remote.

- in Moby-Dick by Herman Melville

c© Matthias Benjamin Siewert, Stockholm University 2016

Cover illustration: Landscape in Kytalyk with carbon storage map c© Matthias B. Siewert

Divider photographs: c© Matthias B. Siewert

ISBN: 978-91-7649-529-2 (print)

978-91-7649-530-8 (pdf)

ISSN: 1653-7211

Type set with LATEX using Department of Physical Geography thesis template

Published articles typeset by respective publishers, reprinted with permission

Printed by: US-AB, Stockholm, 2016

Distributor: Department of Physical Geography, Stockholm University

Abstract

Large amounts of carbon are stored in soils of the northern circumpolar permafrost region. High-resolution mapping of this soil organic carbon (SOC) is important to better understand and predictlocal to global scale carbon dynamics. In this thesis, studies from five different areas acrossthe permafrost region indicate a pattern of generally higher SOC storage in Arctic tundra soilscompared to forested sub-Arctic or Boreal taiga soils. However, much of the SOC stored inthe top meter of tundra soils is permanently frozen, while the annually thawing active layer isdeeper in taiga soils and more SOC may be available for turnover to ecosystem processes. Theresults show that significantly more carbon is stored in soils compared to vegetation, even infully forested taiga ecosystems. This indicates that over longer timescales, the SOC potentiallyreleased from thawing permafrost cannot be offset by a greening of the Arctic. For all study areas,the SOC distribution is strongly influenced by the geomorphology, i.e. periglacial landforms andprocesses, at different spatial scales. These span from the cryoturbation of soil horizons, tothe formation of palsas, peat plateaus and different generations of ice-wedges, to thermokarstcreating kilometer scale macro environments. In study areas that have not been affected byPleistocene glaciation, SOC distribution is highly influenced by the occurrence of ice-rich andrelief-forming Yedoma deposits. This thesis investigates the use of thematic maps from highlyresolved satellite imagery (<6.5 m resolution). These reveal important information on the localdistribution and variability of SOC, but their creation requires advanced classification methodsincluding an object-based approach, modern classifiers and data-fusion. The results of statisticalanalyses show a clear link of land cover and geomorphology with SOC storage. Peat-formationand cryoturbation are identified as two major mechanisms to accumulate SOC. As an alternativeto thematic maps, this thesis demonstrates the advantages of digital soil mapping of SOC inpermafrost areas using machine-learning methods, such as support vector machines, artificialneural networks and random forests. Overall, high-resolution satellite imagery and robust spatialprediction methods allow detailed maps of SOC. This thesis significantly increases the amount ofsoil pedons available for the individual study areas. Yet, this information is still the limiting factorto better understand the SOC distribution in permafrost environments at local and circumpolarscale. Soil pedon information for SOC quantification should at least distinguish the surfaceorganic layer, the mineral subsoil in the active layer compared to the permafrost and further intoorganic rich cryoturbated and buried soil horizons.

Sammanfattning

Den norra cirkumpolära permafrost regionen lagrar stora mängder kol i jordar och sediment.Högupplöst kartläggning av detta markbundna organiskt kolet (MOC) är viktigt för att bättreförstå dynamiken i kol-cykeln. Både lokalt och globalt. I den här avhandlingen visar forskn-ing från fem olika studieområden i permafrostregionen ett mönster av generellt högre lagring avMOC i arktiska tundrajordar jämfört med sub-arktiska eller boreala taigajordar. En stor del avMOC lagrat i den översta meter i arktiska tundrajordar är permanent fryst. I taigan är den årligaupptiningen av det aktiva lagret djupare och därför är mer MOC tillgängligt för omsättning iekosystemprocesser. Betydligt mer kol lagras i marken jämfört med vegetation, även i fulltbeskogade taigaekosystem. Detta tyder på att det MOC som eventuellt frigörs från tinande per-mafrost, inte kan kompenseras genom ett grönare Arktis. För alla studieområden, är fördelnin-gen av MOC starkt påverkad av periglaciala landformer och processer på olika rumsliga skalor.Dessa spänner från kryoturbation av markhorisonter, bildandet av palsar, torvplatåer och olikagenerationer av is-kilar, till thermokarst som skapar kilometer-stora makro miljöer. I studieom-råden som inte har påverkats av pleistocen nedisning, är MOC-fördelningen i hög grad påverkadav förekomsten av is-rika och reliefbildande Yedoma sediment. Tematiska kartor från myckethögupplösta satellitbilder (<6,5 m upplösning) avslöjar viktig information om lokal utbredningoch variation av MOC. Men att generera sådana tematiska kartor kräver avancerade klassificer-ingsmetoder, inklusive objektbaserad klassificering samt moderna klassificeringsalgoritmer ochdatafusion. Statistiska analyser visar en tydlig koppling mellan marktäcke samt geomorfologioch MOC. Torvbildning och kryoturbation identifieras som två viktiga mekanismer för ackumu-lering av MOC. Som ett alternativ till tematiska kartor, visar denna avhandling fördelarna meddigital kartering av MOC permafrostområden baserad på maskininlärningsmetoder (såsom ”sup-port vector machine”, ”artificial neural networks” samt ”random forest”). Högupplösta satellit-bilder och robusta rumsliga prognosmetoder tillåter detaljerade kartor över MOC. Denna avhan-dling bidrar till en väsentligt ökat mängd markprofildata för de individuella studieområdena.Samtidigt är det tydligt att markprofildata fortfarande är den begränsande faktorn för att bättreförstå MOC-fördelningen i permafrostmiljöer på lokal och cirkumpolära skala. Sådana markpro-filer för markbundet kol kvantifiering bör åtminstone särskilja och rapportera data för det ytnäraorganiska lagret, samt separera mineraljord i det aktiva lagret från permafrost och vidare frånorganiska horisonter som påverkats av kryotubation eller begravts på andra sätt.

Thesis content

This doctoral compilation dissertation consists of a summarising text and the IV articleslisted below.

I Siewert, M.B., Hanisch, J., Weiss, N., Kuhry, P., Maximov, T.C., Hugelius,G., 2015. Comparing carbon storage of Siberian tundra and taiga per-mafrost ecosystems at very high spatial resolution: Ecosystem carbon in taigaand tundra. Journal of Geophysical Research: Biogeosciences 1973–1994.doi:10.1002/2015JG002999

II Siewert, M.B., Hugelius, G., Heim, B., Faucherre, S., 2016. Land-scape controls and vertical variability of soil organic carbon storage inpermafrost-affected soils of the Lena River Delta. CATENA 147, 725–741.doi:10.1016/j.catena.2016.07.048

III Siewert. M.B., Lantuit, H., Hugelius, G., Spatial variability of soil organic carbonin tundra terrain at local scale. manuscript.

IV Siewert, M.B., High-resolution digital mapping of soil organic carbon in permafrostterrain using machine-learning: An integrated case study in a sub-Arctic peat-land environment. manuscript.

Author contributions

The contributions from listed authors are divided as follows for each article.

I MBS led the writing with the help of all co-authors. The study was designed byGH, PK and MBS. Field expertise was provided by TCM. The laboratory analysiswas done by JH, MBS and NW. The remote sensing analysis was done by MBSwith additions from JH and NW. The final datasets were compiled and analyzed byMBS.

II MBS led the writing, complied the datasets and did all statistical and remote sensinganalysis. The writing was assisted by all co-authors. The study was designed byMBS with help from GH. BH provided field expertise and helped the interpretationof the remote sensing analysis. SF provided additional laboratory data and helpedwith the landscape interpretation.

III MBS led the writing with the help of GH and HL. MBS and GH designed the studyand did the fieldwork. Field expertise was provided by HL. MBS compiled the finaldatasets and did all analyses.

IV MBS wrote and conceptualized Paper IV, which is based entirely on his own work.

Contents

1 Introduction 1

2 Background 5

2.1 Permafrost, soil organic carbon and the global carbon cycle . . . . . . . 52.2 Pleistocene legacy in permafrost environments . . . . . . . . . . . . . . 62.3 Periglacial landscapes and landforms . . . . . . . . . . . . . . . . . . . 72.4 Permafrost affected soils . . . . . . . . . . . . . . . . . . . . . . . . . 92.5 Classification and grouping . . . . . . . . . . . . . . . . . . . . . . . . 112.6 Spatial data and mapping of soil organic carbon . . . . . . . . . . . . . 11

3 Study areas 13

4 Methods 15

4.1 Field methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154.2 Laboratory methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.3 Data processing and statistical analysis . . . . . . . . . . . . . . . . . . 164.4 Remote sensing methods . . . . . . . . . . . . . . . . . . . . . . . . . 184.5 Digital soil mapping methods . . . . . . . . . . . . . . . . . . . . . . . 18

5 Results 21

5.1 Paper I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215.2 Paper II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225.3 Paper III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245.4 Paper IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.5 Determining an optimal spatial resolution . . . . . . . . . . . . . . . . 29

6 Discussion 31

6.1 Improvements in the creation of thematic maps . . . . . . . . . . . . . 316.2 Digital soil mapping approaches . . . . . . . . . . . . . . . . . . . . . 326.3 Soil organic carbon storage in permafrost terrain . . . . . . . . . . . . . 336.4 General patterns in soil organic carbon distribution . . . . . . . . . . . 346.5 Spatial groupings and vertical subdivisions of soil pedons . . . . . . . . 356.6 Legacy and Future vulnerability . . . . . . . . . . . . . . . . . . . . . 35

7 Conclusions 37

8 Future research opportunities 41

9 Acknowledgement 45

References 47

1 Introduction

It is estimated with high confidence that current global warming will decrease the extentof the circumpolar permafrost zone and lead to thawing of presently frozen carbon (Ciaiset al., 2014). The effect of global warming is amplified by a factor of three in the Arcticand impacts all components of the cryosphere, including changes in permafrost (Comisoand Hall, 2014). Large amounts of soil organic carbon (SOC) have accumulated in thenorthern circumpolar permafrost region (Tarnocai et al., 2009). This was promoted byfrozen conditions, cold temperatures and frequent waterlogging in permafrost affectedsoils, northern peatlands and wetlands, which reduce decomposition rates of SOC andpromote long-term SOC sequestration (Davidson and Janssens, 2006; Ping et al., 2015).Due to the amount of carbon stored in the Arctic, it is clear that this region has an impor-tant role in the global carbon cycle (McGuire et al., 2009). Increasing temperatures andthawing of permafrost caused by global warming can induce significant disturbance tothe pool of carbon stored in permafrost regions (Grosse et al., 2011). It is therefore ex-pected that permafrost regions will shift from a carbon sink to a source of the greenhousegases carbon dioxide (CO2) and methane (CH4), which will accelerate global warming(Koven et al., 2011).

To better understand the positive feedback effect of SOC stored in the permafrostregions to accelerate climate change, integrated regional studies on permafrost carbondynamics and processes are necessary (McGuire et al., 2009). For this, it is important tobetter characterize the permafrost SOC pool (Kuhry et al., 2013). The latest circumpo-lar estimate of the SOC storage in the entire permafrost region was based on 1778 soilpedons for the 0–1 m depth interval (Hugelius et al., 2014). This number is small giventhe relevance of this SOC pool in the global carbon cycle, the size of the circumpolarpermafrost region, the diversity of the region and remaining regional data gaps (Hugeliuset al., 2014; Kuhry et al., 2010; Ping et al., 2015; Tarnocai et al., 2009; Zubrzycki et al.,2014). For the next generation of projections of permafrost carbon dynamics, more in-formation on the vertical and spatial distribution of SOC in permafrost affected soils isrequired. This is important for the quantification and characterization of the SOC pool(Hugelius et al., 2012; Mishra and Riley, 2012; Mishra et al., 2013), to set it in perspec-tive to the pool of phytomass carbon (PC) stored in vegetation (Hugelius et al., 2011), toassess the vulnerability of the SOC pool to top down thaw and to lateral degradation ofpermafrost (Grosse et al., 2011), to assess the microbiological decomposition potential(Schädel et al., 2014) and for projections of its overall vulnerability (Koven et al., 2011;Harden et al., 2012; Hugelius, 2012). As more data becomes available to characterizethe SOC pool, new spatial upscaling methods need to be investigated. This means bothexploring the strengths and limitations of thematic mapping (Hugelius, 2012) and ex-ploring new direct prediction methods, i.e. digital soil mapping (McBratney et al., 2003;Mishra et al., 2013, 2011). This extends to the question how we group and classify soilpedons (Hugelius, 2012; Beaudette et al., 2013; Ping, 2013a) and what spatial data canbe used for upscaling. A multitude of remote sensing data is available for the polar re-

1


gions (Pope et al., 2014), the challenge remains to determine what information is neededfrom this data and how to extract it. Permafrost occurs in a wide range of environments(Fig. 1.1). In particular, tundra environments have a very variable and fragmented landcover compared to other biomes and high resolution remote sensing imagery is neces-sary to accurately map land cover diversity (Virtanen and Ek, 2014). This variabilityis mainly caused by the occurrence of periglacial processes, landforms and patternedground formation. A particular focus of this thesis is the usage of satellite imagery withvery high ground resolutions < 6.5×6.5 m to map SOC in highly variable tundra terrain.

Thesis Objectives

The overarching aim of this PhD thesis is to improve our understanding of SOCstorage in permafrost environments. This aim is accomplished through a set of morespecific objectives.

• Investigate patterns in vertical and spatial SOC distribution and their relationshipto ecological gradients and geomorphological features in permafrost environments

• Investigate thematic mapping of SOC using high resolution remote sensing data

• Investigate new approaches to map SOC in permafrost environments

2

High-resolution mapping and spatial variability of permafrost carbon

Figure 1.1. Permafrost landscapes. a) Lowland tundra landscape in Kytalyk showing the

edge of a drained thermokarst lake basin. b) Close up of the drained lake basin showing

the landscape mosaic with an ice-wedge polygon in the center. c) Ice-wedge exposure

on Herschel Island showing the sub-surface expression of this landform d) Taiga forest

at Spasskaya Pad. f) Alas landscape near Yakutsk (Spasskaya Pad/Neleger study area).

f) Yedoma exposure from the Lena River Delta. The cliff is approximately 50 m high.

Most of these Yedoma deposits consist of pure ice (outlined in yellow for a small section)

and are therefore also called ice-complex. g) Sub-Arctic mountainous terrain overlook-

ing tundra heath and the Stordalen mire in Abisko, Sweden. All photographs by M.B.

Siewert.

3


4

2 Background

2.1 Permafrost, soil organic carbon and the global carbon cycle

The northern circumpolar permafrost region extents to an ice-free land area of around22.79×106 km2 (Fig. 3.1; Zhang et al., 1999). It is subdivided into four zones depend-ing on the occurrence of permafrost from continuous (≥90 to 100%), to discontinuous(≥50 to <90%), to sporadic (≥10 to <50%) and restricted isolated patches (0 to <10%)(Brown et al., 1997). Permafrost is defined as ground material that remains at or belowa temperature of 0◦C for at least two consecutive years. This includes soil, rock, ice andorganic materials (Van Everdingen, 1998). Soil organic matter (SOM) refers to all natu-rally derived biological materials in soil or at the soil surface, except for parts of roots ofaboveground living plants (Baldock and Nelson, 2006). SOM mainly consists of partlydecomposed plant and animal remains. The focus of this thesis is the measurement ofSOC; the fraction of organic carbon stored in soil and SOM. SOC is a key component ofSOM and is of scientific interest as a key source of CO2 and CH4.

Arrhenius (1897) was first to suggest that changes in CO2 concentration in the atmo-sphere could influence global temperatures and that human sources of CO2 may even-tually have influence on the Earth’s climate. By the 1950’s, it became clear that thisis indeed the case and Revelle and Suess (1957) recognized that humans are in factconducting a large-scale experiment. This experiment is to extract carbon stored in sed-imentary rocks for millions of years and to return it to the atmosphere and the oceans.While curiosity first prevailed, the discussion among researchers turned fast towardswarnings over potential global climatic effects of increased emissions from the burningof fossil fuels (Schneider, 1975). Baes et al. (1977) called it an uncontrolled experi-ment and pointed out that more knowledge about the carbon cycle is necessary. A firstimportant step to better characterize the terrestrial SOC pool, was a separate estimateof SOC stocks for individual global ecosystems by Schlesinger (1977). He estimatedthe total SOC storage for the “Tundra and Alpine” ecosystem to 173 Pg, for the “Bo-real forest” to 179 Pg and for “Swamp and marsh” lands to 137 Pg. Combined theseareas would thus store a total of 489 Pg and correspond approximately to a similar geo-graphical region as the northern circumpolar permafrost region. A review by Post et al.

(1990) highlighted a considerable uncertainty for the terrestrial carbon storage in carboncycle studies. The relevance of the permafrost regions for greenhouse gas emissionswas for example pointed out by Zimov et al. (1997), but it was a review by Gruberet al. (2004) that drew considerable attention to the potential significance of frozen SOCstored in permafrost. They found a range of 200–800 Pg carbon stored in wetlands and200–800 Pg stored in permafrost soils. Based on a best estimate of 400 Pg for the lat-ter, they concluded a potential release of 80 Pg to the atmosphere over the next century.Lenton et al. (2008) introduced the notion of tipping elements in the Earth’s climate sys-tem and identified “permafrost and tundra loss” and a “boreal forest” dieback or biomeshift as two critical potential tipping points. This means that small perturbations causing

5


the breach of a critical threshold could change the state of the climate system and itsfuture system trajectory. The vulnerability of the permafrost carbon pool was assessedby Schuur et al. (2008) and around the same time, the Northern Circumpolar Soil Car-bon Database (Hugelius et al., 2013; Tarnocai et al., 2009) was compiled to assess theamount of carbon stored in the soil cover and sediments of the circumpolar permafrostregion as defined by Brown et al.(1997; Fig. 3.1). Using this database, Tarnocai et al.

(2009) estimated a total SOC storage of 1672 Pg for this region. This estimate was aconsiderable increase over previous estimates as it also included deeper soil layers from1–3 m depth (528 Pg), deltaic alluvium (241 Pg) and deep Yedoma deposits (407 Pg).The latest estimate of the SOC pool in the circumpolar permafrost region is 1307 Pgwith an assessed uncertainty range of 1140–1476 Pg (Hugelius et al., 2014). This latestestimate divides into 472±27 Pg for the top meter of soil or 1035±150 Pg for the top3 m of soil, 91±52 Pg in Arctic river deltas and 181±54 Pg for SOC stocks in Yedomadeposits below 3 m depth. Of this around 800 Pg are perennially frozen in permafrost(Hugelius et al., 2014). The study reports much reduced uncertainty for the estimate ofthe SOC pool. However, significant data gaps remain for High Arctic regions, mountain-ous regions and deep deposits including the Yedoma region. Furthermore, a continuedlack of soil pedons is expressed. More soil pedons are needed for the 0–3 m depth inter-val and for regions with thin sediment cover and sampling must address regions with noor very few data points (Hugelius et al., 2014). In a global context, the permafrost SOCpool represents around half of the ca. 3000 Pg global terrestrial SOC pool (Köchy et al.,2015) while covering around 16% of the Earth’s soil area (Tarnocai et al., 2009).

2.2 Pleistocene legacy in permafrost environments

To better understand the present day SOC distribution in the Arctic, a brief look at theLast Glacial maximum (LGM) may help (Fig. 3.1). The LGM lasted approximately from26.5 to 19 ka (Clark et al., 2009). It is estimated that the global-mean temperature wasaround 6◦C colder compared to pre-industrial temperatures. This was most expressedover northern Hemisphere ice-sheets (Schneider von Deimling et al., 2006). The periodof the LGM was also the time of the Last Permafrost Maximum (LPM), a term thatdescribes the maximum extent of permafrost during a period that lasted approximatelyfrom 25 to 17 ka (Vandenberghe et al., 2014). This area was probably around 33% largerthan the present day extent (Lindgren et al., 2015). During this period, sea-level was at itslow stand as water was stored in ice sheets (Peltier and Fairbanks, 2006) that were at theirmaximum extent (Ehlers et al., 2011). These ice sheets covered much of Europe, Green-land, Northern America and Antarctica (Ehlers et al., 2011). However, large land areasin Eastern Asia and Western North America remained unglaciated during this periodand most likely during much of the late Pleistocene (Schirrmeister et al., 2012b). Thisunglaciated area was first described by Hultén (1937) who named it Beringia. Today,this term commonly refers to an area spanning from the Taymyr Peninsula in westernSiberia to the Yukon Territory in western Canada (Schirrmeister et al., 2012b). Duringthe LGM, the area was connected by a land bridge at the Bering Strait, as large areasof the continental shelf were exposed due to the lower sea-level (Romanovskii et al.,2004; Schirrmeister et al., 2012b). It is assumed that periglacial conditions have pre-vailed over much of Beringia during long periods of the Pleistocene. This has led to thedevelopment of a unified regional, cold-adapted flora and fauna and the accumulationof a unique sediment suite called Yedoma (Schirrmeister et al., 2008, 2012b). Yedomabuilds up as intense periglacial weathering, sediment transport and deposition led to theaccumulation of very ice-rich, fine-grained sediments in vast plains and in valley floors.

6


Syngenetic permafrost growth over several tens of thousands of years was associatedwith the formation of massive ice-wedges of several meters in width. Together with avery high ice-content in the remaining sediment, this adds up to a ground ice content of65 to 90% for most Yedoma deposits. These deposits are therefore also referred to as Icecomplex in Russia, while in North America the corresponding deposits are mapped assilt or loess and sometimes called ’muck’ (Schirrmeister et al., 2012b).Today only dis-sected remnants of Yedoma and equivalent deposits remain (Wolfe et al., 2009; Grosseet al., 2013a) and are sometimes exposed in retrogressive thaw slumps, along rivers andthe coast line. These sediments often dominate the landscape evolution, the local geo-morphological setting and soil formation and represent a legacy of organic rich materialvulnerable to thermoerosion. These Beringian periglacial landscapes stand in contrastto periglacial landscapes that were glaciated during the late Pleistocene where soils areyoung and thin as repeated glaciations have rejuvenated the landscapes (André, 2003).

2.3 Periglacial landscapes and landforms

Permafrost environments and periglacial landscapes feature a distinct set of geomor-phological landforms that influence SOC distribution at regional to local scale. Withinthe extent of the Beringian region, Yedoma type deposits are widespread in lowlands,foothills and large river valleys, where eroded remnants form a gentle, low relief land-scape of elevated plateaus. In these ice-rich and fine-grained sediments the formation ofthermokarst lakes is a common phenomenon. These lakes can be several kilometers indiameter. When they are drained they form thermokarst (thaw) lake basins (DTLB), alsocalled Alas by their Siberian term (Fig. 1.1a,b,e; Fig. 2.1a,b; Czudek and Demek, 1970;Soloviev, 1973; Grosse et al., 2013b).

Multiple repeating cycles of strong temperature gradients and hydrological gradi-ents in periglacial environments has caused the formation of patterned ground. Patternedground describes a family of periglacial landforms marked by repeating symmetry of cir-cular or polygonal features forming nets, steps and stripes (Washburn, 1956; Ballantyne,2013; Warburton, 2013). The generally accepted classification distinguishes sorted pat-terned ground for features showing a sorting of soil/sediment grain sizes and non-sortedpatterned ground for features showing no sorting but an uneven but repetitive surfaceand sometimes discontinuous vegetation cover (Washburn, 1956). The later are also of-ten called hummocks. The main driver behind patterned ground formation is the repeatedfreezing and thawing of the ground. In the high latitudes, the active layer thaws duringsummer and refreezes in winter. Patterned ground formation is assumed to be poly-genetic (Washburn, 1956). Most types originate from differential-frost heave and theformation of convection cells in the soil caused by temperature and hydrological gradi-ents (Kessler and Werner, 2003; Peterson and Krantz, 2003, 2008). Thermal contractioncracking is another process that forms ice-wedge polygons. In winter, when soil freezes,cracks form in polygonal blocks. These are filled by snow or in the spring by melt waterthat later refreeze. This repetitive process forms meter-scale, wedge-shaped blocks ofice (Leffingwell, 1915). The formation of regular surface patterns by these landformscan be explained as an expression of self-organization in response to repeated systemdisturbance (Kessler and Werner, 2003). These periglacial phenomena have strong in-fluence on soil formation as they disturb and rework the ground causing the deformation,breaking and subduction of soil material and entire soil horizons, a process also calledcryoturbation (Ping et al., 2008).

Another common feature of northern ecosystems are peatlands. Peatlands are notrestricted to permafrost nor to periglacial environments, but a large portion of the global

7


Figure 2.1. This figure shows the periglacial landscape near Kytalyk. a) Kilometer-

scale variability from a Landsat 5 TM image. In the South a relatively intact Yedoma

plateau dominates, while in the North most of the Yedoma deposits have been incised by

thermokarst lakes. b) Closer view from a GeoEye-1 c©image. At this meso-scale Yedoma

remnants are visible in the middle left and right part. These are undercut by old river

arms in the South and by a thermokarst lake that has drained in the North. c) Ice-wedge

polygon terrain formed in a second generation level of the drained thermokarst lake. d)

Frost mound terrain formed in the first generation thermokarst of the drained lake basin.

e) Hummock dominated tundra terrain on the surface of the Yedoma plateau.

peatlands occur in the northern high latitudes, often associated with permafrost. Peat-lands are generally defined wetland areas with a peat cover of ≥40 cm in NorthernAmerica and ≥30 cm in Russia (Tarnocai and Stolbovoy, 2006) and Sweden, or by theoccurrence of peat-forming plants in Finland (≥75% of Sphagnum moss) (Paavilainenand Päivänen, 1995). Peat is commonly defined as accumulated material with at least17% SOC or 30% organic material (i.e. SOM; Hugelius et al., 2016). The low thermalconductivity and isolating properties of peat promotes permafrost aggregation and theformation of ice lenses. This leads to frost-heave of the organic material and the for-mation of mounds, palsas and peat plateaus that consist of repeated layers of ice andpeat leading to the accumulation of large stocks of SOC (Zoltai, 1972; Åkerman andMalmström, 1986; Seppälä, 2011).

8


2.4 Permafrost affected soils

Soils do not develop uniformly or linearly over time but are a function of different en-vironmental factors that have varying influence. This relationship was first formulatedby the Russian scientist Vasily Dokuchaev (1989, in Jenny, 1980). It was then furtherdeveloped in particular by Hans Jenny (1941, 1980). According to Jenny (1941), a soilproperty (s) is a function of a set of state factors of soil formation. These state factorsinclude climate (cl), organisms (o), topography (r), parent material (p), time (t) and un-specified additional factors (. . . ). Thus, the fundamental equation of soil forming factorscan be formulated as:

s = f (cl,o,r, p, t, ...) (2.1)

This function can be used to address any soil property s in a quantitative way, suchthat the magnitude of s is a function of these factors. This includes the storage of SOCin soils. Each of these soil forming factors can vary independently and take uniqueproperties, giving a wide variety of potential outcomes. Clearly, these factors can alsobe heavily related to and influence each other (Jenny, 1941, 1980). The latter was alsofound for Arctic-patterned ground ecosystems (see Michaelson et al., 2008, for more in-formation). The expression of each of these state factors of soil formation in periglacialenvironments and permafrost regions has been discussed by Ping et al. (2006); Ping(2013b), Bockheim (2015) and Siewert (2015). Johnson et al. (2011) used this frame-work to investigate SOC storage in Alaska. Each of these state factors of soil formationconditions SOC storage in the circum-polar permafrost region in different ways (Ping,2013b). The following section gives an overview on topics relevant to the state factorsof soil formation and SOC storage in the circumpolar permafrost region.

Climate (cl) is considered the most important factor of soil development for coldregion soils (Bockheim, 2015). A wide variety of climates exist in periglacial envi-ronments. These stretch to most combinations of mean annual air temperature from∼-20 to +3◦C and mean annual precipitation from <10 mm in Antarctica to over 2000 mmin some mountainous areas (Humlum, 1998; Bockheim, 2015). Soils in this region expe-rience freezing conditions due to cold temperatures and repeated annual and sometimesdiurnal freeze and thaw cycles. This promotes the occurrence of permafrost as a par-ticular soil thermal regime, when temperatures remain at or below 0◦C for two or moreconsecutive years (Van Everdingen, 1998). Freeze and thaw cycles promote ground-iceof different expressions, which in turn leads cryostructures including ice lenses, gran-ular, platy, lenticular, reticulate, suspended (ataxitic) structures and ice wedges (Pinget al., 2008). This leads to the formation of a set of specific periglacial landforms or-ganized as patterned ground as described earlier. Annual freeze and thaw cycles causecryoturbation, a thermally induced turbic soil process that is effective in transferring soilmaterial from the surface deeper into the soil profile (Ping et al., 2008).

Organisms (o) or the biotic potential as a state factor of soil formation includesflora and fauna. Vegetation and the formation of the surface organic layer has a stronginfluence on permafrost development. Vegetation is an important environmental con-trol factor for permafrost formation and degradation (Shur and Jorgenson, 2007). Theaccumulation of vegetation litter and moss forms a thick surface organic layer in tundraenvironments. This surface organic layer has a strong insulating effect and affects hydro-logical gradients and is therefore an important driver of patterned ground development.Microbial decomposition and root respiration are two processes controlled by organismsthat are responsible for most CO2 production in soils and thus have an important rolein soil development (Davidson and Janssens, 2006). Certain mammals can have strong

9


influence on soil formation. Fox dens create local landforms (Smith et al., 1992) andLemmings have a strong influence on vegetation dynamics to the point that the signal isvisible in satellite imagery (Olofsson et al., 2012). The strongest remaining influence ofmammals on permafrost environment soil may be attributed to the late Pleistocene Mam-moth steppe ecosystem. This provided a high-productivity steppe environment, whichpromoted the build-up of large amounts of SOC in syngenetic permafrost deposits (Zi-mov et al., 2006, 2009, 2012).

Periglacial environments cover the entire topographic (r) range and extent frommountain tops to flat plains with Yedoma hills and to ocean deltas, such as the ArcticLena River Delta discussed in Paper II of this thesis. In most environments, a sequenceof soils with the same parent material and age will develop on a slope due to relief andhydrological influences. This concept is called a catena and is the primary topographiccontrol of soils (Trudgill, 2004). In alpine regions, no or very little soil developmenthas taken place at elevated and exposed topographic positions and soil development andSOC accumulation is limited to slopes and valley floors (Fuchs et al., 2015; Palmtaget al., 2015). Eroded remnants of wide-spread Yedoma type sediments provide a gentlehilly topography with kilometer-scale thaw lakes and drained thermokarst (thaw) lakebasins (DTLB) that formed in these sediments provide a meso-environment for soil de-velopment characterized by wet to waterlogged conditions (cf. Grosse et al., 2013b).Micro-topography associated with decimeter to meter-scale landforms can have a stronginfluence on soil and active layer thickness development. For example the developmentof cryostructures is dependent on the hydrological conditions, which are in many cases afunction of micro-topography, rather than the latitudinal gradient (Ping et al., 2008). Theactive layer thickness of non-sorted circles usually mirrors the surface micro-topography.Different bedrock types have not only strong influence on erosion rates, landscape de-velopment and thus topography (Siewert et al., 2012), but also have strong influence onsoil development as they provide parent material.

All types of parent material (p) can be observed due to the wide range of envi-ronments covered by permafrost environments. This includes various types of sourcebedrock material, grain size ranges and types of deposits including colluvium, alluvium,residuum, eolian deposits, glacial deposits, lacrustine deposits and accumulated organicmatter (Ping, 2013b). The parent material dictates key soil properties such as soil textureand chemical element and nutrient concentrations. This in turn influences vegetation andpatterned ground activity (Michaelson et al., 2008).

The time frame (t) for soil development has not been uniform for all Arctic regions.Large areas in Europe and Northern America have experienced repeated glaciation dur-ing the Pleistocene. Each glacial cycle has eroded and partly stripped off near surfacesoils and reset soil development (André, 2003), while periglacial conditions have pre-vailed over much of Beringia during the Quaternary period (Hultén, 1937; Schirrmeisteret al., 2012b). Periglacial landscapes also show short term dynamics, such as rework-ing of slope deposits by debris flows (Siewert et al., 2012), active-layer detachmentslides (Lewkowicz, 1990) or reworking of alluvial sediments as found in the Lena RiverDelta. On surfaces affected by such processes soil development may be interrupted byerosion/removal or by burial under rapidly accumulating sediments (which may leavepaleosol remnants).

It can be argued that additional state factors of soil formation are relevant to soildevelopment in permafrost environments. While fire is associated with vegetation, it canbe argued that fire is a regional state factor of soil formation on its own. Fire dynamicsare a natural part of the boreal ecosystem. At one of the investigated study areas ofthis thesis (Spasskaya Pad/Neleger), traces of fire of varying severity were visible in

10


all forested areas. Fire over tundra terrain can cause severe disturbance, permafrostdegradation and initiate thermokarst development (Jones et al., 2015). Also, the factorof time in relation to other factors of soil formation can create a distinct historic imprinton soil formation and SOC storage. For example the accumulation of SOC throughsolifluction was related by Becher et al. (2015) to episodes of climatic changes duringthe Holocene.

2.5 Classification and grouping

Permafrost soils have unique thermal properties that differentiate them from other soils.They therefore occupy their own soil order in most soil classification systems, except theRussian system where they are separated at a lower classification level (Ping et al., 2015).Commonly used soil classification systems for permafrost-affected soils are the Cana-dian Soil Classification System (Soil Classification Working Group 1998), the AmericanU.S.D.A. Soil taxonomy system (Soil Survey Staff 2014), the Russian classification sys-tem (Shishov et al., 2004) and the international system of the World Reference Base(WRB) (IUSS Working Group WRB, 2014). In the WRB and in the Canadian soil clas-sification system these soils are called Cryosols, while in the US soil taxonomy they formthe Gelisol soil order. The first key in the US soil taxonomy system specifies Gelisols bythe occurrence of permafrost within the 1 m of the soil surface or signs of cryoturbationor ice segregation and permafrost in the top 2 m. This system distinguishes permafrostsoils into three great groups. These are Histels for organic soils, Turbels for mineral soilsaffected by cryoturbation and Orthels for other mineral soils (Soil Survey Staff 1999).Vertically permafrost-affected soils can be seen in the most simple way as a two layersystem, including the seasonally unfrozen active layer and the permafrost. Often, theactive layer is covered by a surface organic horizon (Ping et al., 2015). As the activelayer thickness varies over the course of years, decades and centuries, a permafrost tran-sition zone can be identified. Depending on prevailing climatic conditions, this zonemay be permanently frozen or part of the annual active layer at any given time (Shuret al., 2005). At the top of this zone, a particularly ice- and SOC-rich transient layer isoften formed. Due to its high ice content this transient layer has increased resistance tothaw and thus acts as a buffer between the active layer and the permafrost (Bockheimand Hinkel, 2005; Shur et al., 2005).

2.6 Spatial data and mapping of soil organic carbon

The varying pre-conditioning of repeated glaciation and non-glacial periglacial condi-tions in the circumpolar permafrost region has created a diverse landscape mosaic in-fluenced by continued cold temperatures. Several landforms unique to these landscapesare prevalent and often visible from satellite images. These landforms cause a high lo-cal scale variability, particularly caused by the occurrence of ice-wedges and patternedground at various scales. This in turn has had a strong influence on soil development andthe accumulation of SOC in these soils (Ping et al., 2015) and soil properties can changeabruptly from one suborder to another (Paper II of this thesis). To quantify SOC stocksin permafrost-affected soils and to understand the spatial distribution of SOC, detailedinput data for spatial scaling of these environments is necessary. One option are soilmaps. However, these are time-consuming and costly to produce and therefore rare forlocal scale study areas. Often, they are only available at very coarse resolutions suchas in the Northern Circumpolar Soil Carbon Database with soil maps at a continentalscale (Hugelius et al., 2013). Land cover classifications, landform classifications and

11


maps of ecological units generated from satellite imagery are commonly used productsfor thematic upscaling of soil pedon values.

Thematic upscaling is a well established method to combine point data and spatialcoverage from a thematic map, such as soil maps or land cover classifications, to producestratified estimates of a variable over an area. Class selection, replication needs, statis-tical uncertainty and general opportunities and limitations of thematic upscaling in per-mafrost regions have been discussed by Hugelius (2012). Several studies have appliedthis method in the circumpolar permafrost region at local level (Hugelius et al., 2010,2011; Zubrzycki et al., 2013; Fuchs et al., 2015; Obu et al., 2015; Palmtag et al., 2015),at regional level (Kuhry et al., 2002; Hugelius and Kuhry, 2009; Bliss and Maurset-ter, 2010; Johnson et al., 2011) and at circumpolar level (Tarnocai et al., 2003, 2009;Hugelius et al., 2014). Hugelius et al. (2011) showed that high resolution mappingis necessary to resolve local scale SOC stocks in periglacial landscapes. In particulartundra landscapes are environments with a very high local scale land cover variabilitycompared to other biomes. This was shown by Virtanen and Ek (2014) who found that2.4 m pixel sizes of QuickBird imagery was necessary to resolve small elongated waterbodies and fen patches that cannot be resolved with lower resolution imagery of Asterwith 15 m and Landsat TM5 with 30 m pixel resolutions.

The internal spectral variability of land cover classes increases as pixel sizes becomefiner than the objects in a remote sensing scene (Woodcock and Strahler, 1987). Thismakes it harder to apply pixel-based classification methods, which led to the increaseduse of object based image classification methods as a way to improve image classifica-tion of geospatial data (Blaschke, 2010; Blaschke et al., 2014). To improve thematicmapping of SOC, quantitative methods need to be developed to design soil groupingsand landscape upscaling classes that preferably combine advantages of satellite derivedland cover maps, like spatial resolution, different sensor types and aerial coverage, whilesufficiently reflecting the subsurface soil properties of interest.

An alternative upscaling approach is to directly link a soil class or soil property toone or several mapped or remotely sensed variables. These variables can be pixel valuesin a satellite image or a digital elevation model (DEM). This direct upscaling or predic-tive modeling approach is generally called digital soil mapping. The main concepts aredirectly derived from Jenny (1941, 1980), but add a spatial component to the traditionalapproach. Based on Jenny’s sclorpt function the conceptual model has been extendedby McBratney et al. (2003) who describe the scorpan model. Where s represents soil ora soil property s at a point, influenced by the climate c at a point, organisms o, topog-raphy r and landscape attributes, parent material p, age a and the time factor and by thespatial position n of each point. Digital soil mapping covers a wide range of methodsand approaches (McBratney et al., 2003; Boettinger et al., 2010). Predictive models canroughly be divided into spatially discrete methods, including machine-learning methods,and into geostatistical interpolation methods that are based on the concept of spatial au-tocorrelation, such as kriging. Combinations of both are also possible. These methodsallow a wide range of spatial input data, including optical satellite and aerial photographydata, elevation data, survey based maps and even radar data (e.g. Bartsch et al., 2016).In this thesis Paper I and II apply thematic mapping to upscale SOC and soil propertiesto landscape scale. Paper III analyses basic assumption of spatial dependency in tun-dra terrain and Paper IV takes a digital soil mapping approach using machine-learningtechniques.

12

3 Study areas

Fieldwork included in this PhD study has been carried out at five study areas (Fig. 3.1).Three of these study areas are located in Eastern Siberia, Yakutia, Russia: SpasskayaPad/Neleger (62◦14’N, 129◦37’E, elevation 220 m.a.s.l.), Kytalyk (70◦49’N, 147◦29’E,11 m.a.s.l.) and the Lena River Delta (72◦22’N, 126◦28’E, 12 m.a.s.l.). The Abisko(68◦21’N, 19◦03’E, 350 m.a.s.l.) study area is located in northern Sweden. HerschelIsland (69◦34’N, 138◦55’W, 1 m.a.s.l.) is situated along the Yukon coastal plain inCanada. Data has been collected at four additional study areas during this PhD. Theseare Tavvavvuoma in northern Sweden, Tiksi located South of the Lena River Delta,the Ogilvie Mountains and the Richardson Mountains, both located along the DempsterHighway in Yukon, Canada. These additional study areas will not be further addressedin this work.

The five study areas of this thesis present a rich picture of the diversity of the cir-cumpolar permafrost region. An excellent general overview on this diversity is providedby Jones et al. (2010). The study areas share some similarities, while they also differ inmany important aspects.

All study areas except Spasskaya Pad/Neleger are located North of the Arctic Circle.The climate in Kytalyk, the Lena River Delta and Herschel Island is hypercontinental.Spasskaya Pad/Neleger is located in central Siberia (Yakutia); a region that featuresthe most extreme continental climate on Earth with a difference in annual minimum tomaximum temperatures of more than 70◦C. Abisko has a much milder eucontinental

climate (Jones et al., 2010).Four of the study areas are located in the zone of continuous permafrost, while

Abisko is located in the zone of discontinuous permafrost according to the circum-Arctic map of permafrost and ground-ice conditions (Brown et al., 1997). However,due to recent and ongoing permafrost degradation Abisko is more typical of the sporadicpermafrost zone (Johansson et al., 2011, 2013).

The Abisko area was covered by the Weichselian ice sheet that carved deep glacialtrough valleys (Holdar, 1959; Grosswald, 1980). Kytalyk, Spasskaya Pad/Neleger andthe Lena river delta are located in Beringia, and have not been glaciated during thePleistocene (Hultén, 1937; Schirrmeister et al., 2012b). Herschel Island has a particu-lar glaciation history, as the island itself is the remainder of a glacial-push moraine ofthe Laurentide ice-sheet (Mackay, 1959; Fritz et al., 2011). In Abisko, moraines andglaciofluvial deposits are also common in the valley floor (Holdar, 1959). At the sametime the surrounding mountain areas feature alpine terrain with weakly developed soilsand bare rock areas.

Spasskaya Pad/Neleger, Kytalyk and the Lena River Delta are fluvial landscapes,but all three areas also feature some remnants of ice-rich Yedoma deposits. The re-search station in Spasskaya Pad is located on a Pleistocene terrace of the Lena river(van Huissteden et al., 2008). In Spasskaya Pad/Neleger the northwestern corner ofthe study area is an Alas landscape developed in Yedoma deposits (cf. Grosse et al.,

13


Figure 3.1. Circumpolar map indicating study areas. The map includes the extent of

different permafrost zones (Brown et al., 1997), the Last Glacial Maximum limits (Ehlers

et al., 2011), the extent of Yedoma deposits in Siberia (Grosse et al., 2013b) and North

American silt and loess deposits (Wolfe et al., 2009).

2013a). In Kytalyk, three landscape scale geomorphological units can be distinguished:Yedoma remnants, Alas and thermokarst lakes that formed in the Yedoma deposits anda floodplain (Schirrmeister et al., 2012a; Weiss et al., 2016). The Lena River Delta iscommonly subdivided into three major river terraces. The third terrace is a Late Pleis-tocene Yedoma remnant (Ice Complex), the second terrace formed as a river responseto the transition from glacial to Holocene conditions and the first terrace is a result ofHolocene deposition (Schwamborn et al., 2002).

Abisko and Spasskya Pad/Neleger are at least partly forested areas, while Kytalyk,the Lena River Delta and Herschel Island are located North of the tree line in tundraterrain (Walker et al., 2005). All study areas have to some extent peat deposits. This ismost expressed in Abisko, with a prevalent occurrence of palsas (Rydén et al., 1980).

14

4 Methods

4.1 Field methods

This section describes a general methodological approach taken for the studies that com-prise this thesis. Deviations from this general approach are discussed in each paperindividually.

During each field campaign a common set of data was collected. Sampling of astudy site included soil samples from soil pedons and a land cover and vegetation survey.The general aim was to replicate the respective landscape heterogeneity of a study area.Following a general field reconnaissance, transects with equidistant sampling sites werelaid out to represent all major land cover types and geomorphological units of a studyarea. Each transect was laid out with equidistant points of 50 to 250 m separation. Inmost cases, transects included 8–10 sites of 100 m equidistance, but other combinationswere used to adapt to local conditions. Once a transect was defined, each study site waslocated and marked at the exact position a hand-held GPS device indicated in terms ofdistance to the first sampling point and compass bearing. Sometimes, additional siteshave been sampled to complement land cover types not covered by the transects. Thissampling scheme provides a compromise between semi-randomness and time efficiencyfor field campaigns in remote locations and difficult terrain.

At each study site, a soil pit was dug to the bottom of the active layer, to the bedrockor to reach a depth of ±50 cm (Fig. 4.1d,e). Notes on the depth distribution of soil hori-zons was made and for strongly cryoturbated soils, a sketch of the horizon distributionwas made (beginning with Paper II). The surface organic layer was sampled completelywith three replicates to reflect its variability, expect for soils where the variability wasderived from horizon sketches. Organic soil was sampled by cutting cubes of knownvolume, while the mineral subsoil was sampled using a fixed volume cylinder (60 cm3

or 100 cm3). Soil samples were taken at 5–10 cm increments or according to soil hori-zons. The general target depth of sampling was 100 cm or to the depth of the bedrockif it was shallower. In Spasskaya Pad/Neleger the target depth was extended to 200 cmbecause the active layer was frequently deeper than >100 cm. For peat deposits the targetsampling depth was to reach below the peat/mineral transition even if this was beyond>100 cm depth. In permafrost free peatlands, a fixed volume Russian peat corer was fre-quently used (Fig. 4.1a). Deeper soil layers and the frozen permafrost were sampled byhammering a steel pipe into the ground and retrieving samples for each depth increment(Fig. 4.1b, c). This resulted in ∼5–15 samples per study site depending on depth. A gen-eral site description was made according to a protocol developed jointly in our researchgroup. This protocol includes information such as land cover, landform, topography, siteconditions, a soil horizon sketch, topographical slope, slope aspect, GPS coordinates anddrainage conditions. The US Soil Taxonomy system was used to classify the soils (SoilSurvey Staff 2014). The vegetation cover was described in terms of relative plant func-

15


tional type coverage for a square of 0.5× 0.5 m or 1× 1 m and samples were taken toestimate phytomass carbon.

4.2 Laboratory methods

The total SOC storage in kg m−2for a sample was calculated from the fraction of organicC (%), dry bulk density (gcm−3, BD) and the depth interval (T) of a sample. This wasfurther corrected for the fraction of coarse fragments (CF, >2 mm) contained in thesample.

SOC = (C×BD× (1−CF)×T )×10 (4.1)

All samples were oven-dried at 65–75◦C for 3 or more days. BD was then calculatedfrom the dry soil mass and the field volume of the sample. We performed loss on ignitionanalysis (LOI) on each sample after drying a smaller subsample at 105◦C overnight. Thissample was then burned at 550◦C for 5 h to estimate the fraction of organic carbon andat 950◦C for two hrs to estimate the fraction of inorganic carbon (CaCO3) (Heiri et al.,2001). For all study areas the fraction of inorganic carbon was comparatively low andwas not further analyzed. In general, a subset of around 15–30% of the samples ofeach study area were submitted for measurement of the elemental carbon and nitrogencontent in an elemental analyzer (EA). Polynomial regressions between LOI 550◦C and%C where used as pedo-transfer functions to estimate the SOC content for samples wereonly LOI was performed. While EA measured carbon concentrations by weight may be amore precise technique, LOI samples are 3 orders of magnitude larger than EA samples.Because perfect soil sample homogenization is challenging, the larger LOI samples aretypically more representative of the original sample. Combining both methods offers agood compromise for this purpose.

The active layer SOC storage was calculated either by consecutive depth incrementor according to the effective thickness of each soil horizons derived from perspective cor-rected photographs and sketches (Kimble et al., 1993; Ping et al., 1997, 2013). Deeperlayers were always calculated by consecutive depth. Unsampled depth increments wereinterpolated based on samples from above or below taking field notes on soil horizonsand property changes into account.

The landscape SOC storage was calculated by multiplying the mean(±SD) value ofall soil pedon corresponding to one class by the area occupied by the class. The 95%confidence interval (CI) was calculated for the landscape mean SOC storage in Kytalyk,Spasskaya Pad and the Lena River Delta.

4.3 Data processing and statistical analysis

A set of statistical analyses was used in this thesis. Most data processing and the sta-tistical tests were performed using R statistical software (Team, 2016). An underlyingquestion in this thesis was the abstraction and data structures needed to represent com-plex entities such as permafrost affected soils. The general data processing procedurewas to subdivide each soil profile into depth increments of 1 mm. Note that this rep-resents only the technical precision, not the precision achieved during field samplingwhich is on the order of ±0.5–1 cm. The 1 mm depth slices were then aggregated for thedesired SOC storage increment. These were calculated for commonly used metric depthincrements (e.g. 0–30 cm, 0–100 cm) and for major SOC storage increments into thesurface organic layer (equivalent to O and OA soil horizons), the mineral active layer,

16


Figure 4.1. Illustration of the field methods used to sample permafrost affected soils.

a) A Russian peat corer is used to sample organic rich and poorly consolidated lake

sediments and peat deposits. b) Permafrost is sampled by hammering a steel pipe into

the ground. c) Steel pipe core sample with high ice content and cryostructures. d) Soil

pit dug into the active layer. f) QR-code linking to a video showing the sampling of a

soil pedon on Herschel Island. All photographs and video by M.B. Siewert.

17


the permafrost and SOC-enriched cryoturbated soil pockets or layers. A set of additionalvariables has been processed, such as the thickness of the organic layer (cm) and the ac-tive layer (cm), water content (%) and ice content (%). Various approaches for samplingdepth increments were compared and evaluated in Paper II.

In Paper I, non-metric multidimensional scaling (NMDS) was applied to analyzepatterns in vertical SOC and PC allocation in relation to LCC classes, geomorpholog-ical classes and different environmental variables. NMDS is a multivariate statisticalmethod for indirect gradient analysis robust against artifacts (Minchin, 1987). Individualstudy sites were ranked using Gower’s dissimilarity-metric developed for soil analyses(Gower, 1971). NMDS projects individual sampling sites in ordinate space by optimiz-ing the agreement of the rank order of the distances and dissimilarities (Henderson et al.,2008).

In Paper II, the non-parametric Kruskal-Wallis rank sum test was used to test forsignificant differences in median values of soil properties for different spatial groupingsof soil pedons (H1). This test is more robust than its parametric equivalent one-wayanalysis of variance (ANOVA), and does not assume normality. In the same paper, thenon-parametric Wilcoxon rank sum test was applied to test for significant differencesin soil property median values (H1) between individual groups of soil pedons and fordifferent vertical soil subdivisions. When appropriate, we applied a false discovery ratecorrection in cross-tables (Benjamini and Hochberg, 1995).

In Paper III spatial autocorrelation of soil pedons was analyzed by calculating ex-perimental variograms as an indicator of spatial dependency (Matheron, 1963; Bivandet al., 2008). A theoretical variogram was fitted on top of the experimental variogram.From this the variogram range was derived. The range indicates the spatial distance atwhich autocorrelation occurs.

4.4 Remote sensing methods

For each study area, a land cover classification (LCC) was created for landscape char-acterization and thematic upscaling purposes (Fig. 4.2; except for Paper III). A core ofthis thesis is to examine the added value of using very high-resolution remote sensingdata for upscaling. Very high-resolution refers here to a spatial resolution of 1× 1 mto 6.5× 6.5 m corresponding to the minimum and maximum resolution of the opticalremote sensing data used in this thesis. This stands in contrast to commonly used lowerresolution satellite sensors such as Landsat TM/ETM+/OLI sensors with a spatial reso-lution of 30×30 m.

Each LCC was created by means of supervised image classification. Representativepixel samples for each land cover class were defined in the original imagery. This train-ing data was then used to automatically classify the entire extent of the spatial dataset.The classes in Kytalyk, Spasskaya Pad/Neleger and Abisko were designed to reflect sur-face land cover and vegetation, while in the Lena River Delta they were designed toreflect major geomorphological land units in form of delta terraces. Therefore, we usethe term landform classification (LFC) in Paper II, rather than LCC.

4.5 Digital soil mapping methods

In Paper IV direct upscaling of SOC storage was explored by comparing four machinelearning algorithms (multiple linear regression, artificial neural networks, support vectormachine and random forest) to predict SOC values from a set of point data, i.e. soilpedons, and a total of 23 environmental datasets. The environmental datasets include

18


Figure 4.2. Land cover classification of the Kytalyk study area. a) Geomorphological

setting (Weiss et al., 2016), b) Land cover classification, c) Satellite imagery with inlet

at a 30×30 m reduced resolution, d) Classification result.

among others a LCC, aerial photography and SPOT5 satellite image bands, a DEM, anda series of derivative products, such as normalized difference vegetation index (NDVI),topographical slope and curvature. A prediction model was trained with SOC storagedata from soil pedons and extracted values of the environmental datasets at each respec-tive pixel location identified by GPS coordinates. The trained prediction model was thenapplied to the stacked environmental dataset to develop a map of continuous SOC (orsoil property) values. The accuracy was assessed by comparing predicted to sampledvalues using linear regression.

19


20

5 Results

This section summarizes the results of this thesis. Paper I is comparing the partitioningof the ecosystem carbon storage in one tundra and one taiga study area using thematicmaps. Paper II statistically investigates the landscape and vertical partitioning of SOCin a study area in detail to draw conclusions for thematic mapping and the quantificationof SOC. In Paper III, we describe how periglacial processes and landforms cause a highspatial variability of the SOC distribution in permafrost environments at multiple scalesand what consequences this has to study and map SOC and ecosystem processes inpermafrost environments. Paper IV investigates the use of digital soil mapping to predictpermafrost carbon at pixel level. A complementary study seeks to determine optimalspatial resolutions to map lowland tundra environments.

5.1 Paper I

Siewert, M.B., Hanisch, J., Weiss, N., Kuhry, P., Maximov, T.C., Hugelius, G., 2015.Comparing carbon storage of Siberian tundra and taiga permafrost ecosystems at veryhigh spatial resolution: Ecosystem carbon in taiga and tundra. Journal of Geophysical

Research: Biogeosciences 1973–1994. doi:10.1002/2015JG002999

This study aimed to improve our understanding on how ecosystem carbon is parti-tioned in continuous permafrost landscapes. We compared the ecosystem carbon stor-age, partitioned into phytomass carbon (PC) and SOC and a set of further subdivisions,in one lowland tundra environment in Kytalyk and a taiga environment in SpasskayaPad/Neleger. The study areas are located in NE and E Siberia. We sampled SOC andPC at a total of 57 individual field sites (24 and 33 sites respectively). At each site asoil pedon was sampled to a target depth of 1 m in the tundra environment and to 2 min the taiga environment. A vegetation survey was undertaken including tree phytomassestimates in the taiga zone. High-resolution satellite imagery with a spatial resolutionof 2×2 m were used to generate a LCC for each study area. The LCCs were generatedusing a pixel-based approach with a maximum-likelihood classifier. LCCs were used toextrapolate SOC and PC point measurements to landscape scale, including calculationof vertical and landscape-scale partitioning of carbon storage. A subset of soil horizonsamples was radiocarbon dated to infer landscape history and SOC storage evolution inthe past. Finally, we explored the relationship of SOC with PC, land cover, geomor-phology and a set of soil- and permafrost-related variables using NMDS multivariatestatistics.

We showed that significantly more carbon is stored in soils than in the phytomass.In Kytalyk, PC amounts to only 0.7 ± 0.1 kg m−2and in Spasskaya Pad/Neleger to3.9 kg m−2, including the tree phytomass. Therefore, more than >86% of the ecosys-tem carbon is stored in soils. In Kytalyk the mean ecosystem carbon storage is esti-mated to 28.6± 2.9 kg m−2(mean± 95CI) including the top 1 m of soil. In Spasskaya

21


Figure 5.1. The effect of noise reduction filters as post-processing techniques on the LCC

of Kytalyk. Top left unprocessed LCC, top right processed LCC.

Pad/Neleger, the SOC storage in the top 1 m of soil is 23.4±3.3 kg m−2. In SpasskayaPad/Neleger the active layer (≥127 ± 15 cm) is significantly deeper than in Kytalyk(40±3 cm). Thus, here the total ecosystem carbon storage refers to the SOC stored in thetop 2 m of soils and the PC. This amounts to a total of 41.6±0.2 kg m−2(mean±95CI).

The satellite image resolution of 2× 2 m was sufficient to resolve individual geo-morphological landforms such as frozen mounds and ice-wedge polygons. In the taiga,classification problems occurred for areas with low tree stand densities. The high reso-lution of the satellite imagery meant that post-processing was necessary to remove noisein the original LCCs (Fig. 5.1). Regardless of these issues the high resolution was verybeneficial for characterizing ecosystem carbon stocks at local scale.

The results of the multivariate ordination analyses using NMDS showed that land-scape scale geomorphic features dictate the distribution of land cover classes and distri-bution of SOC (Fig. 5.2). These geomorphic features are mainly thermokarst lakes anddrained lake basins dictated by the occurrence of ice-rich Yedoma deposits. The SOCstorage is overprinted by within-class variability caused by periglacial landforms, i.e.earth hummocks and ice-wedges in the tundra. In the taiga, forest succession, soil tex-ture and moisture distribution seem to control the SOC distribution. The results showedthat for the given response variables (including different SOC and PC subdivisions) thedefined land cover classes formed distinct groups of ecosystem carbon storage.

5.2 Paper II

Siewert, M.B., Hugelius, G., Heim, B., Faucherre, S., 2016. Landscape controls andvertical variability of soil organic carbon storage in permafrost-affected soils of the LenaRiver Delta. CATENA 147, 725–741. doi:10.1016/j.catena.2016.07.048

The aim of this study was to better understand the vertical and landscape scale dif-ferentiation of permafrost-affected soils. We analyzed a total of 50 soil pedons collectedin the Lena River Delta. These were classified according to the U.S.D.A. Soil Taxon-omy. We then grouped the soil pedons according to soil taxonomy, geomorphological

22

Hig

h-reso

lutio

nm

ap

pin

ga

nd

spa

tial

varia

bility

of

perm

afro

stca

rbo

n

Figure 5.2. Results of the NMDS analysis in Paper I for Kytalyk (a,b) and Spasskaya Pad/Neleger (c,d). Individual soil pedons are represented by symbols

grouped by land cover (a,c) and by geomorphological setting (b,d). The symbol size indicates the square of the total SOC. The response variables that control

the ordination diagram are represented by black labels: Blue arrows and labels mark environmental variables. Soil texture is represented by green labels.

23


units (corresponding to delta terrace levels), soil taxonomy classification (into soil order,soil sub-order and soil great group), according to patterned ground type and position,drainage and according to the position in a Landsat 7 ETM+ based LCC (Schneideret al., 2009). We also extracted highly resolved vertical information on SOC storage,C%, N%, CN ratio, bulk density, visible ice and water content according to a set ofseven different vertical subdivisions. These include subdivisions in several variationsaccording to: functional layers into organic layer, active layer, permafrost and furtherinto the permafrost transition zone and into buried and cryoturbated SOC; into soil Mas-ter horizons with and without a separation of cryoturbated Ojj and Ajj horizons; and intometric depth increments (e.g. 0–30, 30–50, 50–100).

The data was further analyzed by plotting these averaged soil properties with depthand as density plots of C% by functional layer (organic layer, active layer and per-mafrost). A LFC covering the central Lena River Delta was constructed by combining a6.5×6.5 m resolution Rapideye c©satellite image with a DEM. This classification aimedto differentiate major delta terrace levels for upscaling purposes. The classification wasconstructed by first aggregating spectrally homogeneous areas into segments that werethen classified using a SVM classifier.

Out of 50 soil pedons, 46 fell into the Gelisol soil order and 4 into the Entisol soilorder. The analysis of different groupings of soils showed the best differentiation whenclassified following geomorphological units. However, other groupings also performedwell, namely according to soil suborder, soil great group and according to vegetation.Some soil properties were better differentiated by certain groupings than others, i.e.SOC stored in the organic layer was only well differentiated if the pedons were groupedaccording to soil taxonomy, while the amount of SOC stored in the active layer wasonly significantly differentiated when pedons were grouped according to land cover. Afollow-up comparison of different soil properties across each of the five distinguishedgeomorphological units, revealed that geomorphologically stable surfaces, i.e. the firstand third terrace, are more similar to each other while landscape types where sedimentsare affected by erosion and accumulation are also more similar.

The comparison of different vertical subdivisions of aggregated soil profiles showedthat a differentiation according to organic layer, mineral active layer, permafrost and cry-oturbated or buried organic soil horizons can extract most information on SOC storage.A subdivision according to soil master horizons also yields meaningful differentiationand information, while a subdivision according to metric depth intervals fails to capturemuch of the important information on SOC stocks in permafrost affected soils.

High resolution plots of vertical changes reveal a strong variability of soil proper-ties with depth (Fig. 5.3). The curves show distinct signals for individual geomorphicterraces. These properties indicate that soil development is most advanced on the oldPleistocene third terrace, followed by the stable Holocene first terrace, while the erodingslopes of the Pleistocene third terrace, and the present day floodplain show much lesssoil development and also store considerably less SOC. The inclusion of a DEM in theconstruction of the LFC proved very beneficial for the differentiation of individual deltaterraces. Thematic upscaling of the soil pedon data using the LFC resulted in a meanlandscape SOC storage of 19.2±2 kg m−2 for the entire study area.

5.3 Paper III

Siewert, M.B. Lantuit, H. and Hugelius, G., Spatial variability of soil organic carbon intundra terrain at local scale. Manuscript.

24


Figure 5.3. Different soil properties of soil pedons from the Lena River Delta grouped

by geomorphological unit (Paper II).

Figure 5.4. Three experimental variograms of soil organic carbon and the corresponding

variogram models fitted for each of the analyzed tundra types (Paper III).

25


Paper III is a conceptual study in which we investigate spatial dependency of SOC intundra terrain on Herschel Island, Canada. Spatial dependency can be expressed as in To-bler’s first law of geography: “everything is related to everything else, but near things aremore related than distant things” (Tobler, 1970). Spatial dependency is a key assumptionfor the analysis of spatial ecosystem processes and for many spatial prediction methods(Legendre and Fortin, 1989; Hengl, 2009). We describe four hierarchical levels at whichSOC and ground ice vary in tundra landscapes. These four levels are decimeter-scalevariability due to cryoturbation, meter-scale variability due to late-Holocene ice-wedgeformation, decameter-scale (tens of meters) variability due to mid-Holocene ice-wedgesand hectometer-scale (hundreds of meters) variability due to geomorphological distur-bance and catenary shifts in landscape.

We set up three major transects of soil sampling points in commonly occurring tun-dra terrain types to analyze the variability of SOC storage. The sampling-layout wasdesigned to generate a large number of distances between soil sampling points to sup-port the generation of variograms. The three terrain types can be characterized as hum-mocky tussock tundra, upland tundra dominated by non-sorted circles and poorly drainedice-wedge polygon terrain. We exemplify high spatial variability of SOC by delineat-ing soil-horizon boundaries in three representative soil pedon profiles from each tundratype. We compute experimental variograms for each tundra type to analyze spatial au-tocorrelation and the variability caused by ice-wedges is shown from crosscut schemesfor ice-wedge polygon terrain and from a retrogressive thaw slump headwall that wasdigitized to calculate the proportion of the active-layer, ice-wedge ice and remainingpermafrost.

The soil horizons in hummocky tussock tundra and in non-sorted circle tundra tendto show a high variability of SOC densities at decimeter scale and the variograms showspatial autocorrelation to a maximal range of 2 m (Fig. 5.4). Soils in polygonal terrainhave homogeneous properties across polygon centers, while polygon rims are underlainby meter-scale ice-wedges. The variogram for this tundra type shows a spatial rangeof 9.5 m. This ice-wedge generation is of late Holocene age and only occurs in wetlower catenary position. Ice-wedges also occur in upland tundra terrain. These have alarger diameter of ∼15–30 m and have likely been active from the middle Holocene andonwards (Fritz et al., 2012). The quantification of a retrogressive thaw slump headwallshows several meters thick, massive ice-wedges entirely displacing the soil for 17.5%of the surface area. To describe the variability at landscape (hectometer) scale, we alsosummarize SOC storage and soil properties for all tundra terrain types.

The described four levels of soil property variability in these tundra terrain typeschallenge the validity of application of Tobler’s law and methods for spatial predictionand ecosystem analysis based on spatial dependency. We show that the observed (some-times repetitive) spatial structure in these permafrost landscapes are fundamentally dif-ferent from most other natural landscapes. These different scale-levels of soil variabilityshould be accounted for when investigating permafrost carbon dynamics at local scales.

5.4 Paper IV

Siewert, M.B., manuscript. High-resolution digital mapping of soil organic carbon inpermafrost terrain using machine-learning: An integrated case study in a sub-Arcticpeatland environment. Manuscript.

This study explores machine learning techniques as a method of direct upscaling ofSOC storage in a sub-Arctic permafrost environment. A total of 47 soil sampling sites

26


Figure 5.5. A subset of the study area of Paper IV surrounding the Stordalen mire com-

plex, Abisko, Sweden. a) shows an illumination corrected orthophoto ( c©Lantmäteriet,

I2014/00691), b) shows the land cover classification created for this study and c) shows

the predicted total SOC storage using the random forest prediction model.

with information on SOC storage was available from an area surrounding the Stordalencatchment near Abisko, northern Sweden. This data is used to investigate the useful-ness of four machine learning prediction algorithms for digital soil mapping of SOCstorage for an area of 65 km2. A total of 23 spatially referenced environmental datasetswere compiled. This includes 1×1 m spatial resolution aerial photography (Fig. 5.5a),a SPOT5 multi-spectral satellite image, a 2× 2 m spatial resolution DEM and avail-able survey-based vector maps on vegetation and quaternary deposits with a scale of1:250 000. Further, a set of derivative datasets was generated from the SPOT5 image(e.g. NDVI, SAVI) and from the DEM (slope, curvature, topographic wetness index).A LCC was created using an object-based classification approach (Fig. 5.5b). The or-thophoto and the DEM were used at 1× 1 m resolution to segment the study area intohomogeneous regions of at least 130 m2. This was combined with a subset of five envi-ronmental datasets using a support vector machine (SVM) classifier to generate a LCC.

All 23 environmental datasets were included in the digital soil mapping process.The 47 soil sampling sites together with a set of 10 pseudo-training sites (with a valueof 0.0 kg m−2for bare rock areas), were used to train four prediction models. Theseare a multiple linear regression model (MLR; Wilkinson and Rogers, 1973), an artificial

27


Figure 5.6. Performance of different prediction models used for digital mapping of SOC

in Paper IV. Each predicted SOC value is compared to the sampled SOC value.

neural network (ANN; Ripley, 1996), a support vector machine (SVM; Chang and Lin,2011) and a random forest model (RF; Breiman, 2001; Liaw and Wiener, 2002). Modelperformance was evaluated by comparing sampled total SOC values against predictedvalues using linear regression. RF was the prediction model that performed best witha coefficient of determination (R2) of 0.936 (Fig. 5.6), followed by SVM (R2=0.866),MLR (R2=0.731) and ANN (R2=0.695). A visual inspection of the developed mapsconfirmed the superior performance of RF. The RF algorithm was further evaluated bytraining only subsets of input pedons including even and uneven numbered pedons andusing only input data from one of two field seasons. These were compared to the re-maining, thus independent, study sites. RF was then used to predict the SOC stored inthe organic layer, top 0–30 cm, top 0–100 cm and the total SOC including deep peatdeposits (<1.5 m; Fig. 5.6c).

Radiocarbon dated soil samples (N = 9) indicate that the bulk of the SOC stored inthis area has accumulated during the past 2000 years. The mean landscape total SOCstorage is estimated to 8.6 ± 7.8 kg m−2 and to 7.6 ± 6.0 kg m−2 for the top meter(0–100 cm). Wetland and peatland areas (including peat plateaus) store on average thelargest amount of SOC with a range of 29.7±7.0 kg m−2 to 38.0±8.4 kg m−2 comparedto non-wetland classes with a range of 2.0± 2.2 kg m−2 to 9.9±3.1 kg m−2. Wetlandareas cover 6.0% of the landscape, but store 23.6% of the total SOC. This highlightstheir overall significance.

All models overestimate low values and underestimate high values of SOC and theRF model is restricted by design to estimates within the range of the input variable,meaning that it does not feature a trend extrapolation for situations where significantlyhigher or lower SOC values could be expected. The LCC is the most important predictivevariable, complemented by vegetation sensitive bands and band composites, such asNDVI, together with the topographic wetness index, elevation and slope.

28


Figure 5.7. Local variance plots are used to analyze the textural signal of different

landforms for the Kytalyk study area. Peaks indicate 1/2 to 3/4 of the object size of

the scene. These object sizes can be linked to distinct periglacial landforms. Note the

logarithmic scale of the x-axes.

5.5 Determining an optimal spatial resolution

As part of this thesis, the spatial structure of satellite scenes has been analyzed forperiglacial terrain. A wide choice of remote sensing products is available. This choicecomes with a compromise in spatial resolution and the size of the scene footprint (amongother factors). A large footprint and fine spatial resolution are desirable, but increasecost, computer processing time and complicate the classification process. Choosingan optimal spatial resolution depends on the type of environment under investigation,the information desired and data processing restrictions. Local variance plots are amethod to reveal the spatial structure of a remote sensing scene (Woodcock and Strahler,1987).This can guide the choice of an optimal spatial resolution to map SOC in periglacialterrain.

Such local variance plots for the Kytalyk study area show the textural dominanceof different geomorphological landforms at specific scales (Fig. 2.1 and 5.7). The localvariance peaks at 7 m for polygonal tundra terrain. Less regular frost-mound dominatedwet tundra terrain has a peak at 5 m. The curve increases sharply towards the lowestmeasured resolution in hummocky tundra terrain on a Yedoma plateau indicating a peak<0.5 m. An analysis of the entire region reflects the curve of the overall dominatingwetland areas covered by ice-wedges and frost mounds. A second curve appears towardslower resolutions with a peak at a resolution of ∼ 2000 m. This peak reflects the spatialstructure of thermokarst lakes and drained lake basins. A peak at 2–3 m was found forthe forested areas in Spasskaya Pad/Neleger (data not shown).

29


30

6 Discussion

6.1 Improvements in the creation of thematic maps

Upscaling of soil pedon data to landscape scale using thematic maps depends both onthe quality of the soil pedon database and on the quality of the thematic map (Hugelius,2012). In this thesis a continuous effort was made to improve the quality of the thematicmaps produced for upscaling of SOC and other soil properties.

Papers I, II and IV used thematic maps derived from high resolution remote sensingimagery. The improvements made in the classification process are within several fields.Paper I uses a traditional pixel-based classification approach, where a class is assignedto each pixel according to a statistical decision rule, i.e. by an image classificationalgorithm. This approach gave results with a high noise level and was addressed using aset of noise reducing post-processing methods (Fig. 5.1). These include sieving of singlepixels and clumping of classes according to majority rules in a moving window.

Several improvements to prevent these problems were applied in Paper II and IV.A wide range of image classifiers are available. Maximum likelihood is the most com-monly used image classification algorithm and was used in Paper I. This classifier as-sumes a normal distribution for the pixel values in each class and the training data(Campbell and Wynne, 2011). This may not be the case in periglacial environmentsthat are marked by strong contrasts in spectral signature of the land surface. A SVMclassifier was used in Paper II and IV. This classifier is capable of creating nonlinearclass boundaries (Hastie et al., 2009). The visual inspection of the results showed gen-erally very satisfactory classification results that were confirmed by good Kappa values.It is suggested that future classification attempts should also investigate the RF.

A further improvement in Paper II and IV was the use of an object based imageclassification approach. For this, the image is first segmented into areas with a similarspectral signal (Comaniciu and Meer, 2002). These segments are then classified basedon the mean spectral signature of the area the segment covers.

In Paper I, the LCCs were generated using 4 spectral bands from optical imagery.The combination of different remote sensing products, also called data fusion, is an in-creasingly popular method for image classification Liao et al. (2015). In Paper II, thecombination of optical satellite imagery and a DEM showed a marked classification im-provement compared to the sole use of multispectral satellite imagery. In Paper IV, thismethod was further refined. Image segmentation was applied in a first step using onlyvery high 1 m resolution orthophoto and a DEM. In a second step, additional satelliteimagery data from SPOT5 with a lower resolution (10 m and 20 m depending on theband) and topographical slope were added to classify the segments derived in the firststep.

Papers I and IV show a visual comparison of the classification results with the orig-inal high-resolution imagery. In Paper I, the original 2 × 2 m resolution imagery iscompared to a downscaled inlet with a resolution of 30×30 m. This shows marked im-

31


provement for the estimation of SOC and PC. The thematic maps for Kytalyk, SpasskayaPad/Neleger and Abisko were designed to primarily reflect land cover and vegetation.For Paper II, the thematic map was designed to reflect geomorphological units ratherthan vegetation. In general, the separation of spectral distinct classes worked well, e.g.for bare ground or peat bog. It was often a challenge to define classes that reflect SOCstorage coherently and to draw the distinctive line between two classes that transitioninto each other along a natural continuum. This was in particular problematic in thespectral poorly differentiating lowland tundra environments of Kytalyk and the LenaRiver Delta, but also for different forest classes in Spasskaya Pad/Neleger and Abisko.

For this thesis, we looked for the optimal spatial resolution to map tundra terrain.The peak in local variance curves occurs approximately at 1/2 to 3/4 of the size of thedominating object in a remote sensing scene (Woodcock and Strahler, 1987). The con-structed local variance plots for three tundra terrain types in Kytalyk (Fig. 2.1 and (Fig.5.7), show that for mapping and modeling of SOC, a resolution of ∼ 5 m or finer isneeded to distinguish between frost mounds and wet fen areas, and between polygoncenters and polygon rims, respectively. To account for kilometer-scale thermokarst fea-tures, a resolution below ∼ 2000 m is recommended. Similar results have been suggestedby other researchers in such environments. For example, Muster et al. (2012) indicatethat a resolution of 4 m is necessary to map the local scale heterogeneity of ice-wedgeterrain in the Lena River Delta.

6.2 Digital soil mapping approaches

One aim of this thesis is to improve upscaling methods for SOC in periglacial envi-ronments. Improvements in the generation of land cover classifications and work ongrouping of SOC at landscape scale are two ways to improve thematic mapping. Digitalsoil mapping is an entirely different approach to upscaling of SOC.

Many spatial prediction models are based on geostatistical methods that assume spa-tial autocorrelation. The assumption that adjacent observations are more similar to ob-servations further apart in permafrost environments is critically tested in Paper IV. Theresults indicate that the adoption of this assumption based on Tobler’s first law of ge-ography (Tobler, 1970), cannot be uncritically transferred to measurements of SOC inpermafrost environments. Essentially it means that very highly resolved spatial dataand supporting soil pedon data, georeferenced at a similar scale, would be necessary toreplicate patterns in SOC in tundra environments and that there is a substantial naturalvariability in the distribution of SOM in these soils that may not be predicted in detailby any upscaling method. An alternative approach are machine learning methods.

In Paper IV it is demonstrated, that machine learning algorithms can predict SOCstocks in periglacial landscapes even if the amount of input ground data is limited. Fourmodels were compared. RF was the best performing model with a R2 of 0.936 forthe comparison of predicted SOC values to sampled SOC values. The model indicatesthat land cover is the primary predictor of SOC stocks in this landscape, overprinted byvariability of vegetation and topographic influences. This approach has the potential tosignificantly improve upscaling of SOC data in permafrost environments compared tothematic mapping and needs to be further investigated.

Predictive models offer numerous potential approaches. Li and Heap (2011) founda total of 72 spatial interpolation methods and sub-methods applied in environmentalsciences. Often these methods come with several parameters that can be adjusted. Re-searcher also use a multitude of environmental predictor variables. Ließ et al. (2016)generated 236 predictor variables from digital elevation data and satellite imagery. How-

32


17.1 %

4.3 %

8.2 %

41.1 %

10.2 %

1.9 %1.3 % 1.2 %1.5 %

13.1 %

0.1 %0

10

20

30

40

Bareground

Alpineheathtundra

Dwarfshrubs

Alpinewillow

Sparsebirchforest

Birchforest

Forestedwetland

Sedgewetland

Lowlandshrub

wetland

Sphagnumwetland

Peatbog

SO

C (

%)

an

d L

CC

are

a c

ove

rag

e (

%)

SOC Organic topsoil

SOC Mineral subsoil

Area %

Figure 6.1. Landscape partitioning of the total SOC storage for the Abisko study area.

The crosses indicate the landscape fraction of the land cover class, while the columns

represent the fraction of SOC stored in that class subdivided into organic topsoil and

mineral subsoil.

ever, they also found, that the best prediction results were achieved using only a subsetof the environmental variables. Clearly, with all this choice also a large margin for erroris present. This includes the choice of an inadequate prediction model. Paper III and IVindicate that the model needs to be able to reflect very strong environmental gradientsand potentially non-linear effects in SOC storage. Models also depend on appropriateparametrization and there is a risk of over-fitting of models (McBratney et al., 2003). Asfor thematic maps, there should be an empirical relationship between the environmentalvariable, i.e. spatial map and the predicted variable, i.e. SOC. These sources of errormay lead to completely unrealistic upscaling results, in particular in studies with limitedpedon input datasets. The amount of soil pedons in the circum-polar permafrost regionis often very limited and below 50 pedons for a single study area. Some models maycope better with a limited input datasets.

6.3 Soil organic carbon storage in permafrost terrain

This thesis reports landscape estimates of SOC storage for a total of five study areasin the circumpolar permafrost region. In Paper I, the SOC storage for the top 1 m ofsoil is estimated to 27.9± 2.9 kg m−2 for the tundra study area in Kytalyk. This areaincludes Yedoma upland areas, an Alas and a floodplain. The taiga environment inSpasskaya Pad/Neleger is characterized by larch forest on a Pleistocene river terraceand Yedoma sediments with Alas formation. Here the SOC storage was estimated to23± 3.3 kg m−2 for the top 1 m and for the top 2 m it amounts to 41.6± 6.2 kg m−2.

33


Paper II estimates the SOC stock for a subregion at the apex of the Lena River Delta to19.2±2.0 kg m−2. In Paper IV, a predictive model was used to upscale the SOC storagein Abisko. Here the landscape mean SOC storage amounts to 7.6±6.0 kg m−2 for the top1 m covering an extended catchment area with alpine tundra, mountain birch forest andpeat plateaus (Fig. 6.1). Except for the Lena River Delta study area, no previous stratifiedlocal or regional landscape estimates of the SOC were available. Each of studies havesignificantly increased the amount of pedon data available for the respective region.

6.4 General patterns in soil organic carbon distribution

In Paper I, we analyzed patterns of ecosystem carbon distribution using multivariatestatistical ordination diagrams (NMDS method). Major landscape trends in SOC stor-age were caused by the presence of Yedoma hills, thermokarst depressions (Alas) andfloodplains. SOC storage varied significantly at local scale, where the landscape is over-printed by landforms unique to periglacial environments such as ice-wedges and frostmounds in the tundra and by catenary position, soil texture and forest succession in thetaiga. We show that coherent land cover classes of SOC distribution are possible.

In Paper II, statistical tests were undertaken to compare different groupings of SOC.These indicated that the SOC and other soil properties relevant to permafrost-affectedsoils are best differentiated according to major terrace levels of the Lena River Delta, i.e.geomorphological units. The description of soil pedons reveals a tremendous variabilityof these soils at local scale. This was further investigated in Paper III, were we showthat soil horizon variability is associated with differences in SOC density. In Paper IV,land cover is the primary predictive variable of SOC followed by vegetation sensitivesatellite imagery and derivatives and the topographic wetness index. In Paper III, wefind significant differences in soils of three different ecological units that correspondto variations in catenary position and geomorphic disturbance regimes. Paper III alsodescribes the variability of SOC stocks caused by ice wedges. In Paper IV, the relevanceof peatlands and wetlands for the total SOC storage is highlighted. Similarly, it wasfound for Paper I, that Alas-associated land cover classes with a tendency for wettersoil conditions showed proportionally higher SOC stocks in both study areas. In thisPaper, SOC accumulation could be related to either peat formation or cryoturbation.Thus providing two mechanisms for SOC accumulation, of which one applies to organicrich soils in wetland areas and one to mineral upland soils. While in general peatlandsstore proportionally higher amounts of SOC per m2, they are mostly restricted to thelowest catenary position. Thus, in the Abisko study area and in the taiga study area nearSpasskaya Pad/Neleger, the bulk of the SOC is in fact stored in upland mineral soils, butpeatlands can form local hot spots of SOC storage.

All four papers of this thesis point towards geomorphology as a primary control ofSOC allocation and other soil properties. Geomorphology seems to control the overallspatial structure of these landscapes and the distribution of land cover to which veg-etation adapts, and is thus the main factor controlling SOC distribution at local-scale.For example, Johansson et al. (2013) describes vegetation adaption to degrading peatplateaus caused by experimental induced warming of the permafrost. However, there arealso clear signs that vegetation has significant influence on geomorphology in periglacialenvironments (Williams, 1988). For instance, Shur and Jorgenson (2007) describe howecosystems can protect permafrost in the discontinuous permafrost zone and Phillips(2016) argues that geomorphological landforms, such as peat bogs, can be seen as anextended phenotype of biota.

34


6.5 Spatial groupings and vertical subdivisions of soil pedons

In Paper I it became apparent, that many classes are not unique and may be defined indifferent ways. For example the boundary from one forest type to another type may beset with different weights. Also, the dominance and importance of geomorphology be-came apparent for the definition of classes. For example, in Kytalyk the vegetation coverof ice-wedge polygon rims changes, while the underlying ice content should be similar.To address class definition uncertainties found in Paper I, Paper II investigated land-scape controls and the vertical variability of SOC and key soil properties including (C%,N%, CN ratio, bulk density, water and ice content) for a set of 50 soil pedons from theLena River Delta. We compared different groupings, i.e. classes, of soil pedons, to findthe best differentiation at landscape-scale. The goal was to determine whether soil tax-onomy, land cover, geomorphological units at landscape-scale, meter-scale periglaciallandforms, or differences in moisture level would be the best upscaling tool for thesesoils. Soil pedons were best differentiated into geomorphological units correspondingto different delta terraces, followed by land cover and soil taxonomy. An interestingresult was that only geomorphological and land cover based groupings differentiateddifferences in soil C/N ratio; a primary index of carbon decomposability in permafrostenvironments (Schädel et al., 2014).

Furthermore, we addressed the question of how pedons of permafrost affected soilsare best subdivided vertically. For this, a data processing method was developed thatsubdivides soil pedons into depth slices of 1 mm, which allowed a free aggregation ofSOC data according to desired depth increments and across several pedons. Few publi-cations have analyzed soil property changes with depth at such a high vertical resolution.Soils differentiate mainly in the permafrost subsection and in the amount of cryoturbatedSOC, while the organic layer and active layer are fairly uniform. It could be shown, thata subdivision into the organic layer, the active layer, cryoturbated and buried pockets andthe remaining mineral subsoil retained most information. Such a subdivision is thereforerecommended to inventory and analyze permafrost soils.

6.6 Legacy and Future vulnerability

This thesis is primarily addressing the current SOC distribution in the investigated re-gions. Yet, a reoccurring question was the age of the SOC stored in permafrost-affectedsoils and its potential future vulnerability.

The surfaces on Herschel Island (Paper III) and in Abisko (Paper IV) are condi-tioned by the extent of ice sheets at the LGM. In Abisko the deposition of peat in theStordalen mire started around 6000–5000 yrs ago. The majority of the bulk SOC in thebirch forest is likely younger than ∼1350 yrs cal BP as shown by 14C data. Similarly,Becher et al. (2013) found that the oldest cryoturbated SOC on slopes has accumulatedduring the last 2000 yrs. On Herschel Island and along the Beaufort coast, a transitiontowards a warmer climate led to favorable conditions for peat accumulation around 11–10 cal ka BP (Fritz et al., 2012). The three Siberian landscapes stand in contrast to theseareas as they have not been glaciated during the Pleistocene. Radiocarbon dates of thesoils in Spasskaya Pad/Neleger revealed very old soils with an age of the bulk SOC upto 37600± 600 BP 14C yrs. Other analyzed pedons indicated a mostly Holocene ageof the basal peat samples or cryoturbated soil pockets. In the Lena River Delta, SOCstorage is highest for the soils on top of the Pleistocene aged Yedoma deposits. Yet, 14Csamples of cryoturbated SOC indicates a modern age (unpublished data). The first ter-race that accumulated during the late Holocene also stores significant amounts of SOC.

35


Reworked soils from the degrading Ice Complex and present day floodplain level andalluvial sands show significantly lower SOC stocks. In general it seems, surface stabilityand time favors SOC accumulation and cryoturbation seems to play an important role forthe accumulation of SOC. Yet, disturbance can lead to fast depletion of the accumulatedSOC stocks.

In both study areas of Paper I, SOC stocks are significantly larger than PC stocks.Thawing of this presently frozen SOC due to top-down permafrost degradation can likelynot be compensated by uptake into the above ground phytomass as >86% of the ecosys-tem carbon is stored as SOC even in the forested areas in Spasskaya Pad/Neleger. InKytalyk, Spasskaya Pad and the Lena River Delta large scale thermokarst features haveformed in Yedoma sediments. These ice-rich deposits and the SOC stored in the up-per meter of soil are very prone to ongoing lateral permafrost degradation. Consideringthe large amounts of SOC stored in these deposits (Strauss et al., 2013), it is likely thatlandscape scale geomorphological changes associated in particular with thermokarst andlateral degradation of these sediments will have more relevance for the overall ecosys-tem carbon dynamics than a potential increase in above ground vegetation and uptake ofcarbon into phytomass.

36

7 Conclusions

This thesis contributes to efforts to better constrain and understand the circumpolar SOCpool using highly resolved landscape studies in five different permafrost environments.A detailed study of the total ecosystem carbon storage shows that the amount of SOCstored both in Tundra and in Taiga ecosystems is significantly larger than the amount ofcarbon stored in vegetation. The general trend is that SOC storage is higher in Arctic tun-dra areas compared to sub-Arctic areas, but as the active layer is deeper, more unfrozenSOC may be available for turnover in sub-Arctic ecosystems. All studies show that thedistribution of SOC and soil properties is highly variable and influenced by periglaciallandforms and processes at multiple scales. This includes cryoturbation of soil horizonsin non-sorted circles, the formation of several meter large ice-wedge polygons, palsasand peat plateaus and thermoerosion by thermokarst lakes forming macro environmentsof several kilometers in diameter. The occurrence of ice-rich Yedoma deposits vulnera-ble to thermoerosion strongly affects landscape patterns of SOC distribution in regionsthat have not been affected by Pleistocene glaciation. Yet, most of the SOC in the topmeter of soil is of Holocene age in all study areas, but can be older when samplingreached into stable Pleistocene deposits.

Satellite imagery with very high ground resolutions (< 6.5× 6.5 m) was used forthematic upscaling of SOC, PC and other key properties of permafrost-affected soils.Such imagery is shown to be very beneficial for mapping and understanding the SOCdistribution in permafrost environments. It further enabled the investigation of a veryhigh variability of land cover observed at local scale. It was possible to resolve thedistinct land cover in ice-wedge polygon terrain in Siberia and Canada and to highlightthe importance of very small peatlands away from major mire complexes in NorthernSweden. The generation of thematic maps was continuously developed in this thesis byusing an object-based classification method, applying a support vector machine classifierand trough data-fusion of satellite imagery, digital elevation data and several derivativelayers such as topographical slope. This made it possible to improve the quality ofthematic maps. Using local variance plots it is shown that a resolution of <5 m is acrucial threshold to map the variability of SOC in this terrain.

The analysis also showed that a clear link between land cover mapped with satelliteimagery over large areas, geomorphology and SOC storage exists. Overall, the resultsemphasize the role of geomorphology to dictate soil properties at landscape scale, whileprevious studies mainly highlight the relevance of vegetation. Two major mechanismsof SOC accumulation were identified. These are peat formation in poorly drained towaterlogged conditions and cryoturbation in upland mineral soils. The results showthat most information on carbon stocks is extracted when soils are vertically subdividedinto the surface organic layer, the mineral active layer, cryoturbated or buried organicenriched soil horizons and pockets and the remaining mineral permafrost. At the sametime, it was documented that not all sub-surface local scale variability may be mappedfrom satellite imagery.

37


The variability of SOC and soil properties in permafrost environments at local scalehas been described quantitatively before, but only now can it be mapped over severalsquare kilometers at very high-resolution. This means a significant improvement overprevious studies in these remote areas. Importantly, these maps will enable to upscalemeasurement results from the plot scale, at which they are often measured, to the land-scape scale. For instance, carbon release from ecosystem monitoring is often measuredfrom emission chambers or eddy covariance towers that resolve local scale variabilitydepending on land cover changes like dry to wet conditions at meter scale. In previousstudies the challenge was often to bridge measurements from the plot scale to the satel-lite imagery scale and to find meaningful averages and groups of soil pedons and soilproperty values that can be linked to satellite pixels with resolutions of ≥ 30 m. Thisthesis investigates these groups, but also shows that satellite imagery with sufficient spa-tial resolution may bridge these scales. The extent of mapped wetlands was higher thanwould be expected from lower resolution maps.

Previous studies have rarely emphasized critical scales of investigation as identifiedhere and most studies focus on variability caused by one landform or process but do nottake the overlap of several landforms into account or describe the potential relevancefor the SOC distribution and vulnerability. Many ecosystem processes are organizedand vary with land cover at meter scale and it is documented how different periglacialprocesses and landforms become relevant as emerging properties at specific scales andmay have consequences for the geomorphological development. For example, uplandtundra terrain soils may have a dry trajectory for polygon centers and a water saturatedtrajectory for areas over ice-wedges. Such different trajectories may cause significantdifferent emissions of greenhouse gases. This questions simple pathways for carbon re-lease derived from laboratory results. For instance waterlogging of sediment and carbonrelease estimated from incubation results may not follow a simple wet trajectory.

The high spatial variability at this scale is so far largely unaccounted for. For ex-ample, the implementation of cryoturbation (including solifluction) is only now beingaddressed for earth system models. This process is found to be relevant in for nearlyall permafrost environments. This is slightly better addressed for peatlands. Further-more, This thesis highlights that permafrost environments often show sharp contrastsand thresholds in several aspects, e.g. the binary distribution of soil and massive groundice and wetland to upland tundra soils. While this thesis indicates that SOC release frompermafrost may not be offset by a greening of the Arctic purely by mass, this does notimply that vegetation will have no role in the fate of the SOC, since root respiration mayrelease significant amounts of carbon without uptake as plant phytomass.

In the Abisko study area the application of machine learning methods for regressionmodeling of SOC was explored. The best performing prediction model (RF) had an R2

of 0.936 when comparing predicted SOC values with sampled SOC values. Digital soilmapping offers a much more precise picture of the local distribution of SOC, as it canaccommodate pixel-scale variations of the environmental input variables. This meansa significant improvement of upscaling results with only few preceding studies at suchhigh-resolution and in permafrost regions. This approach showed very powerful results,but needs to be further tested and critically assessed. It is important that the digital soilmapping method addresses the fragmented and non-linear nature of SOC distributionin periglacial environments described by soil investigations in this thesis. Digital soilmapping may be a better upscaling method when basic requirements, such as a minimumamount of soil pedons are fulfilled. Yet, thematic maps keep their importance for thestratified extraction of SOC values from maps generated using digital soil mapping as

38


done in Paper IV. Without this thematic information of the SOC the interpretation of anyresult would be very difficult.

Evermore remote sensing data is available and robust methods for the creation ofthematic maps and spatial prediction methods can be developed. Therefore, all studiesin this thesis have concluded that the major limiting factor in high-resolution upscal-ing studies of SOC in these environments is the amount of soil pedons available to theresearcher.

39


40

8 Future research opportunities

Increasing the amount of empirical point information on SOC and soil properties in per-mafrost regions should be a primary research priority for the future. This is in particularthe case for SOC end members, i.e. soils with very low or very high SOC stocks, as theseseem to be the most challenging for prediction-models. While covering the entire envi-ronmental gradient of a particular landscape, pedons should be statistically independentand be collected and analyzed with sufficient precision. An in depth study could ana-lyze appropriate sampling strategies. This also applies to the geographical distribution.Most sub-Arctic and Arctic research facilities are located in remote areas and logisticsare difficult. At the same time they represent the location that is most accessible in thisparticular environment and thus not necessarily representative for the overall landscape.Often only a low amount of soil pedons (<20–50) are available even for key study areas.At circumpolar level, much research has been done in carbon rich areas, but regions withlower SOC storage in the High Arctic and Boreal forests have been neglected. Further,the transition into the non-permafrost regions must also be covered.

In this thesis, very good upscaling results have been achieved with thematic map-ping using land cover and landform classifications. This is in line with work from otherauthors that have been using land cover as a mean to upscale SOC in different studyareas of the circumpolar permafrost region. However, this far no attempt to apply up-scaling using land cover maps at the circumpolar level has been undertaken and shouldbe considered for the future. This will require an in-depth analysis of the suitability ofcircumpolar available land cover maps and grouping of available soil pedons into cor-responding land cover classes. Paper II offers a strategy to investigate what upscalingclasses or groups may be most appropriate. Alternatively, (semi-)automatic soil pedonclassifications based on similarities or dissimilarities of soil pedons or land cover shouldbe investigated (see Beaudette et al., 2013). With an increasing size of databases, thismay turn into a necessity.

Any manual classification process for soil or land cover classes is to a certain degreesubjective and will thus influence the upscaling result, as class definition will have in-fluence on the relative and total estimated amount of SOC. Digital soil mapping erasesthis particular subjectivity and the estimate is continuous, rather than discrete. Manydigital soil mapping approaches exist and are well tested in lower latitude environments,but very few studies have applied these methods in high latitude research environments.Permafrost-affected soils are in many ways unique and these environments are markedby additional challenges. In general, low amounts of soil pedon data and less remotesensing data are available. Therefore, digital soil mapping methods need to be testedwith these restrictions in mind. The response of different prediction algorithms to thedistinct periglacial setting needs to be investigated. For example, can these methodsreplicate Yedoma landscapes with thermokarst features and ice-wedge polygons? Theminimum amount of input data can be investigated using boot-strapping methods. Futurestudies should include confidence intervals and uncertainty estimates for the prediction

41


result. A thorough investigation of potential environmental predictor variables at localscale is necessary. The field of digital soil mapping provides a multitude of different al-gorithms and optimization parameters. Therefore, an in depth review will be necessaryto guide future research. Paper III indicates that spatial prediction methods based on spa-tial autocorrelation may suffer from significant challenges at local scale. Alternatively,machine learning algorithms as investigated in Paper IV, provide promising results as atool for upscaling of SOC in permafrost environments.

This thesis has quantitatively estimated the current distribution of SOC storage in therespective study areas and found a large variability at local scale. Interesting researchbuilding on top of this could address how the SOC storage developed in the past, fromthe LGM over the course of the Holocene and in the future under different scenariosof global warming. Past states of the SOC storage can be investigated by combiningenvironmental reconstructions with 14C dates of SOC and estimates of accumulationrates.

In the Abisko study area permafrost disappearance and active layer thickening as aresponse to climatic changes is ongoing (Åkerman and Johansson, 2008). Many otherstudy areas in the circumpolar region report ongoing permafrost degradation even incold-continuous permafrost (Liljedahl et al., 2016). Therefore, it is important to betterunderstand the geomorphic response of the landscape to permafrost degradation. Thisposes the question what measurable and quantifiable soil variables need to be investi-gated and how remote sensing data can be translated into maps reflecting the vulnerabil-ity of permafrost carbon. Permafrost landscape dynamics and the distribution of groundice have been identified by young researchers as two of the five most pressing researchpriorities for permafrost science (Fritz et al., 2015).

Paper III addressed the high spatial variability of permafrost environments acrossseveral scales from decimeters to hectometers (several hundreds of meters). Such vari-ability needs to be included into models, as it is associated with strong contrasts in SOCstorage, soil and ice-distribution and other sub-surface properties relevant to the vulner-ability of permafrost carbon. This variability affects geomorphological processes thatoccur over large areas, such as waterlogging of soil or the shrubification and greening ofthe Arctic, but are at the same time highly heterogeneous at local scale. Such processesneed to be addressed in earth system models that operate at much coarser scale with pixelsizes measured sometimes in kilometers to degrees and ecological studies. One option isto characterize these properties statistically and to implement them for example as frac-tions in pixel properties. This means to find the right level of abstraction and averaging.Another option to bridge scales would be to investigate simple process indicators andrules of SOC vulnerability derived at the scale of high-resolution satellite imagery or,even better, scale independent. For instance, are there any simple relationships of land-form or land cover and carbon release? If such rules are found at the local scale, it maybe possible to apply them at coarser scales too. However, complexities that would governsuch rules must be investigated. Some processes and relationships may scale linearly,while others may have positive or negative feedback mechanisms, show saturation or de-pletion or other underlying sources of complexity. Wet polygon troughs may get wetteruntil the system drains, while dry centers may get continuously drier. Such simply ex-pressed relationships could be addressed using GIS models and coupled with permafrostmodels. The latter are very good to model subsurface temperatures for simple settings,but only now start to take complex interactions of temperature, hydrology and periglaciallandforms in one or two dimensional space into account. Three dimensional effects areeven more complicated and computationally demanding. The high-resolution approachin this thesis highlights the importance to understand geomorphological dynamics in this

42


respect. Earth system models must take into account geomorphological processes andlandforms that emerge at different spatial scales. These include cryoturbation, ice-wedgedegradation and thermokarst lake formation. However, one will be at the risk to fall intothe reductionist trap. Simple laboratory results such as carbon release from incubationmay not explain landscape scale carbon releases. For instance, in upland tundra not allSOC stored in permafrost will automatically be exposed to wetter conditions, but onlya smaller fraction along the troughs of ice-wedge polygons. Thus, a simple rule at thelowest scale may result in gross misrepresentation if applied inappropriately. Field basedresearch can assist this, by understanding and mapping processes and properties at evenfiner detail such as the thawing of the ice-rich transient layer, which may lead to a switchin ecosystem response and influence the release of cryoturbated SOC.

43


44

9 Acknowledgement

This research has been funded and supported by the EU FP7 PAGE21 project (282700)and the ESF-CryoCarb project. Fieldwork was supported by the NordForsk Defrostproject (23001) and the EU FP7 INTERACT Integrating Activity, the Swedish ResearchCouncil (VR) and the EU JPI-Climate COUP project. The Bolin Centre (Stockholm Uni-versity) provided funding for many activities including field work, course and conferenceparticipations. Other financial support came from Mannerfelts fond and Ahlmanns fond(Stockholm University).

Writing this thesis has been a challenging, but also very interesting and enjoyabletask. I would like to thank in particular my supervisor Gustaf Hugelius for his continuoussupport, his scientific input and expertise that he shared with me, the fruitful discussionsover the years, his sizable patience, for being there whenever needed and for the goodtimes in the field.

I am also very thankful to my co-supervisor Peter Kuhry for his assistance, feedback,support and for the many good and valuable advises I have received from him and forthe good times in the field.

I would also like to thank my former supervisors and university teachers that havepaved my way up this thesis and who’s advice and knowledge has helped and will helpme in the future. These are in particular Hans-Joachim Rosner, Brynhildur Davíðsdóttir,Richard Dikau, Michael Krautblatter and Hanne Christiansen.

I would like to thank all my co-authors for their respective contributions to the ar-ticles in this thesis. Many other co-workers, collaborators and field assistants have alsocontributed to this work and the results. They deserve a big thank you. In particular thelocal partners and people maintaining the field stations that we visited throughout thedifferent field seasons deserve a special thanks. Without their efforts much of our workwould not be possible. I have to mention that the food in Spasskaya Pad will alwayshave a place in my heart.

Over the course of the past years I have met many people at conferences, workshopsand through collaborations. This has made it an interesting avenue for me. A particularthank in this regard goes to Elin Högström and Stefanie Weege for sharing the task ofbeing the representatives of the PAGE21 young researchers for more than 3 years. Weorganized two young researchers workshops for the project and did a lot of the workaround the outreach of the project. I think we accomplished a lot.

During this PhD I have been part of a team of motivated young researchers thatworked on three projects: organizing the young researchers workshop at the 4th Euro-pean Conference on Permafrost (EUCOP) in Évora, Portugal, writing the paper Fritz et

45


al. (2015) and now, as an official IPA action group, working on permafrost comics. Iwould like to thank everyone involved in these efforts very much for the great time andthe constructive and fruitful work we have accomplished.

Thanks to all my other co-workers, colleagues and members of the department forthe enjoyable time.

Thank you to my friends and family for their support and distraction.

I wish to thank Alev for being there for me.

46

References

André, M.-F., 2003. Do periglacial landscapes evolve under periglacial conditions? Geomor-

phology, 52(1-2), 149–164.Arrhenius, S., 1897. ON THE INFLUENCE OF CARBONIC ACID IN THE AIR UPON THE

TEMPERATURE OF THE EARTH. Publications of the Astronomical Society of the Pacific,9(54), 14–24.

Baes, C. F., Goeller, H. E., Olson, J. S., and Rotty, R. M., 1977. Carbon Dioxide and Climate:The Uncontrolled Experiment: Possibly severe consequences of growing CO 2 release fromfossil fuels require a much better understanding of the carbon cycle, climate change, and theresulting impacts on the atmosphere. American Scientist, 65(3), 310–320.

Baldock, J. A. and Nelson, P. N., 2006. Soil organic matter. In M. E. Sumner, editor, Handbook

of Soil Science, pages B25–B84. CRC Press, Boca Raton, FL, USA.Ballantyne, C. K., 2013. PERMAFROST AND PERIGLACIAL FEATURES | Patterned Ground

A2 - Elias, Scott A. In C. J. Mock, editor, Encyclopedia of Quaternary Science (Second

Edition), pages 452–463. Elsevier, Amsterdam.Bartsch, A., Widhalm, B., Kuhry, P., Hugelius, G., Palmtag, J., and Siewert, M. B., 2016. Can

C-band synthetic aperture radar be used to estimate soil organic carbon storage in tundra?Biogeosciences, 13(19), 5453–5470.

Beaudette, D. E., Roudier, P., and O’Geen, A. T., 2013. Algorithms for quantitative pedology: Atoolkit for soil scientists. Computers & Geosciences, 52, 258–268.

Becher, M., Olid, C., and Klaminder, J., 2013. Buried soil organic inclusions in non-sortedcircles fields in northern Sweden: Age and Paleoclimatic context. Journal of Geophysical

Research: Biogeosciences, 118(1), 104–111.Becher, M., Olofsson, J., and Klaminder, J., 2015. Cryogenic disturbance and its impact on

carbon fluxes in a subarctic heathland. Environmental Research Letters, 10(11), 114006.Benjamini, Y. and Hochberg, Y., 1995. Controlling the False Discovery Rate: A Practical and

Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society. Series B

(Methodological), 57(1), 289–300.Bivand, R. S., Pebesma, E. J., and Rubio, V. G., 2008. Applied spatial data: analysis with R.

Springer.Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS Journal of Pho-

togrammetry and Remote Sensing, 65(1), 2–16.Blaschke, T., Hay, G. J., Kelly, M., Lang, S., Hofmann, P., Addink, E., Queiroz Feitosa, R.,

van der Meer, F., van der Werff, H., van Coillie, F., and Tiede, D., 2014. Geographic Object-Based Image Analysis – Towards a new paradigm. ISPRS Journal of Photogrammetry and

Remote Sensing, 87, 180–191.Bliss, N. B. and Maursetter, J., 2010. Soil organic carbon stocks in Alaska estimated with spatial

and pedon data. Soil Science Society of America Journal, 74(2), 565–579.Bockheim, J. G., 2015. Cryopedology. Springer International Publishing, Cham.Bockheim, J. G. and Hinkel, K. M., 2005. Characteristics and significance of the transition zone

in drained thaw-lake basins of the Arctic Coastal Plain, Alaska. Arctic, pages 406–417.Boettinger, J. L., Howell, D. W., Moore, A. C., Hartemink, A. E., and Kienast-Brown, S., editors,

2010. Digital Soil Mapping. Springer Netherlands, Dordrecht.Breiman, L., 2001. Random forests. Machine learning, 45(1), 5–32.Brown, J., Ferrians, O. J., Heginbottom, J. A., and Melnikov, E. S., 1997. Circum-Arctic map of

permafrost and ground-ice conditions. National Snow and Ice Data Center, Boulder, Colorado.

47


Campbell, J. B. and Wynne, R. H., 2011. Introduction to Remote Sensing. Guilford Press, NewYork, USA, 5 edition. Google-Books-ID: NkLmDjSS8TsC.

Chang, C.-C. and Lin, C.-J., 2011. LIBSVM: A library for support vector machines. ACM

Transactions on Intelligent Systems and Technology (TIST), 2(3), 27.Ciais, P., Sabine, C., Bala, G., Bopp, L., Brovkin, V., Canadell, J., Chhabra, A., DeFries, R.,

Galloway, J., Heimann, M., and others, 2014. Carbon and other biogeochemical cycles. InClimate Change 2013: The Physical Science Basis. Contribution of Working Group I to the

Fifth Assessment Report of the Intergovernmental Panel on Climate Change, pages 465–570.Cambridge University Press.

Clark, P. U., Dyke, A. S., Shakun, J. D., Carlson, A. E., Clark, J., Wohlfarth, B., Mitrovica, J. X.,Hostetler, S. W., and McCabe, A. M., 2009. The Last Glacial Maximum. Science, 325(5941),710–714.

Comaniciu, D. and Meer, P., 2002. Mean shift: a robust approach toward feature space analysis.IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(5), 603–619.

Comiso, J. C. and Hall, D. K., 2014. Climate trends in the Arctic as observed from space. Wiley

Interdisciplinary Reviews: Climate Change, 5(3), 389–409.Czudek, T. and Demek, J., 1970. Thermokarst in Siberia and its influence on the development of

lowland relief. Quaternary Research, 1(1), 103–120.Davidson, E. A. and Janssens, I. A., 2006. Temperature sensitivity of soil carbon decomposition

and feedbacks to climate change. Nature, 440(7081), 165–173.Ehlers, J., Gibbard, P. L., and Hughes, P. D., 2011. Quaternary glaciations-extent and chronol-

ogy: a closer look. Elsevier.Fritz, M., Wetterich, S., Meyer, H., Schirrmeister, L., Lantuit, H., and Pollard, W. H., 2011.

Origin and characteristics of massive ground ice on Herschel Island (western Canadian Arctic)as revealed by stable water isotope and Hydrochemical signatures. Permafrost and Periglacial

Processes, 22(1), 26–38.Fritz, M., Wetterich, S., Schirrmeister, L., Meyer, H., Lantuit, H., Preusser, F., and Pollard, W. H.,

2012. Eastern Beringia and beyond: Late Wisconsinan and Holocene landscape dynamicsalong the Yukon Coastal Plain, Canada. Palaeogeography, Palaeoclimatology, Palaeoecology,319–320, 28–45.

Fritz, M., Deshpande, B. N., Bouchard, F., Högström, E., Malenfant-Lepage, J., Morgenstern,A., Nieuwendam, A., Oliva, M., Paquette, M., Rudy, A. C. A., Siewert, M. B., Sjöberg, Y.,and Weege, S., 2015. Brief Communication: Future avenues for permafrost science from theperspective of early career researchers. The Cryosphere, 9(4), 1715–1720.

Fuchs, M., Kuhry, P., and Hugelius, G., 2015. Low below-ground organic carbon storage in asubarctic Alpine permafrost environment. The Cryosphere, 9(2), 427–438.

Gower, J. C., 1971. A General Coefficient of Similarity and Some of Its Properties. Biometrics,27(4), 857.

Grosse, G., Harden, J., Turetsky, M., McGuire, A. D., Camill, P., Tarnocai, C., Frolking, S.,Schuur, E. A. G., Jorgenson, T., Marchenko, S., Romanovsky, V., Wickland, K. P., French, N.,Waldrop, M., Bourgeau-Chavez, L., and Striegl, R. G., 2011. Vulnerability of high-latitudesoil organic carbon in North America to disturbance. Journal of Geophysical Research, 116.

Grosse, G., Robinson, J. E., Bryant, R., Taylor, M. D., Harper, W., DeMasi, A., Kyker-Snowman,E., Veremeeva, A., Schirrmeister, L., and Harden, J., 2013a. Distribution of late Pleistoceneice-rich syngenetic permafrost of the Yedoma Suite in east and central Siberia, Russia. US

Geological Survey Open File Report, 2013(1078), 1–37.Grosse, G., Jones, B., and Arp, C., 2013b. Thermokarst Lakes, Drainage, and Drained Basins.

In J. Shroder, F., R. Giardino, and J. Harbor, editors, Glacial and Periglacial Geomorphology,number 8 in Treatise on Geomorphology, pages 325–353. Elsevier, San Diego.

Grosswald, M. G., 1980. Late Weichselian ice sheet of Northern Eurasia. Quaternary Research,13(1), 1–32.

Group, S. C. W., 1998. The Canadian system of soil classification. Agriculture and Agri-FoodCanada Publication. NRC Research Press, National Research Council of Canada, Ottawa,Canada.

Gruber, N., Friedlingstein, P., Field, C. B., Valentini, R., Heimann, M., Richey, J. E., Lankao,P. R., Schulze, E.-D., and Chen, C.-T. A., 2004. The vulnerability of the carbon cycle in the

48


21st century: An assessment of carbon-climate-human interactions. Scope-Scientific commit-

tee on problems of the environment international council of scientific unions, 62, 45–76.Harden, J. W., Manies, K. L., O’Donnell, J., Johnson, K., Frolking, S., and Fan, Z., 2012. Spa-

tiotemporal analysis of black spruce forest soils and implications for the fate of C. Journal of

Geophysical Research: Biogeosciences, 117(G1), 9.Hastie, T., Tibshirani, R., and Friedman, J., 2009. The Elements of Statistical Learning. Springer

Series in Statistics. Springer New York, New York, NY.Heiri, O., Lotter, A. F., and Lemcke, G., 2001. Loss on ignition as a method for estimating or-

ganic and carbonate content in sediments: reproducibility and comparability of results. Jour-

nal of Paleolimnology, 25(1), 101–110.Henderson, B., Webster, R., and McKenzie N.J., N., 2008. Statistical Analysis. In N. McKenzie,

M. J. Grundy, R. Webster, and A. Ringrose-Voase, editors, Guidelines for Surveying Soil and

Land Resources, number 2 in Australian Soil and land survey handbooks, pages 327–348.Csiro Publishing, Melbourne, 2nd edition.

Hengl, T., 2009. A practical guide to geostatistical mapping of environmental variables, 2nd

Edition. Publications Office, Luxembourg.Holdar, C. G., 1959. The Inland Ice in the Abisko Area. Geografiska Annaler, 41(4), 231–235.Hugelius, G., 2012. Spatial upscaling using thematic maps: An analysis of uncertainties in

permafrost soil carbon estimates. Global Biogeochemical Cycles, 26(2).Hugelius, G. and Kuhry, P., 2009. Landscape partitioning and environmental gradient analyses

of soil organic carbon in a permafrost environment. Global Biogeochemical Cycles, 23(3),GB3006.

Hugelius, G., Kuhry, P., Tarnocai, C., and Virtanen, T., 2010. Soil organic carbon pools ina periglacial landscape: a case study from the central Canadian Arctic. Permafrost and

Periglacial Processes, 21(1), 16–29.Hugelius, G., Virtanen, T., Kaverin, D., Pastukhov, A., Rivkin, F., Marchenko, S., Romanovsky,

V., and Kuhry, P., 2011. High-resolution mapping of ecosystem carbon storage and poten-tial effects of permafrost thaw in periglacial terrain, European Russian Arctic. Journal of

Geophysical Research, 116(G3).Hugelius, G., Routh, J., Kuhry, P., and Crill, P., 2012. Mapping the degree of decomposition

and thaw remobilization potential of soil organic matter in discontinuous permafrost terrain.Journal of Geophysical Research: Biogeosciences, 117(G2), G02030.

Hugelius, G., Tarnocai, C., Broll, G., Canadell, J. G., Kuhry, P., and Swanson, D. K., 2013. TheNorthern Circumpolar Soil Carbon Database: spatially distributed datasets of soil coverageand soil carbon storage in the northern permafrost regions. Earth System Science Data, 5(1),3–13.

Hugelius, G., Strauss, J., Zubrzycki, S., Harden, J. W., Schuur, E. A. G., Ping, C.-L., Schirrmeis-ter, L., Grosse, G., Michaelson, G. J., Koven, C. D., O’Donnell, J. A., Elberling, B., Mishra,U., Camill, P., Yu, Z., Palmtag, J., and Kuhry, P., 2014. Estimated stocks of circumpolar per-mafrost carbon with quantified uncertainty ranges and identified data gaps. Biogeosciences,11(23), 6573–6593.

Hugelius, G., Kuhry, P., and Tarnocai, C., 2016. Ideas and perspectives: Holocene thermokarstsediments of the Yedoma permafrost region do not increase the northern peatland carbon pool.Biogeosciences, 13(7), 2003–2010.

Hultén, E., 1937. Outline of the history of arctic and boreal biota during the quaternary period :

their evolution during and after the glacial period as indicated by the equiformal progressive

areas of present plant species. Thule, Stockholm.Humlum, O., 1998. The climatic significance of rock glaciers. Permafrost and Periglacial

Processes, 9(4), 375–395.IUSS Working Group WRB, I. W., 2014. World reference base for soil resources 2014 interna-

tional soil classification system for naming soils and creating legends for soil maps. Number106 in World Soil Re- sources Reports. FAO, Rome, Italy.

Jenny, H., 1941. Factors of Soil Formation, A System of Quantitative Pedology. McGraw-Hill,New York, NY.

Jenny, H., 1980. The Soil Resource, volume 37 of Ecological Studies. Springer New York, NewYork, NY.

49


Johansson, M., Akerman, J., Keuper, F., Christensen, T. R., Lantuit, H., and Callaghan, T. V.,2011. Past and Present Permafrost Temperatures in the Abisko Area: Redrilling of Boreholes.Ambio, 40(6), 558–565.

Johansson, M., Callaghan, T. V., Bosiö, J., Åkerman, H. J., Jackowicz-Korczynski, M., andChristensen, T. R., 2013. Rapid responses of permafrost and vegetation to experimentallyincreased snow cover in sub-arctic Sweden. Environmental Research Letters, 8(3), 035025.

Johnson, K. D., Harden, J., McGuire, A. D., Bliss, N. B., Bockheim, J. G., Clark, M., Nettleton-Hollingsworth, T., Jorgenson, M. T., Kane, E. S., Mack, M., O’Donnell, J., Ping, C.-L.,Schuur, E. A. G., Turetsky, M. R., and Valentine, D. W., 2011. Soil carbon distribution inAlaska in relation to soil-forming factors. Geoderma, 167–168, 71–84.

Jones, A., Stolbovoy, V., Tarnocai, C., Broll, G., Spaargaren, O., and Montanarella, L., editors,2010. Soil Atlas of the Northern Circumpolar Region. European Commission, PuplicationsOffice, Luxembourg.

Jones, B. M., Grosse, G., Arp, C. D., Miller, E., Liu, L., Hayes, D. J., and Larsen, C. F., 2015.Recent Arctic tundra fire initiates widespread thermokarst development. Scientific Reports, 5,15865.

Kessler, M. A. and Werner, B. T., 2003. Self-organization of sorted patterned ground. Science,299(5605), 380–383.

Kimble, J. M., Tarnocai, C., Ping, C. L., Ahrens, R., Smith, C. A. S., Moore, J. P., and Lynn,W., 1993. Determination of the amount of carbon in highly cryoturbated soils. In Post-

seminar proceedings, Joint Russian-American Seminar on Cryopedology and Global Change,

Pushchino, Russia, volume 15, page 16.Koven, C. D., Ringeval, B., Friedlingstein, P., Ciais, P., Cadule, P., Khvorostyanov, D., Krinner,

G., and Tarnocai, C., 2011. Permafrost carbon-climate feedbacks accelerate global warming.Proceedings of the National Academy of Sciences, 108(36), 14769–14774.

Kuhry, P., Mazhitova, G. G., Forest, P. A., Deneva, S. V., Virtanen, T., and Kultti, S., 2002.Upscaling soil organic carbon estimates for the Usa Basin (Northeast European Russia) usingGIS-based landcover and soil classification schemes. Geografisk Tidsskrift-Danish Journal of

Geography, 102(1), 11–25.Kuhry, P., Dorrepaal, E., Hugelius, G., Schuur, E. A. G., and Tarnocai, C., 2010. Potential

remobilization of belowground permafrost carbon under future global warming. Permafrost

and Periglacial Processes, 21(2), 208–214.Kuhry, P., Grosse, G., Harden, J. W., Hugelius, G., Koven, C. D., Ping, C.-L., Schirrmeister, L.,

and Tarnocai, C., 2013. Characterisation of the Permafrost Carbon Pool: Permafrost Carbon.Permafrost and Periglacial Processes, 24(2), 146–155.

Köchy, M., Hiederer, R., and Freibauer, A., 2015. Global distribution of soil organic carbon –Part 1: Masses and frequency distributions of SOC stocks for the tropics, permafrost regions,wetlands, and the world. SOIL, 1(1), 351–365.

Leffingwell, E. d. K., 1915. Ground-Ice Wedges: The Dominant Form of Ground-Ice on theNorth Coast of Alaska. The Journal of Geology, 23(7), 635–654.

Legendre, P. and Fortin, M. J., 1989. Spatial pattern and ecological analysis. Vegetatio, 80(2),107–138.

Lenton, T. M., Held, H., Kriegler, E., Hall, J. W., Lucht, W., Rahmstorf, S., and Schellnhuber,H. J., 2008. Tipping elements in the Earth’s climate system. Proceedings of the National

Academy of Sciences, 105(6), 1786–1793.Lewkowicz, A. G., 1990. Morphology, frequency and magnitude of active-layer detachment

slides, Fosheim Peninsula, Ellesmere Island, NWT. In Proceedings of the 5th Canadian per-

mafrost conference, volume 54, pages 111–118.Li, J. and Heap, A. D., 2011. A review of comparative studies of spatial interpolation methods

in environmental sciences: Performance and impact factors. Ecological Informatics, 6(3-4),228–241.

Liao, W., Huang, X., Van Coillie, F., Gautama, S., Pizurica, A., Philips, W., Liu, H., Zhu, T.,Shimoni, M., Moser, G., and Tuia, D., 2015. Processing of Multiresolution Thermal Hyper-spectral and Digital Color Data: Outcome of the 2014 IEEE GRSS Data Fusion Contest. IEEE

Journal of Selected Topics in Applied Earth Observations and Remote Sensing, PP(99), 1–13.Liaw, A. and Wiener, M., 2002. Classification and regression by randomForest. R news, 2(3),

18–22.

50


Ließ, M., Schmidt, J., and Glaser, B., 2016. Improving the Spatial Prediction of Soil OrganicCarbon Stocks in a Complex Tropical Mountain Landscape by Methodological Specificationsin Machine Learning Approaches. PLoS ONE, 11(4), 1–22.

Liljedahl, A. K., Boike, J., Daanen, R. P., Fedorov, A. N., Frost, G. V., Grosse, G., Hinzman,L. D., Iijma, Y., Jorgenson, J. C., Matveyeva, N., Necsoiu, M., Raynolds, M. K., Romanovsky,V. E., Schulla, J., Tape, K. D., Walker, D. A., Wilson, C. J., Yabuki, H., and Zona, D., 2016.Pan-Arctic ice-wedge degradation in warming permafrost and its influence on tundra hydrol-ogy. Nature Geoscience, 9(4), 312–318.

Lindgren, A., Hugelius, G., Kuhry, P., Christensen, T. R., and Vandenberghe, J., 2015. GIS-basedMaps and Area Estimates of Northern Hemisphere Permafrost Extent during the Last GlacialMaximum. Permafrost and Periglacial Processes, page 11.

Mackay, J. R., 1959. Glacier ice-thrust features of the Yukon coast. Geographical Bulletin, 13,5–21.

Matheron, G., 1963. Principles of geostatistics. Economic geology, 58(8), 1246–1266.McBratney, A. B., Mendonça Santos, M. L., and Minasny, B., 2003. On digital soil mapping.

Geoderma, 117(1–2), 3–52.McGuire, A., Anderson, L., Christensen, T., Dallimore, S., Guo, L., Hayes, D., Heimann, M.,

Lorenson, T., Macdonald, R., and Roulet, N., 2009. Sensitivity of the carbon cycle in theArctic to climate change. Ecological Monographs, 79(4), 523–555.

Michaelson, G. J., Ping, C. L., Epstein, H., Kimble, J. M., and Walker, D. A., 2008. Soils andfrost boil ecosystems across the North American Arctic Transect. Journal of Geophysical

Research, 113(G3), G03S11.Minchin, P. R., 1987. An evaluation of the relative robustness of techniques for ecological

ordination. Vegetatio, 69(1-3), 89–107.Mishra, U. and Riley, W. J., 2012. Alaskan soil carbon stocks: spatial variability and dependence

on environmental factors. Biogeosciences, 9(9), 3637–3645.Mishra, U., Lal, R., Clay, D., and Shanahan, J., 2011. Predictive mapping of soil organic carbon:

A case study using geographic weighted regressional approach. GIS Applications in Agricul-

ture—Nutrient Management for Improved Energy Efficiency. CRC Press. Forthcoming.Mishra, U., Jastrow, J. D., Matamala, R., Hugelius, G., Koven, C. D., Harden, J. W., Ping, C. L.,

Michaelson, G. J., Fan, Z., Miller, R. M., McGuire, A. D., Tarnocai, C., Kuhry, P., Riley,W. J., Schaefer, K., Schuur, E. A. G., Jorgenson, M. T., and Hinzman, L. D., 2013. Empiricalestimates to reduce modeling uncertainties of soil organic carbon in permafrost regions: areview of recent progress and remaining challenges. Environmental Research Letters, 8(3),035020.

Muster, S., Langer, M., Heim, B., Westermann, S., and Boike, J., 2012. Subpixel heterogeneityof ice-wedge polygonal tundra: a multi-scale analysis of land cover and evapotranspiration inthe Lena River Delta, Siberia. Tellus B, 64(17301).

Obu, J., Lantuit, H., Myers-Smith, I., Heim, B., Wolter, J., and Fritz, M., 2015. Effect of TerrainCharacteristics on Soil Organic Carbon and Total Nitrogen Stocks in Soils of Herschel Island,Western Canadian Arctic. Permafrost and Periglacial Processes, page 16.

Olofsson, J., Tømmervik, H., and Callaghan, T. V., 2012. Vole and lemming activity observedfrom space. Nature Climate Change, 2(12), 880–883.

Paavilainen, E. and Päivänen, J., 1995. Peatland Forestry, volume 111 of Ecological Studies.Springer Berlin Heidelberg, Berlin, Heidelberg.

Palmtag, J., Hugelius, G., Lashchinskiy, N., Tamtorf, M., Richter, A., Elberling, B., and Kuhry,P., 2015. Storage, landscape distribution, and burial history of soil organic matter in contrast-ing areas of continuous permafrost. Arctic, Antarctic, and Alpine Research, 47(1), 71–88.

Peltier, W. R. and Fairbanks, R. G., 2006. Global glacial ice volume and Last Glacial Max-imum duration from an extended Barbados sea level record. Quaternary Science Reviews,25(23–24), 3322–3337.

Peterson, R. A. and Krantz, W. B., 2003. A mechanism for differential frost heave and itsimplications for patterned-ground formation. Journal of Glaciology, 49(164), 69–80.

Peterson, R. A. and Krantz, W. B., 2008. Differential frost heave model for patterned ground for-mation: Corroboration with observations along a North American arctic transect. J. Geophys.

Res, 113.

51


Phillips, J. D., 2016. Landforms as extended composite phenotypes: Landforms as ExtendedComposite Phenotypes. Earth Surface Processes and Landforms, 41(1), 16–26.

Ping, C., 2013a. Gelisols: Part I. Cryogenesis and State Factors of Formation. Soil Horizons,54(3), 0.

Ping, C., 2013b. Gelisols: Part II. Classification and Related Issues. Soil Horizons, 54(4), 0.Ping, C. L., Michaelson, G. J., and Kimble, J. M., 1997. Carbon storage along a latitudinal

transect in Alaska. Nutrient Cycling in Agroecosystems, 49(1-3), 235–242.Ping, C.-L., Michaelson, G., Chapin, F. S., Kimble, J., Oechel, W., Shur, Y., Tarnocai, C., and

Walker, D. A., 2006. The State Factors of Soil Formation in Arctic Tundra. In The 18th World

Congress of Soil Science.Ping, C.-L., Michaelson, G. J., Kimble, J. M., Romanovsky, V. E., Shur, Y. L., Swanson, D. K.,

and Walker, D. A., 2008. Cryogenesis and soil formation along a bioclimate gradient in ArcticNorth America. Journal of Geophysical Research: Biogeosciences (2005–2012), 113(G3).

Ping, C.-L., Clark, M. H., Kimble, J. M., Michaelson, G. J., Shur, Y., and Stiles, C. A., 2013.Sampling Protocols for Permafrost-Affected Soils. Soil Horizons, 54(1), 13.

Ping, C. L., Jastrow, J. D., Jorgenson, M. T., Michaelson, G. J., and Shur, Y. L., 2015. Permafrostsoils and carbon cycling. SOIL, 1(1), 147–171.

Pope, A., Rees, W. G., Fox, A. J., and Fleming, A., 2014. Open Access Data in Polar andCryospheric. Remote Sensing, 6(7), 6183–6220.

Post, W. M., Peng, T.-H., Emanuel, W. R., King, A. W., Dale, V. H., DeAngelis, D. L., andothers, 1990. The global carbon cycle. American scientist, 78(4), 310–326.

Revelle, R. and Suess, H. E., 1957. Carbon Dioxide Exchange Between Atmosphere and Oceanand the Question of an Increase of Atmospheric CO2 during the Past Decades. Tellus, 9(1),18–27.

Ripley, B. D., 1996. Pattern recognition and neural networks. Cambridge university press.Romanovskii, N. N., Hubberten, H. W., Gavrilov, A. V., Tumskoy, V. E., and Kholodov, A. L.,

2004. Permafrost of the east Siberian Arctic shelf and coastal lowlands. Quaternary Science

Reviews, 23(11-13), 1359–1369.Rydén, B. E., Fors, L., and Kostov, L., 1980. Physical properties of the tundra soil-water system

at Stordalen, Abisko. Ecological Bulletins, pages 27–54.Schirrmeister, L., Kunitsky, V. V., Grosse, G., Kuznetsova, T. V., Derevyagin, A. Y., Wetterich,

S., and Siegert, C., 2008. The Yedoma Suite of the Northeastern Siberian Shelf Region Char-acteristics and Concept of Formation.

Schirrmeister, L., Pestryakova, L., Wetterich, S., and Tumskoy, V., 2012a. Joint Russian-Germanpolygon project: East Siberia 2011-2014; the expedition Kytalyk 2011. Berichte zur Polar-

und Meeresforschung= Reports on polar and marine research, 653, 160.Schirrmeister, L., Froese, D., Tumskoy, V., and Wetterich, S., 2012b. Yedoma: Late Pleistocene

ice-rich syngenetic permafrost of Beringia. The Encyclopedia of Quaternary Science, pages542–552.

Schlesinger, W. H., 1977. Carbon Balance in Terrestrial Detritus. Annual Review of Ecology and

Systematics, 8, 51–81.Schneider, J., Grosse, G., and Wagner, D., 2009. Land cover classification of tundra environments

in the Arctic Lena Delta based on Landsat 7 ETM+ data and its application for upscaling ofmethane emissions. Remote Sensing of Environment, 113(2), 380–391.

Schneider, S. H., 1975. On the carbon dioxide-climate confusion. Journal of the Atmospheric

Sciences, 32(11), 2060–2066.Schneider von Deimling, T., Ganopolski, A., Held, H., and Rahmstorf, S., 2006. How cold was

the last glacial maximum? Geophysical Research Letters, 33(14).Schuur, E. A. G., Bockheim, J., Canadell, J. G., Euskirchen, E., Field, C. B., Goryachkin, S. V.,

Hagemann, S., Kuhry, P., Lafleur, P. M., Lee, H., Mazhitova, G., Nelson, F. E., Rinke, A.,Romanovsky, V. E., Shiklomanov, N., Tarnocai, C., Venevsky, S., Vogel, J. G., and Zimov,S. A., 2008. Vulnerability of Permafrost Carbon to Climate Change: Implications for theGlobal Carbon Cycle. BioScience, 58(8), 701–714.

Schwamborn, G., Rachold, V., and Grigoriev, M. N., 2002. Late Quaternary sedimentationhistory of the Lena Delta. Quaternary International, 89(1), 119–134.

Schädel, C., Schuur, E. A. G., Bracho, R., Elberling, B., Knoblauch, C., Lee, H., Luo, Y., Shaver,G. R., and Turetsky, M. R., 2014. Circumpolar assessment of permafrost C quality and its

52


vulnerability over time using long-term incubation data. Global Change Biology, 20(2), 641–652.

Seppälä, M., 2011. Synthesis of studies of palsa formation underlining the importance of localenvironmental and physical characteristics. Quaternary Research, 75(2), 366–370.

Shishov, L. L., Tonkonogov, V. D., Lebedeva, I. I., and Gerasimova, M. I., 2004. Classificationand diagnostics of soils of Russia. Oecumene, Moscow (in Russian).

Shur, Y., Hinkel, K. M., and Nelson, F. E., 2005. The transient layer: implications for geocryol-ogy and climate-change science. Permafrost and Periglacial Processes, 16(1), 5–17.

Shur, Y. L. and Jorgenson, M. T., 2007. Patterns of permafrost formation and degradation inrelation to climate and ecosystems. Permafrost and Periglacial Processes, 18(1), 7–19.

Siewert, M. B., 2015. High-resolution mapping of soil organic carbon storage and soil properties

in Siberian periglacial terrain. Licentiate, Stockholm University, Stockhom, Sweden.Siewert, M. B., Krautblatter, M., Christiansen, H. H., and Eckerstorfer, M., 2012. Arctic rock-

wall retreat rates estimated using laboratory-calibrated ERT measurements of talus cones inLongyeardalen, Svalbard. Earth Surface Processes and Landforms, 37(14), 1542–1555.

Smith, C. A. S., Smits, C. M. M., and Slough, B. G., 1992. Landform Selection and Soil Modi-fications Associated with Arctic Fox (Alopex lagopus) Den Sites in Yukon Territory, Canada.Arctic and Alpine Research, 24(4), 324–328.

Soloviev, P. A., 1973. Thermokarst phenomena and landforms due to frost heaving in centralYakutia. Biuletyn Peryglacjalny, 23, 135–155.

Staff, S. S., 1999. Soil Taxonomy: A Basic System of Soil Clas- sification for Making and

Interpreting Soil Surveys, volume 436 of Agriculture Handbook. United States Department ofAgriculture & Natural Resources Conservation Service, Washington, DC, 2nd ed. edition.

Staff, S. S., 2014. Keys to Soil Taxonomy. United States Department of Agriculture & NaturalResources Conservation Service, Washington, DC, 12th ed. edition.

Strauss, J., Schirrmeister, L., Grosse, G., Wetterich, S., Ulrich, M., Herzschuh, U., and Hub-berten, H.-W., 2013. The deep permafrost carbon pool of the Yedoma region in Siberia andAlaska. Geophysical Research Letters, 40(23), 2013GL058088.

Tarnocai, C. and Stolbovoy, V., 2006. Northern Peatlands: their characteristics, developmentand sensitivity to climate change. In I. P. Martini, A. Martínez Cortizas, and W. Chesworth,editors, Peatlands Evolution and Records of Environmental and Climate Changes, volume 9of Developments in Earth Surface Processes, pages 17–51. Elsevier.

Tarnocai, C., Kimble, J., Broll, G., Philips, M., Springman, S. M., and Arenson, L., 2003. De-termining carbon stocks in cryosols using the Northern and Mid Latitudes soil database. InPermafrost, pages 1129–1134.

Tarnocai, C., Canadell, J. G., Schuur, E. A. G., Kuhry, P., Mazhitova, G., and Zimov, S., 2009.Soil organic carbon pools in the northern circumpolar permafrost region. Global Biogeochem-

ical Cycles, 23(2), 11.Team, R. C., 2016. R: A language and environment for statistical computing. R Foundation for

Statistical Computing, Vienna, Austria.Tobler, W. R., 1970. A Computer Movie Simulating Urban Growth in the Detroit Region. Eco-

nomic Geography, 46, 234–240.Trudgill, S., 2004. Catena. In A. S. Goudie, editor, Encyclopedia of Geomorphology, pages

122–124. Routledge, New York.Van Everdingen, R. O., editor, 1998. Multi-language glossary of permafrost and related ground-

ice terms. National Snow and Ice Data Center, Boulder, CO.van Huissteden, J., Maximov, T., Kononov, A., and Dolman, A., 2008. Summer soil CH4 emis-

sion and uptake in taiga forest near Yakutsk, Eastern Siberia. Agricultural and Forest Meteo-

rology, 148(12), 2006–2012.Vandenberghe, J., French, H. M., Gorbunov, A., Marchenko, S., Velichko, A. A., Jin, H., Cui, Z.,

Zhang, T., and Wan, X., 2014. The Last Permafrost Maximum (LPM) map of the NorthernHemisphere: permafrost extent and mean annual air temperatures, 25-17 ka BP. Boreas,43(3), 652–666.

Virtanen, T. and Ek, M., 2014. The fragmented nature of tundra landscape. International Journal

of Applied Earth Observation and Geoinformation, 27, Part A, 4–12.Walker, D. A., Raynolds, M. K., Daniëls, F. J., Einarsson, E., Elvebakk, A., Gould, W. A.,

Katenin, A. E., Kholod, S. S., Markon, C. J., Melnikov, E. S., Moskalenko, N. G., Talbot,

53


S. S., Yurtsev, B. A., and other members of the CAVM Team, T., 2005. The CircumpolarArctic vegetation map. Journal of Vegetation Science, 16(3), 267–282.

Warburton, J., 2013. 8.19 Patterned Ground and Polygons. In Treatise on Geomorphology, pages298–312. Elsevier.

Washburn, A. L., 1956. Classification of Patterned Ground and Review of Suggested Origins.Geological Society of America Bulletin, 67(7), 823–866.

Weiss, N., Blok, D., Elberling, B., Hugelius, G., Jørgensen, C. J., Siewert, M. B., and Kuhry, P.,2016. Thermokarst dynamics and soil organic matter characteristics controlling initial carbonrelease from permafrost soils in the Siberian Yedoma region. Sedimentary Geology, 340,38–48.

Wilkinson, G. N. and Rogers, C. E., 1973. Symbolic Description of Factorial Models for Analysisof Variance. Journal of the Royal Statistical Society. Series C (Applied Statistics), 22(3), 392–399.

Williams, R. B. G., 1988. The biogeomorphology of periglacial environments. In H. A. Viles,editor, Biogeomorphology, pages 222–252. Basil Blackwell, Oxford, UK ; New York, NY,USA.

Wolfe, S. A., Gillis, A., and Robertson, L., 2009. Late Quaternary eolian deposits of northernNorth America: Age and extent. Technical Report 6006.

Woodcock, C. E. and Strahler, A. H., 1987. The factor of scale in remote sensing. Remote

sensing of Environment, 21(3), 311–332.Zhang, T., Barry, R. G., Knowles, K., Heginbottom, J. A., and Brown, J., 1999. Statistics and

characteristics of permafrost and ground-ice distribution in the Northern Hemisphere. Polar

Geography, 23(2), 132–154.Zimov, N. S., Zimov, S. A., Zimova, A. E., Zimova, G. M., Chuprynin, V. I., and Chapin, F. S.,

2009. Carbon storage in permafrost and soils of the mammoth tundra-steppe biome: Role inthe global carbon budget: CARBON STORAGE OF THE MAMMOTH STEPPE. Geophysi-

cal Research Letters, 36(2), 6.Zimov, S., Zimov, N., Tikhonov, A., and Chapin, F., 2012. Mammoth steppe: a high-productivity

phenomenon. Quaternary Science Reviews, 57, 26–45.Zimov, S. A., Voropaev, Y. V., Semiletov, I. P., Davidov, S. P., Prosiannikov, S. F., Chapin, F. S.,

Chapin, M. C., Trumbore, S., and Tyler, S., 1997. North Siberian Lakes: A Methane SourceFueled by Pleistocene Carbon. Science, 277(5327), 800–802.

Zimov, S. A., Davydov, S. P., Zimova, G. M., Davydova, A. I., Schuur, E. A. G., Dutta, K., andChapin, F. S., 2006. Permafrost carbon: Stock and decomposability of a globally significantcarbon pool. Geophysical Research Letters, 33(20), L20502.

Zoltai, S. C., 1972. Palsas and Peat Plateaus in Central Manitoba and Saskatchewan. Canadian

Journal of Forest Research, 2(3), 291–302.Zubrzycki, S., Kutzbach, L., Grosse, G., Desyatkin, A., and Pfeiffer, E.-M., 2013. Organic

carbon and total nitrogen stocks in soils of the Lena River Delta. Biogeosciences, 10(6),3507–3524.

Zubrzycki, S., Kutzbach, L., and Pfeiffer, E.-M., 2014. Permafrost-affected soils and their carbonpools with a focus on the Russian Arctic. Solid Earth, 5(2), 595–609.

Åkerman, H. J. and Johansson, M., 2008. Thawing permafrost and thicker active layers in sub-arctic Sweden. Permafrost and Periglacial Processes, 19(3), 279–292.

Åkerman, H. J. and Malmström, B., 1986. Permafrost Mounds in the Abisko Area, NorthernSweden. Geografiska Annaler. Series A, Physical Geography, 68(3), 155–165. ArticleType:research-article / Full publication date: 1986 / Copyright c© 1986 Swedish Society for An-thropology and Geography.

54

high-resolution mapping and spatial variability of soil

Documents