Characterization of the spatial and temporal variability in pan-Arctic, terrestrial hydrology

University of New Hampshire

University of New Hampshire Scholars' Repository

Doctoral Dissertations

Student Scholarship

Winter 2006

Characterization of the spatial and temporal

variability in pan-Arctic, terrestrial hydrology

Michael A. Rawlins

University of New Hampshire, Durham

Follow this and additional works at:

https://scholars.unh.edu/dissertation

Recommended Citation

Rawlins, Michael A., "Characterization of the spatial and temporal variability in pan-Arctic, terrestrial hydrology" (2006).Doctoral Dissertations. 358.

C H A R A C T E R IZ A T IO N OF T H E SPA T IA L A N D

T E M P O R A L V A R IA B IL IT Y IN P A N -A R C T IC ,

T E R R E S T R IA L H Y D R O L O G Y

BY

M ichael A. R aw lins

Bachelor of Science, University of Delaware, 1996

Master of Science, University of Delaware, 2001

DISSERTATION

Submitted to the University of New Hampshire

in partial fulfillment of

the requirements for the degree of

Doctor of Philosophy

in

Earth and Environmental Science

UMI Number: 3241649

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction.

In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.

®

UMI

Copyright 2007 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against

unauthorized copying under Title 17, United States Code.

ProQuest Information and Learning Company 300 North Zeeb Road

P.O. Box 1346 Ann Arbor, Ml 48106-1346

This dissertation has been examined and approved.

i /■

Dissertation Director, Dr. Charles J. Vorosmarty

Research Pyofessor of Earth Sciences and Earth, Oceans,

and Space 1

/

Dr. S. Lawrence Dingman

Professor Emeritus of Hydrology and Water Resources

Dr. Mark Fahnestock

Research Associate Professor of Earth, Oceans, and

Space

Dr. Steve Frolking

Research Associate Professojf of Earth Sciences and

Earth, Oceans, and Space

Dr. Ernst Linder

Professor of Mathematics

A C K N O W L E D G E M E N T S

I would like to acknowledge and thank Charlie Vorosm arty for encouraging and advising me over the past several years. Dr. Vorosm arty always suggested th at I think broadly, and he challenged m e and to pursue interesting and exciting research. I have benefited trem endously as a result of our work together and for that I am thankful.

I thank members of the Water System s Analysis Group including Richard Lammers, Balazs Fekete, Alexander Shiklomanov, and W il Wollheim. Stanley Glidden and Alexander Prusevich are thanked for their technical assistance, and Darlene D ube has been helpful in too m any ways to mention here. I also thank Cort W illm ott, Mark Serreze, and Kyle M cDonald for the fruitful collaborations and discussions we have had over the past several years. I thank com m ittee members Larry Dingm an, Mark Fahnestock, Steve Frolking, and Ernst Linder for their assistance with m y dissertation studies.

This research was supported under N SF grants O0094532, O 0230243, O PP-9818199, and OPP-9910264; NA SA grants NAG5-9617, NAG5-6137, NAG5-11256, and NAG5-11750; and the ARCUS Program.

Lastly, I would like to thank my fam ily for their love and support. A special word of thanks is reserved for my mother for encouraging me to attend college when I seemed reluctant.

TA B L E OF C O N T E N T S

IN TR O D U C T IO N

1

1 SIM ULATING PA N -A R C TIC R U N O FF W ITH A MACRO-SCALE TE R ­

RESTRIAL W ATER BA LA N C E MODEL

8

1.2 The Pan-Arctic Water Balance M o d e l... 11

1.3 M odel Results ...15

1.3.1 Active-layer M o d e lin g ...15

1.3.2 Pan-Arctic Runoff ... 18

1.4 Sensitivity A n a ly sis...24

2 REM OTE SENSING OF SNOW THAW AT TH E PA N -A R C TIC SCALE

USING THE SEAW INDS SCATTEROM ETER

31

2.1 Rem ote Sensing of Snow ...31

2.2 D ata Sources and M e t h o d s ... 36

2.2.1 Radar B a c k s c a tte r ... 38

2.2.2 Hydrological D a ta ... 41

2.2.3 Land Surface D a t a ... 43

2.3 Comparison of SeaW inds-Derived Thaw T im ing and Hydrological R e sp o n se ... 44

2.4 Comparison of Snow Thaw Tim ing across the Pan-Arctic D o m a in ... 49

2.5 Discussion of Results ...57

3 EVALUATION OF TR EN D S IN D ER IV ED SNOWFALL A N D R A IN ­

FALL ACROSS EU R A SIA

60

3.2 D ata and M ethods ...61

3.3 Linkages between Precipitation and D ischarge... 63

4 EFFECTS OF U N C ER TA IN TY IN CLIMATE IN P U T S ON SIM ULATED

EV A PO TR A N SPIR A TIO N A N D R UNO FF IN THE W EST ER N ARC­

TIC

71

4.2.4 M odel Application and A n a ly s is ...78

4.3 R esu lts...81

4.3.1 Simulated E v a p o tra n sp ira tio n ... 81

4.3.2 Simulated R u n o f f ... 82

4.4 D iscu ssio n ... 86

4.5 C onclusions... 90

5 ON THE EVALUATION OF SNOW WATER EQUIVALENT ESTIMATES

OVER THE TERRESTRIAL ARCTIC D R A IN A G E B A SIN

91

5.3 R e su lts... 97 5.4 C onclusions... 105

SUM M ARY

107

BIBLIO G R A PH Y

114

A PPE N D IC E S

138

A P P E N D IX A PW B M

139

A P P E N D IX B PE T FU N C T IO N S

146

A P P E N D IX C SOURCE CODE

148

LIST OF T A B L E S

1.1 Sim ulated long-term annual ru n off... 21

3.1 Trends in annual precipitation, annual snowfall, and annual ra in fa ll...66

4.1 Clim ate data sets used in PW B M sim u la tio n s...76

4.2 Long-term mean simulated ET, runoff, and clim ate across the WALE

domain ... 76

4.3 Observed June-A ugust ET at 2 sites in Alaska and simulated June-A ugust

E T ...78

4.4 Mean absolute difference (M AD) in sim ulated annual runoff as compared to

observed r u n o f f ... 87

5.1 Mean February-March SWE, percent negative correlations, and minimum,

maximum, and mean coefficient of d eterm in a tio n ... 102

LIST OF F IG U R E S

1-1 Pan-Arctic domain by sea basin b ou n d aries... 13

1-2 Schematic o f the PW B M soil zones and water fluxes in W inter/Spring and S u m m er...14

1-3 Simulated active layer th ic k n e ss...16

1-4 PW BM long-term m onthly runoff climatology, 1 9 8 0 -2 0 0 1 ... 19

1-5 PW BM sim ulated annual, long-term m ean runoff for years 1980-2001 ...20

1-6 D istribution of sim ulated and observed annual (long-term) ru n off... 23

1-7 Long-term seasonal runoff for Arctic sea b a sin s ...24

1-8 Sensitivity of sim ulated annual r u n o ff... 27

2-1 M oderate Resolution Imaging Spectroradiometer (MODIS)-derived vegetation co v er... 37

2-2 Behavior of Ku-band radar backscatter w ith increasing snow wetness ... 39

2-3 Observed basin-average runoff, m odel snow water content, and daily SeaW inds backscatter... 45

2-5 Primary thaw date ( t p) for year 2000 derived from the SeaW inds

b a c k sc a tte r ... 50

2-6 Snow water initiation date (tj[,f) for year 2000 derived from the pan-Arctic

Water Balance M o d e l... 51

2-7 Difference between PW B M snow water initiation date (ijv/) and

scatterom eter derived primary thaw date ( t p )...52

2-8 Observed basin-average runoff, m odel snow water content, and daily

SeaW inds backscatter for the Hanbury b a s in ...53

2-9 Cumulative area histogram s for Im — t p discrepancies...55

2-10 Average signal to noise value (R ) grouped by percent tree cover and total

1999-2000 winter s n o w fa ll... 56

3-1 Five-year running means of spatially averaged river d is c h a r g e ... 64

3-2 Spatially averaged water equivalent of annual rainfall and sn ow fall... 65

3-3 Trends in derived annual snowfall (a) and in derived annual rainfall (b),

1936-1999 ... 67

3-4 Annual total P for 1972 ... 68

4-1 Annual total precipitation (upper panel) and air tem perature (lower panel)

across the WALE domain ... 77

4-3 PW B M sim ulated annual E T across the WALE d o m a in ... 83

4-4 Simulated annual runoff across the WALE d o m a in ... 85

4-5 Long-term mean m onthly runoff across the Yukon b a s in ... 86

4-6 Observed and sim ulated annual runoff across the Yukon b a s in ... 87

5-1 Locations of stations used in s t u d y ... 95

5-2 M onthly total SW E and mean discharge (Q) across the Yukon b a s in ... 98

5-3 Explained variance (R2) for pre-melt SW E and spring Q com p a riso n s...100

A B S T R A C T

C H A R A C T E R IZ A T IO N OF TH E SPATIAL A N D T E M PO R A L

V A R IA B IL IT Y IN P A N -A R C T IC , T E R R E ST R IA L H Y D R O L O G Y

by

M ichael A . R aw lins

University of New Hampshire, December, 2006

Arctic hydrology represents an important component of the larger global clim ate sys­ tem , and there are signs that significant water-cycle changes, involving complex feedbacks, have occurred. This dissertation explores the m ethods to estim ate com ponents of the arctic hydrological cycle, the numerous biases and uncertainties associated with the techniques, and suggestions for future research needs. T he studies described here focus on quantita­ tive m odels and m ethods for predicting the spatial and tem poral variability in pan-Arctic hydrology.

This dissertation discusses pan-Arctic water budgets drawn from a hydrological model which is appropriate for applications across the terrestrial Arctic. Including effects from soil-water phase changes results in increases in sim ulated annual runoff of 7% to 27%. A sensitivity analysis reveals that sim ulated runoff is far more sensitive to the time-varying cli­ m ate drivers than to param eterization of the landscape. W hen appropriate clim ate data are used, the Pan-Arctic Water Balance M odel (PW BM ) is able to capture well the variability in seasonal river discharge at the scale of arctic sea basins.

T his dissertation also dem onstrated a m ethod to estim ate snowpack thaw tim ing from radar data. Discrepancies between thaw tim ing inferred from the microwave backscatter data and th e hydrological model are less than one week. The backscatter signal-to-noise values are highest in areas of higher seasonal snow accumulation, low to m oderate tree cover and low topographic complexity. An evaluation of snow water equivalent (SW E) estim ates drawn from land surface m odels and microwave remote sensing data suggests that simulated SW E from a hydrological model like PW BM , when forced with appropriate clim ate data, is far superior to current snow m ass estim ate derived from passive microwave data.

Biases arising from interpolations from sparse, uneven networks can be significant. A bias of well over + 1 0 mm yr-1 was found in the early network representations of spatial precipitation across Eurasia. W hen exam ining linkages between precipitation and river discharge, these biases limit our confidence in the accuracy of historical precipitation re­ constructions. This dissertation assess our current capabilities in estim ating com ponents of arctic water cycle and reducing the uncertainties in predictions of arctic clim ate change.

IN T R O D U C T IO N

The terrestrial Arctic is vast and rem ote region which has has experienced unprecedented change. T he arctic water cycle is an integral com ponent of the larger, global energy and water cycles, and alterations in the arctic system can feedback or impact global clim ate in ways which are not fully understood. A lthough the Arctic Ocean contains only 1% of the world ocean water, it receives 11% of the world runoff (Shiklomanov et al., 2000). It also has the largest contributing basin area, relative to the ocean surface area, of any of the world’s oceans. Yet, despite its prominent role in the global clim ate system , observations o f key hydrological quantities across the terrestrial arctic drainage basin have recently declined. Between 1986 and 1999, the area monitored has declined approximately 7%, from 74% to 67% of the pan-Arctic (Shiklomanov et al., 2002). This loss of information— vitally important for m odeling calibration and validation efforts— seriously compromise our ability to understanding the pan-Arctic system at this critical tim e.

More than a century ago, scientists speculated that combustion of fossil fuels will in­

crease the level of CO2 in the atmosphere. Over the past century, mean global, surface

air tem perature has increased by about 0.6°C (Houghton et ah, 2001). M odel projections suggest an increase in global tem perature, relative to 1990, of about 2°C by 2100. General circulation m odels (GCMs) generally agree that warming will be m ost pronounced across northern high latitudes during winter. Observations since 1960 confirm that warming in the Arctic has been strongest in winter and spring, amounting to as much as 2°C decade-1

for areas of greatest warming. Although the m ost recent warming may be attributable to multiple causes, paleoclim atic reconstructions indicate that the Arctic was warmer during the 20th century than at any tim e since 1600 (Overpeck et al., 1997).

Although polar amplification of global warming is a well known feature, of potentially greater im portance are hydrologic changes th at are likely to accompany a warmed Arctic. Recent assessm ents of 20th-century precipitation (Houghton et al., 1996) show that precip­ itation has increased by a greater percentage in the Arctic (65°N -85°N ) than in any other latitudinal zone on the globe. The Arctic has becom e wetter as well as a warmer, sug­ gesting an acceleration o f the hydrologic cycle. Annual river discharge across the 6 largest Eurasian river basins has increased by 7% from 1936 to 1999 (Peterson et al., 2002), and there are indications that river discharge is occurring earlier in the spring in m any arctic rivers (Lammers et al., 2001).

Climate change has the potential to affect arctic hydrology further through complex linkages and feedbacks. Warming air tem perature has been implicated in the record sea ice minimums of the past several years. T he significant downward trend of 8% decade-1 since the late 1970s has led to a reduction of approximately 20% in sea ice extent in September, when the annual minimum occurs (Manabe et al., 2005). Thawing of ice-rich permafrost— a consequence of increasing air and soil tem peratures— may lead to significant increases in river discharge. Simulations w ith the Community Land M odel (CLM3) show an increase in discharge of 28% by 2100, m ostly due to increases in precipitation that exceed increases in evaporation (Lawrence and Slater, 2005). Approxim ately 15% of the increase, however, is attributed to contributions from thawing permafrost. These increases in river discharge

m ay lead to a freshening of the Arctic Ocean, reduction of the North Atlantic Deep Water formation, and a slowing o f the Atlantic thermohaline circulation (Broecker, 1997).

Current clim ate m odels are unable to capture sufficiently the spatial and tem poral dy­ namics in pan-Arctic hydrology (Waliser et al., 2005). M ost state-of-the-art clim ate models significantly overestim ate the snow m ass across the Northern Hemisphere, particularly in spring (Roesch, 2006). Regional clim ate m odels (RCM s) or offline land-surface m odels (LSMs) run at higher spatial scales may offer improvements over GCMs. However, since they are usually driven by a GCM, water balances predicted by RCMs and LSMs may not be much better. On th e other hand, hydrological m odels— data-rich and suitably physically based— currently offer the best tools for estim ating arctic water budgets at small to medium scales. Although simple bucket m odels (Robock et al., 1995) have performed comparably to complex biosphere m odels, the equations used are often empirical, which lim its their use in simulations involving future clim ate scenarios. The desire to implement physically-based algorithms is challenged by the lack of input data required to parameterize the relevant m odel equations. Moreover, hydrological m odels and LSMs are more sensitive to climate-driven perturbations than to the chosen biotic surface parameters (Beringer et al., 2002; Federer et al., 2003). Understanding the degree of com plexity in m odel structure neces­ sary to obtain reasonable water budget predictions is central to our goal of developing a predictive capability in arctic change studies.

Across the terrestrial Arctic, in situ snow depth observations are biased toward pop­

ulated areas and lower elevations, and gauge undercatch can be severe (Groisman et al., 1991; Yang et al., 2005). R em ote sensing techniques offer the potential to overcome m any of the challenges in direct m onitoring across remote arctic lands. Evapotranspiration has been

estim ated using the M oderate Resolution Imaging Spectroradiometer (MODIS) instrument (Nishida et al., 2003). Low spatial resolution, high tem poral revisit microwave radiometers and scatterom eters are well-suited for quantifying the tim ing of landscape freeze/thaw state (Kimball et al., 2001; M cDonald et al., 2004).

Among all hydrological quantities, estim ation of snow m ass currently has the best po­ tential from a remote-sensing perspective. A critical component of th e arctic hydrological cycle, seasonal snow cover stores large am ounts of energy and provides the principle source of freshwater in m any arctic com m unities (W hite et al., 2004). Snow cover is involved in m any feedbacks (Randall, 1994), the m ost critical being the surface albedo feedback (Hall, 2004). Rem ote sensing of snow m ass at microwave wavelengths can be achieved without the lim itations of optical-infrared sensors such as MODIS. Passive microwave sensors such as the Special Sensor Microwave Imager (SSM /I) have proven particularly useful for re­ trieving information on snowpack water storage (Armstrong and Brodzik, 2001; Derksen et al., 2003). Snowpack thaw tim ing can be inferred from passive radiometers and from active instruments (eg., SeaW inds on N A SA Q uikScat). Uncertainties in snow water equiv­ alent (SW E) and snow thaw tim ing estim ates arise due to the integration of m any different terrain and land cover features into a single grid-cell brightness tem perature value. These sub-pixel-scale features include grain size variability, high-density snow layers or ice lenses, and a significant lake cover fraction (Rees et al., 2006). Filling the gap in snow observations w ith remotely sensed data is dependent on a better understanding of regional biases due to landcover effects.

The primary objective of this stu dy is to evaluate our current ability to characterize the arctic terrestrial hydrological cycle. The following research objectives help achieve this goal:

• To m odify an existing hydrological m odel for application to the pan-Arctic basin

• To estim ate the tim ing in landscape thaw from remote sensing data

• To understand the driving m echanisms behind observed hydrological changes

B y assembling the requisite input data, a hydrological model, and rem ote sensing fields, the main objectives of this study can be addressed through a series of specific research questions:

1. W hat is the spatial and tem poral distribution of runoff across the pan-Arctic drainage basin

2. W hat is the effect of incorporating soil freeze/thaw processes in a hydrological model? Are the effects on simulated runoff due to clim ate driver data more im portant than differences due to specification of landscape properties?

3. How do uncertainties and errors in m odel forcing data affect sim ulated water budgets?

4. Can remote sensing provide information about landscape thaw tim ing sufficient to improve hydrological forecast and biogeochemical modeling of the Arctic? W hat are the primary influences on thaw tim ing estim ates across the pan-Arctic basin?

5. Have changes in precipitation played a prominent role in the observed discharge trends across Eurasia? How does the sparse observing network impact our ability to monitor the region?

6. Are current SW E estim ates derived from passive microwave observations able to cap­ ture the spatial and temporal variability in pan-Arctic SWE? Are SW E estim ates derived from a suitable hydrological m odel superior at this time?

This dissertation is organized into five chapters. T he first chapter describes the develop­ ment, coding, and testing of a version of the Water Balance M odel (W BM , Vorosm arty et al., 1996) suitable for simulations across the pan-Arctic drainage basin. D etails of a new soil moisture phase-change subm odel— added to more accurately represent the daily changes to soil liquid and solid water fractions— are also presented. Results of a sensitivity experim ent illustrate the m ost important model com ponents for sim ulating the arctic water cycle. A

version of this chapter was published in the journal Hydrological Processes (Rawlins et al.,

2003) in 2003.

The second chapter presents an evaluation pan-Arctic snow thaw tim ing estim ates which are derived from the SeaW inds instrum ent. Tim ing of thaw inferred from SeaW inds backscatter is compared w ith observed river discharge and runoff from the hydrological model described in Chapter 1. An exam ination of landscape factors which influence the backscatter signal provides information on the applicability of this instrum ent for monitor­ ing the tim ing of landscape thaw at the pan-Arctic scale. This study was published in the

The third chapter exam ines the role of precipitation in the observed discharge trends across Eurasia. Several recent studies (Ye et al., 1998; Frey and Smith, 2003; M cClelland et al., 2004) suggest that increased precipitation is the m ost plausible source for the dis­ charge anomaly. In this study, trends is both annual snowfall and rainfall— derived from common precipitation data sets— is examined and compared with the discharge trends. The effect of changing station networks on trends drawn from common gridded data sets is also

explored. A version of this study was published in Geophysical Research L etters (Rawlins

et al., 2006) in 2006.

The fourth chapter describes how several configurations of clim ate driver and potential evapotranspiration function affect sim ulated water budgets. This study also examine several biases in common clim ate data sets and how these uncertainties propagate through the simulated water budgets. Identifying these biases is important given known problems in commonly-used, large-scale precipitation data sets. Results from this study were published

in the journal E arth Interactions (Rawlins et al., 2005) in 2006.

The fifth chapter presents a m ethod for evaluating SW E data through the use of m onthly river discharge data. Comparisons of agreements between river discharge and snow mass drawn from (i) the hydrological model described in Chapter 1, (ii) another land surface m odel, and (iii) SSM /I data are presented. The study also explores the linkages between precipitation input to the landscape and the freshwater flux through the basin. A journal

article subm itted to a special issue of Hydrological Processes (Rawlins et al., 2005) focusing

C H A P T E R 1

S IM U L A T IN G P A N -A R C T IC R U N O F F W IT H A

M A C R O -SC A L E T E R R E S T R IA L W A T E R B A L A N C E

M O D EL

1.1

H yd rological M od elin g

Global-change scenarios have predicted significant positive increases in surface air tem per­ ature, with the greatest increases expected to occur in the Arctic (M anabe et al., 1991; Nicholls et al., 1996). Although much speculation surrounds the causes, feedbacks, and uncertainty in Arctic environmental change, a large body of evidence suggests that major changes have already occurred (Serreze et al., 2000; Vorosmarty et al., 2001). Increases in surface air tem perature over the next several decades may lead to significant changes in per­ mafrost active-layer thickness (Anisim ov et al., 1997). Thawing of permafrost-rich soils can dramatically alter landscape patterns, w ith a potential to release water and carbon stored in soils (Hinzman and Kane, 1992; Waelbroeck et al., 1997). Given the linkages between Arctic hydrology and numerous geophysical system s over a wide range of scales, along with recent evidence of significant change (Chapman and Walsh, 1993; Groisman et al., 1994; Oechel et al., 1993; SEARCH SCC, 2001), the mechanisms underlying major hydrological processes across the Pan-Arctic deserve considerable attention. Large variations in

river-ine exports to the Arctic Ocean have the potential to alter global ocean and atmospheric circulations (Broecker, 1997; Schiller et al., 1997) as well as oceanic net carbon storage (Anderson et al., 1998). Major changes in runoff and freshwater export can also affect the biogeochem istry of Arctic aquatic ecosystem s (Holmes et al., 2000; W olheim et al., 2001). And although it contains only 1 % of the world ocean water, the Arctic Ocean receives 11 % of th e global river runoff (Shiklomanov, 1998).

Models that sim ulate water budgets at continental and global scales have been widely used in hydrology and earth science research (Roads et al., 1994; Vorosm arty et al., 1998; Nijssen et al., 2001). Mintz and Walker (1993) applied a sim ple bucket m odel to derive global fields of m onthly soil moisture. Pitm an et al. (1999) employed a land surface model to estim ate the effects of frozen soil m oisture parameterizations on sim ulated runoff. A hydrology m odel was used to evaluate the water budgets of clim ate m odel simulations (Maurer et al., 2001), revealing significant biases in the clim ate m odel fields. A similar overprediction of evapotranspiration and underprediction of runoff from a clim ate model land-surface scheme was found across the Yenisei, Lena, and Amur basins in Asia (Arora, 2001). Improvement in global estim ates of river discharge were obtained using a new m ethod to determine runoff calibration parameters (Nijssen et al., 2001). Given the lack of observed river discharge data across large portions of the Pan-Arctic basin (Shiklomanov et al., 2002; Lammers et al., 2001), hydrological m odels which adequately capture the Arctic water cycle are needed to provide accurate benchmarks and aid in environmental change studies. Further, given the significant bias in runoff generated by current GCMs (Walsh et al., 1998), accurate tim e series of sim ulated seasonal runoff routed through a Simulated Topological

Network (STN ) (Vorosmarty et al., 2000a; Vorosmarty et al., 2000b) offers the potential to improve freshwater forcing in coupled ocean models.

Thawing and freezing of Arctic soils is affected by many factors, w ith soil surface tem ­ perature, vegetation, and soil m oisture among th e more significant (Zhang and Stam nes, 1998). Soil texture and slop e/asp ect also strongly influence active-layer dynam ics, which can vary considerably over short lateral distances. Indeed, differences in end-of-season mean thaw depths up to 50 % have been found when comparing two sites even in close ( < 1 0 km) proximity (Nelson et al., 1997).

Investigations of active-layer thickness (ALT) have traditionally been performed through field studies at point locations (Rom anovsky and Osterkamp, 1995; Zhang et al., 1996). Given the difficulty in compiling spatially coherent data sets of key input drivers, few studies have been conducted to model seasonal active-layer changes at the regional scale. Anisimov et al. (1997) applied a semi-empirical m ethod to calculate the depth of seasonal freezing and thawing using annual air tem perature, snowcover, vegetation, soil m oisture, and thermal conductivity parameterizations. More com plicated models, which sim ulate heat flow and phase change, have been used to investigate the sensitivity of soil therm al processes to air tem perature, seasonal snow cover, and soil m oisture (Zhang and Stam nes, 1998), Although detailed models are helpful in understanding the effects of climatic and landscape factors, sim ple m odels may be useful in estim ating changes in ALT, particularly for large-scale applications. The Stefan solution to the differential equation of heat transfer with phase change under constant conditions (e.g. Lunardini, 1981) shows that ALT progresses as the square root of tim e. This approach has been applied across the Kuparuk basin (2100 km2) in northern Alaska (Nelson et al., 1997). Klene et al. (2001) found that incorporating the

effects of vegetation on soil tem peratures could improve this m ethod. In addition to the difficulty in compiling accurate input data sets to m odel soil thawing and freezing, a lack of empirical observations for validation of sim ulated estim ates presents a further challenge for Pan-Arctic applications (Vorosmarty et al., 2001).

Our focus in this paper is the estim ation of runoff across the Pan-Arctic drainage basin for the period 1980-2001. A simple sub-m odel for estim ating phase changes in soil m oisture is described and evaluated by comparing m odel-estim ated active-layer thickness to field measurements from several locations in Alaska. M odel simulated runoff is then presented and compared to observed data. Sensitivity of sim ulated runoff to variations in clim ate inputs and m odel parameterization are also investigated to identify the m ost sensitive model requirements.

1.2

T h e P a n -A rctic W ater B alan ce M od el

Large-scale numerical m odels which sim ulate the hydrologic cycle have recently been devel­ oped to characterize m oisture fluxes and storage across diverse landscapes (Nijssen et al.,

2001; Zhuang et al., 2001). A modified version of the water balance m odel (W B M )

(Vorosmarty et al., 1996; Vorosmarty et al., 1998) was applied across the Pan-Arctic to study the spatial and tem poral variability of the high-latitude terrestrial water cycle, with significant changes incorporated into this version— henceforth referred to as the Pan-Arctic

W a ter B a la n c e M o d e l (P W B M )— d e ta ile d in th e A p p e n d ix A.

Models which sim ulate water and energy balance at fine vertical resolution within the soil have been developed and show promise in estim ating soil thermal regimes (Zhuang et al.,

2001) as well as water balance (Bruland et al., 2001) in Arctic regions. Simple “bucket” m odels, however, have been shown to perform comparably to com plex biosphere m odels in estim ating soil m oisture (Robock et al., 1995). A fundam ental premise in the development and modification of the PW B M is that for large-scale spatial applications there are severe lim itations in basic data quality needed to parameterize and drive a hydrological m odel (e.g., precipitation, soil properties, vegetation characteristics such as leaf area and rooting depth), so developing a simple, suitably-scaled m odel is appropriate. T he m odel should balance physically-based sim ulations of hydrological processes w ith the practical lim its of soil and vegetation parameterizations and meteorological drivers. To this end, PW BM is data-rich, suitably physically based, and well scaled to the challenges of water budget estim ation over the Pan-Arctic. Our model does not explicitly sim ulate glacier accumulation and melt. Therefore, runoff for areas dom inated by glaciers and ice fields are expected to have substantial error.

In this study, estim ates of snow water equivalent, soil ice and water stores, along w ith fluxes such as evaporation, evapotranspiration, and runoff are made with the PW B M at explicit daily tim e steps across the Pan-Arctic drainage basin, defined as all land areas draining to the Arctic Ocean in Russia and Canada, as well as Hudson Bay and the Bering Sea (Figure 1-1). T he PW BM requires spatial data sets of vegetation cover, plant rooting depth, soil texture, soil depth, and soil carbon content. Gridded fields of daily air tem pera­ ture and precipitation drive the PW BM . Input data (parameter fields, air tem perature, and precipitation) and m odel output is gridded at 25 km resolution on the Lambert Azim uthal equal area Grid (NSIDC, 1995; Brodzik and Knowles, 2002). A total of 39,926 EASE-Grid pixels defines the Pan-Arctic drainage basin, which extends as far south as 45° N in

Figure 1-1: Pan-Arctic domain by sea basin boundaries. Basin boundaries are derived from a digital river network at 30 m inute grid cell resolution.

southern Canada (Nelson Basin) and southern Siberia (Ob Basin). Air tem perature and precipitation inputs are derived from the National Center for Environmental Prediction (NCEP) reanalysis project (Kalnay et al., 1996; Uppala et al., 2000). The N C EP-N C A R reanalysis constitutes a retrospective record of numerical weather prediction (N W P) anal­ ysis and forecasts, with the added advantage of being constantly updated with minimal (1 month) tim e lag. Six-hourly N C E P data are aggregated to daily means and interpolated to the 25 km EASE-Grid using a statistical downscaling approach (Serreze et al., 2002). Usage of data sets for vegetation cover (Mellilo et al., 1993), soil texture (Food and Agriculture O rganization/UN ESC O , 1995), and rooting depth is based on the m ethodologies originally reported in Vorosmarty et al. (1989). D ata for soil organic content were obtained from the Oak Ridge National Laboratory (ORNL) (Global Soil D ata Task, 2000).

Snowpack Snow Content

Water Content

Sum m er Profile

r

M

r&sT’

ss

_______ i

Figure 1-2: Schematic of the PW B M soil zones and water fluxes in W inter/Spring and Sum­ mer. Root Zone represents the spatially-variable vegetation rooting depth. Water in root and deep soil zones can be all frozen, partially frozen, or all liquid. In some locations/cells, th e deep zone never fully thaws and in others, it never fully freezes.(M odified from (Holden, 1999)).

PW BM has two soil layers, a root zone that gains water from infiltration and loses water via evapotranspiration and horizontal and vertical drainage, and a deep zone that gains water via root zone vertical drainage and loses water via horizontal drainage (Figure 1-2). Seasonal changes in soil w ater/ice content are an important component of Arctic hydrology (Woo, 1998), so specification of phase changes in soil moisture is a key component of the PW BM . Soil liquid water and ice contents of each soil layer are calculated in a submodel referred to as the Thaw-Freeze M odel (TFM ). Because PW BM does not sim ulate vertical heterogeneity within either soil layer, it does not explicitly track the depth of thawing or freezing, but instead uses the Stefan solution to update daily changes in the amount of liquid and frozen water of each soil layer. T he sign of the daily thaw /freeze increment determines the exchange of water and ice w ithin each layer. Details of the TFM are presented in A ppendix A.

1.3

M o d el R esu lts

1.3.1 A ctive-layer M odeling

To evaluate the efficacy of the Stefan solution, sim ulated active-layer estim ates from the two-layer TFM (equations A .4 and A .5) w ithin the PW B M framework are examined. Simulated active-layer development (for a single grid located in northern Alaska) progresses from the organic layer (depth = 23 cm) to mineral soil through the warm season (Figure l-3 a ). This Figure also shows the effect of soil m oisture variability. After 600°C-days had accumulated, ALT for this grid was 38.4 cm for 1999 (drier) and 42.9 cm for 1996 (wetter) conditions, with the variation attributed to differences in thermal conductivity of wet and dry soils. For the purposes of comparison to data in Zhang (1997), a linear regression m odel fit through the estim ates in Figure l-3 a reveals a rate of change approximated by ALT = 0.058 D D T (r = 0.99), where D D T is accumulation of degree days of thawing (°C-day). Zhang et al. (1997) examined 17 observation at 3 locations across northern Alaska (1987-1992), and found ALT = 0.046 D D T (r = 0.75), a difference of 1.2 mm per 10°C-days from the TFM estim ates.

The Stefan solution to heat transfer w ith phase change in one dimension (vertical) provides a sim ple estim ate of ALT (Lunardini, 1981). Nelson et al. (1997) used an empirical ALT similar to the Stefan solution to determ ine ALT estim ates w ithin ~ 6 c m of observed values. Here we compare gridded estim ates from the two-layer Stefan solution (in the TFM ) to a set of observed data from the Circumpolar A ctive Layer Monitoring network (CALM) (Brown et al., 2000). Various sampling strategies are represented in the CALM data set, w ith maxim um summer ALT determined as an average of samples across relatively small areas (10 m lattice within a 100 m 2 area) as well as larger sampling designs (100m lattice

1996 1997 1998 1999 (°) organic soil mineral soil 100 200 300 400

D egree Days (°C d ay)

500 600 700 BOO s o CO 'e_o o <0 ■o £ o 3 E in s 8 s ' ,4* — 1km grid •

/ 100m (or smoller) grid ♦

Average of CALM standard deviations ” 12.0 cm Average absolute difference (sim vs. obs) = 12.4 cm

o

0 20 40

Observed ALT (cm )

60 80 100

Figure 1-3: Simulated active layer thickness (ALT) as a function of degree days for one EASE-Grid over North Slope, AK (a); and m odel predicted active-layer thickness vs. CALM observed depth (b ). Horizontal lines in lower panel represent one standard deviation on each side of observed ALT for the 1 km2 CALM sites.

within a 1 km2 area) in some locations. Simulated ALT values on the 25 km EASE-Grid encompassing each CALM validation site are compared to the maxim um summer CALM ALT for 27 sites in Alaska (years 1999 and 2000). Specifically, we use the TFM m odel value for the day on which the CALM estim ate was made. M odel estim ates are generally within one standard deviation of the observed value (Figure l-3 b ). A bias (underestimation) in simulated ALT is evident, which is likely attributable, in part, to a bias in the air temperature field that is adjusted to 25 km grid mean elevation, which is higher, and thus cooler, than CALM site elevation in m ost cases (data not shown). It should be noted that the CALM value used in each comparison represents a single point sample within the 625 km2 EASE-Grid and grid-to-point comparisons are known to create interpretation problems (Bloschl and Sivapalan, 1995; Klene et al., 2001; Vorbsmarty et al., 1998). And although an increasing number of CALM sites have begun to employ a gridded sampling design, the use of observed active-layer thickness estim ates made from a single observation should be undertaken with caution (Brown et al., 2000). Nonetheless, the average absolute error of the TFM model versus observations is 12.4cm (observed data range 32 cm to 72 cm), while the average standard deviation of samples from the 1 km2 grids (100 m lattice) is 12.0 cm. Given the variance in the observed data, TFM estim ates are within the variability seen in these field samples. Improvements in ALT estim ates using the Stefan solution in this manner have been achieved using higher resolution data sets across the Kuparuk basin in Alaska (Klene et al., 2001). Although there is no apparent bias between the comparisons with the 1 km2 CALM sites and those from smaller areas, there is evidence that the 100 m spacing is unable to resolve the variability in ALT at upland (North Slope, AK) sites (Nelson et al., 1999).

1.3.2

P an-A rctic R unoff

To estim ate runoff over the Pan-Arctic drainage basin, the PW B M was used to sim ulate the water cycle at daily tim e steps for each EASE-Grid across the domain. The model was run with inputs of air tem perature and precipitation for the year 1980, repeated for 50 years to stabilize soil m oisture content, followed by a transient run for the years 1980-2001. Climatologies of m onthly total runoff (Figure 1-4) show the progression of the annual pattern of runoff, from the spring snowmelt pulse, to low flow conditions, to freeze-up. M onthly runoff is relatively low across much of the terrestrial Arctic in winter w ith the exception of coastal western Canada and southern Alaska. Snowmelt contributes to higher runoff across Eurasia in April. Runoff increases in both m agnitude and extent during May in both hemispheres. T he m ost northern areas of Eurasia see the snowmelt driven runoff peak in June. In a general sense, this peak runoff progresses northward toward the Arctic Ocean through spring in central Eurasia, indicative of seasonal changes in surface air temperature. Summer rainfall then contributes to runoff through summer, however, higher evapotranspiration tends to produce relatively dry conditions. T he water cycle in fall and winter is dom inated by snowpack accumulation and low runoff am ounts.

Simulated long-term annual runoff (1980-2001) is highest across southern Alaska, coastal Norway, and Iceland. Higher runoffs are also found across southern parts of Canada in the Nelson basin and the Eurasian part of Russian. Lower runoffs are evident across the Canadian archipelago and Siberia (Figure 1-5). Runoffs exceeding 400 mm year-1 are noted across northeastern Canada and southern Alaska. Simulated long-term annual runoff across the largest Arctic drainage basins is approximately 100 to 180m m year-1 (Table 1.1). More

0 10 25 40 55 70 85 100 115 130 mm month-1

Figure 1-4: PW BM long-term m onthly runoff clim atology 1980-2001. T his figure includes runoff for southern Alaska, which is not part of the Pan-Arctic drainage per se. Grids with zero runoff for the m onth are shaded in blue. Runoff for areas w ith glaciers should be inter­ preted w ith caution, as the PW B M does not model glacier accumulation and m elt/ablation.

Figure 1-5: PW BM sim ulated annual, long-term mean runoff for years 1980-2001 across the 2 5 x l 0 6 km 2 land area of the Pan-Arctic drainage basin. Annual runoff is estim ated at the 39,926 EASE-Grids (equal area 25 km x 25 km) defining th e domain. Grids with zero annual runoff are shaded in blue.

variability, however, is seen in runoff to individual Arctic sea basins; runoff ranges from 90 mm year-1 (East Siberian Sea, Table 1.1) to as much as 300 mm year-1 (Hudson Strait).

Spatially-averaged sim ulated runoff is approximately 180 mm year-1 across the entire Pan-Arctic drainage basin. Sim ulated annual runoff is approximately 40-70 % of the annual downscaled precipitation (Serreze et al., 2002) across m any regions and is highly correlated w ith precipitation (r = 0 .9 0 ). Higher variability in observed runoff is apparent (Figure 1-6),

while the correlation w ith precipitation is lower (r = 0 .5 6 ). Observed runoff is generated by

distributing a basin’s discharge across the monitored region, including areas between gaging stations (inter-station areas) (Lammers et al., 2001). Observed discharge estim ates for the period 1980-1997 are from a data set of 650 gaging stations across the Pan-Arctic (Lammers et al., 2001; Shiklomanov et al., 2002). Discrepancies due to th e use of different averaging periods (simulated runoff from 1980-2001, observed from 1980-1997) are assumed to be

River Basin Basin Sizea Gaged Area6 S im u la t e d A n n u a l R unofiF (km2) (km2) (mm year- 1 ) Ob 2,994,238 2,965,100 180 Yenisei 2,537,404 2,452,300 170 Lena 2,460,742 2,460,000 100 Mackenzie 1,783,972 1,769,200 150 Yukon 833,232 831,391 120 Nelson 1,106,578 1,050,300 160 S e a B a s in C o n t r ib u t in g A r e a a (km2) G a g e d A r e a 6 (km2) S im u la te d A n n u a l R unoflff (mm year- 1 ) Arctic Archipelago 1,134,856 209,270 40 H udson Bay 3,304,025 2,613,320 270 Barents Sea 1,322,741 984,830 300 Hudson Strait 468,050 285,480 410 Beaufort Sea 2,139,635 1,860,100 130 South Greenland 1,174,444 10,800 250 Bering Strait 1,205,234 1,010,940 170 Kara Sea 6,631,308 5,159,700 200 Chukchi Sea 282,143 56,160 190 Laptev Sea 3,639,584 3,232,480 100

East Siberian Sea 1,329,025 941,500 90

Pan-Arcticd 22,611,659 16,460,080 180

Table 1.1: Simulated long-term annual runoff for selected Arctic drainage basins and ter­ restrial runoff integrated across basins draining to selected Arctic Ocean sea basins. aTotal area for river or sea basin on the EASE-Grid (NSIDC, 1995)

bArea captured by observed gaging stations (Lammers et al., 2001).

cAnnual runoffs in Table represent an integration across all EA SE grids in a given basin. dT he Pan-Arctic value represents the spatially-averaged runoff for all land areas draining to the Arctic Ocean in Russia, Canada, and Alaska, as well as Hudson Bay and the northern Bering Sea.

negligible. Mean values of the sim ulated and observed runoff distributions for precipitation between 300-900 mm (80% of total samples) are comparable (Figure 1-6). PW B M simu­ lated runoff is conservative (does not exceed precipitation) and the residual of precipitation minus runoff represents modeled evapotranspiration. Observed runoff, however, exceeds the inter-station-area precipitation in som e regions, im plying either considerable interbasin groundwater transfers or significant problems w ith the spatial precipitation and/or runoff data (Vorosmarty et al., 1998; Fekete et al., 1999).

Simulated and observed runoffs are further compared by aggregating to long-term sea­ sonal runoff for each Arctic sea basin. Seasonal runoff represents the total stock of freshwater which contributes to the sea basin’s seasonal riverine input. Here we compare the integrated runoff across all EA SE grids in a given basin to the observed runoff over the monitored por­ tion of that basin. Although underestim ates are again more comm on than overestimates, good correlation is evident (r = 0.84, Figure 1-7). The PW BM runoff estim ates are near zero in winter, and underestim ate m ost observed basin values. T his is likely due to sev­ eral factors, including groundwater inputs that do not freeze in winter, and lags in water transport that generate winter flow (observed at gaging stations) from fall runoff. For some basins, the m onitored area is only a small fraction of the total basin (Table 1.1), which could introduce a bias into the observed runoff, evidence that the decline in river discharge m onitoring across both North America and A sia (Shiklomanov et al., 2002) com plicates our m odel verification efforts.

o

o

o

Distribution of Obser ved and Sim ula te d Runoff

N = 4 8 2 1 5 * 2 3 3 63 20 O O 00 O O co O O CM O b serv ed Z ' / S im u late d

!

I

!

I

Figure 1-6: D istribution of simulated and observed annual (long-term) runoff at (n=579) inter-station areas for groupings of annual long-term precipitation. The top and bottom of each box are the 25th and 75th percentiles, respectively. Boxplot whiskers represent the 5th and 95th percentiles. T he spatial m ean is the thick line and the median is the thin line. Maximum and minimum runoff for each distribution is marked w ith an asterisk. The number of observed and simulated runoffs in each grouping is listed along the top of the Figure. M aximum value of observed runoff for 500-700 precipitation is 1915 mm and is not plotted. N ote that in all bins except 900-1100 mm yr- 1 , the maximum observed runoff is greater than annual precipitation.

S im u late d a n d O b serv ed S e a s o n a l R unoff by A rctic S e a B asin w inter • sp rin g • s u m m e r/f a ll * o o «M O O o 0 50 100 150 O b s e r v e d R u n o ff ( m m / s e a s o n ) 250 200 300

Figure 1-7: Long-term seasonal runoff for Arctic sea basins. Winter is December, January, February, March; spring is April, May, June, July; sum m er/fall is August, September, October, November.

1.4

S en sitiv ity A n alysis

Biogeophysical characteristics such as plant rooting depth, organic-layer thickness, and soil texture affect water flow paths and are integral factors in hydrological models. In addition, climate data (precipitation and air tem perature) are essential inputs with both spatial and temporal variations and uncertainties. Precipitation and air tem perature data for the Arc­ tic, however, are more poorly resolved (owing to the the sparsity of meteorological stations), relative to other parts of the world. In addition to being undersampled, biases in Arctic precipitation records are known to be large, particularly at higher latitudes. Underestim ates

of 20-25 % (Karl et al., 1993) and 10-140 % (Yang et al., 1998) have been determined across

North America and at 10 locations in Alaska, respectively. Substantial gage undercatch across the Arctic has also been estim ated by applying a hydrological model (Fekete et al.,

1999). Although N C EP reanalysis data have been adjusted for measurement biases (Serreze et al., 2002), some uncertainty in the model input can be assumed. Comparison of daily gridded N C EP air tem peratures in summer with observed meteorological data yielded ab­ solute differences from 1.6°C for an Arctic continental location and 6.6°C at a coastal site. In general, NC EP air tem peratures are consistently cooler than the station observations throughout summer.

To investigate the sensitivity of the PW BM to m odel param eterizations and clim ate inputs, long-term annual runoff (1980-2001) was compared against the long-term annual runoff produced in a series of m odel perturbations runs. T he following perturbation ex­ periments were performed: (i) m odel organic-layer depths were halved [0.5x Org] and (ii) doubled [2x Org]; (iii) vegetation rooting depths were halved [0.5x RD] and (iv) doubled [2x RD]; (v) soil field capacity was increased by 0.05 cm 3/c m 3 of pore space [FC + 25%]; (vi) horizontal and downward water flux from rooting zone (Figure 1-2) was increased to 40 % (from 20%) of water over field capacity [RF = 40%]; (vi) TFM subm odel was not applied [No TFM]; (viii) summer air tem peratures were increased by 4°C [T + 4°C] and (ix) daily precipitation was increased by 25 % [P + 25 %]. T he control run represents our best esti­ m ate of annual runoff (e.g., Figure 1-5) (with associated error characteristics as discussed above) using available fields of soil characteristics, NCEP-derived air tem perature and pre­ cipitation, and the TFM -generated active-layer behavior. Comparisons were examined for the Yukon, Nelson, M acKenzie, Ob, Yenisei, and Lena River basins. Differences between the control and sensitivity runs also were determined for the entire Pan-Arctic basin.

O f the nine sensitivity experim ents performed, an increase in precipitation produces the m ost significant changes in basin-average runoff. Adding 25% to each daily precipitation

occurrence increases Arctic-wide and basin runoff well over 50% (Figure 1-8), as additional precipitation is more likely to be diverted to runoff than evapotranspiration. Bias in pre­ cipitation inputs has been noted as a primary source of error in other large-scale hydrology m odels (Nijssen et al., 2001).

Increasing daily summer air tem peratures by 4°C also has a significant effect (albeit smaller than precipitation) across much of the Pan-Arctic. Annual runoff is reduced by more than 20 % across the Yukon, Lena, and Nelson basins (Figure 1-8). Warmer air temperatures result in higher rates of evapotraspiration (Equation A .l) , enhanced development of the active layer each spring/sum m er, an increased water holding capacity (which allows for more evapotranspiration) and, therefore, less runoff. The larger changes across the Yukon and Lena basins are expected considering the greater extent of permafrost conditions in these areas, relative to the other basins. A similar mechanism and m agnitude of effect occurs when the TFM sub-m odel is not used, effectively neglecting the seasonal thawing and freezing (i.e., changes to water-holding capacity) of Arctic soils. In this case, the absence of a shallow active-layer in late spring and early summer results in higher infiltration and summer evaporation, with a resultant reduction in annual runoff of 7 % for the Yenisei basin (least effect) to 27% for the Yukon basin (greatest effect). This result is consistent w ith a recent investigation of soil frost effects on catchm ent runoff, which found that ignoring soil frost tends to decrease total runoff (Stahli et al., 2001). However, two recent studies have questioned the importance of m odeling soil ice for runoff estim ation in forested environs

(Nyberg et al., 2001) and at large basin scales (Pitm an et al., 1999).

A s opposed to the perturbations to clim ate inputs (and the T F M ), changes to other m odel parameterizations result in relatively smaller changes (generally < 15 %) in annual

Sensitivity of Annual Runoff Climatology ( 1 9 8 0 —2 0 0 1 )

fro m Change to Model Geophysical Data Layer or Climate Input sensitivity o cn Wetter O in a> o c Q> L_ _o a> o o ro o c o 3 Cs| (X s — V*— o 3 c c < Drier O ro 2x RD FC + 5% RF - 4056 No TFM T + 4°C P + 25X Model P a ra m eter 0 .5 Org. 2x Org. 0 .5 RD

Figure 1-8: Sensitivity of sim ulated annual runoff to various alterations of input parameters.

Relative percent differences are defined by AR O = ( R O c — R 0P) / R O c * 100 % where AR O

is the relative difference in percent, R O c is the 1980-2001 runoff climatology, and R O p is the

1980-2001 runoff clim atology for perturbation or sensitivity run. Basins in analysis are O =

Ob, Y = Yenisei, L = Lena, M = Mackenzie, N = Nelson, K = Yukon, A —» = Pan-Arctic.

Perturbation experim ents are as follows: m odel organic-layer depths are halved [0.5x Org], doubled [2x Org], vegetation rooting depths are halved [0.5x RD], and doubled [2x RD], soil field capacity is increased by 0.05 cm 3/c m 3 of pore space [FC + 25%], horizontal and downward water flux from rooting zone is increased to 40 % (from 20 %) [RF = 40%], TFM subm odel is not applied [No TFM], summer air tem peratures are increased by 4°C [T + 4°C], and daily precipitation is increased by 25% [P + 25%].

runoff. Reducing organic layer depths enhances active-layer development, reducing runoff (Figure 1-8). A doubling of rooting depths, (as well as an increase in field capacity) also increases soil water-holding capacity, which lowers runoff. Runoff changes are negligible when the flux from the rooting zone is increased. These differences from the control runoff are significantly less than the standard deviation in annual runoff, further em phasizing the relatively small im pact of these model param eterization compared to the clim ate inputs.

1.5

Sum m ary and D iscu ssion o f R esu lts

A comprehensive understanding of Arctic hydrological system s has become im portant in light of recent evidence of the region’s environmental changes. Given the linkages involving water and carbon in terrestrial landscapes, the atmosphere and oceans, quantifying the Arc­ tic water cycle at continental scales allows us to establish baseline conditions and explore changes predicted to occur (SEARCH SCC, 2001). River discharge is highly undersampled across many of the higher latitudes in the Pan-Arctic drainage basin. W ith recent closures to a number of observed discharge m onitoring stations (Shiklomanov et al., 2002), m odel­ ing efforts which sim ulate runoff and freshwater flux to the Arctic Ocean can provide the requisite inputs to ocean m odels in lieu of observed data.

A Pan-Arctic Water Balance Model (PW BM ) has been developed and applied to esti­ m ate the water cycle at daily tim e steps for the 25 million km2 land area of the Pan-Arctic

d ra in a g e b a sin for t h e p e r io d 1 9 8 0 -2 0 0 1 . P h a s e c h a n g e s in so il m o istu r e w ere s im u la te d

w ith the Thaw-Freeze M odel (TFM ). These linked m odels utilize spatial fields of vegetation rooting depth, organic-layer depth, and soil textures, and are driven w ith clim ate data from

the N C EP reanalysis project. These spatial data sets are of varying quality, and m any regions are severely undersampled in all variables. Nonetheless, their com pilation and anal­ ysis provides a framework for evaluating consistencies between data sets (eg. runoff and precipitation) as well as a means for sim ulating the hydrological cycle.

Active-layer thicknesses generated w ith the TFM were compared with observed data from the Circumpolar A ctive Layer M onitoring (CALM) network. Simulated end-of-season active-layer thickness was generally w ithin the range of variability seen in the observed data; m odel biases were 12.4 cm, while the average standard deviation of the observed CALM estim ates is 12.0 cm. In large-scale studies of this nature, observed data validation sites, even 1 km2 grid sampling, represent point estim ates w ithin the larger (625 km2) PW BM grid. Considerable variability exist at this scale in all biophysical parameters, including seasonal n-factors, w ith soil-surface degree-day sum s varying up to 100 % within 1-ha plots (Klene et al., 2001). A lthough sim ulated m aximum summer ALT estim ates are generally w ithin the variability observed in the field samples, our interest centers on the day-to-day changes in active layer development used to determine phase changes of soil water. Recent studies (Anisimov et al., 1997; Klene et al., 2001) have suggested that improvements in active-layer simulation are dependent on the development of more spatially coherent data sets of air tem perature, vegetation, and soil moisture.

Simulated m onthly runoff is relatively low during winter when precipitation accumu­ lates as snow. A spring m elt pulse is evident in a south-to-north progression across the Arctic basin, with the m ajority of runoff occurring between the m onths of A pril-June. An­ nual long-termrunoff is highest across coastal western Canada, northeastern Canada and west central Eurasia. Lowest annual runoffs are seen across the Canadian archipelago and

Siberia. Simulated long-term annual runoff is less variable than observed runoff, and is highly correlated with precipitation. G ood correspondence was found when comparing sim­

ulated and observed seasonal runoff at individual Arctic sea basins (r = 0 .8 4 ). T his suggests

that the PW B M has the potential to provide the seasonal (tem poral) variations in freshwa­ ter discharge to ocean circulation m odels. Our sensitivity analyses show this m odel to be strongly influenced by clim ate drivers as well as the absence or presence of modeled active-layer changes, and that uncertainties in parameters such as rooting depth and organic active-layer thickness may be less problematic. As important as m odel development, it is essential for the research comm unity to work to improve spatial data sets for fundam ental biophysical variables and climatic drivers, and to maintain and expand river gaging in the Pan-Arctic to provide more com plete data sets for m odel evaluation. Simulation of the Arctic water cycle is notably influenced by specification of active-layer changes. T his finding suggests that modeling and analyses which depend on hydrological drivers such as ocean circulation, coastal processes, ecosystem biogeochemistry, and clim ate m odels will benefit from incor­ poration of thawing and freezing of Arctic soils. Gridded runoff fields are available from the Water System s Analysis Group, University of New Hampshire (h ttp ://w sa g .u n h .ed u ).

C H A P T E R 2

R E M O T E S E N S IN G OF S N O W T H A W A T T H E

P A N -A R C T IC SC A L E U S IN G T H E S E A W IN D S

S C A T T E R O M E T E R

2.1

R em o te S en sin g o f Snow

T he clim ate and hydrological cycle of the Arctic have undergone rapid change in recent decades (Serreze et al., 2000). Am ong the changes are increased winter air tem peratures (Chapman and Walsh, 1993; Rawlins and W illm ott, 2003; Robeson, 2004), reduced snow cover (Serreze et al., 2000), warming of permafrost (Osterkamp and Romanovsky, 1999; Romanovsky et al., 2002), increased runoff and river discharge (Peterson et al., 2002), and reduced extent of sea ice cover (Serreze et al., 2002; Rothrock et al., 2003). These sig­ nificant changes are occurring over large spatial domains. The combination