Hydrological Sciences Journal

ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: http://www.tandfonline.com/loi/thsj20

Extreme hydrological events, palaeo-information

and climate change

J. C. KNOX & Z. W. KUNDZEWICZ

To cite this article: J. C. KNOX & Z. W. KUNDZEWICZ (1997) Extreme hydrological events,
palaeo-information and climate change, Hydrological Sciences Journal, 42:5, 765-779, DOI:
10.1080/02626669709492071

To link to this article: http://dx.doi.org/10.1080/02626669709492071

Published online: 25 Dec 2009.

Submit your article to this journal

Article views: 400

View related articles

Citing articles: 18 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=thsj20

Download by: [190.91.77.208] Date: 04 September 2017, At: 11:00

 Hydrological Sciences-Journal-des Sciences Hydrologiques, 42(5) October 1997 765

Extreme hydrological events, palaeo-information

and climate change*

J. C. KNOX
Department of Geography, University of Wisconsin, Madison, Wisconsin 53706-1491, USA

Z. W. KUNDZEWICZ
Research Centre of Agriculture and Forest Environment Studies, Polish Academy of Sciences,
Poznan, Poland*

Abstract Extreme flood events have been and continue to be one of the most
important natural hazards responsible for deaths and economic losses. Extreme
Downloaded by [190.91.77.208] at 11:00 04 September 2017

floods result in direct destructive effects during the time of the event, and they also
may be followed by a related chain of indirect calamities such as famines and
epidemics that produce additional damages and suffering. Extreme hydrological
events that have occurred in the historical past may also occur in the future.
Knowledge about magnitudes and recurrence frequencies of past extreme
hydrological events in most regions are too short to adequately evaluate potential
magnitudes and recurrence frequencies of extreme hydrological events. Stationary
climate in which the mean and variance do not change over time is a basic
underlying assumption of standard methodological procedures for estimating
recurrence probabilities of extreme hydrological events. Palaeo-archives contained in
river and lake sediments, fossil plant and animal matter, ice layers, and otJier natural
archives show that the assumption of stationary climate is not valid when the time
scale is extended beyond centuries and millennia. Records of past extreme floods
that occurred long before the period of instrumentation can be reconstructed from
the distribution of slackwater flood deposits or from derivation of water depths
competent to transport the largest rocks found in flood deposited sediment. Palaeo-
flood records reconstructed from the Upper Mississippi and Lower Colorado River
systems in the United States confirm nonstationary behaviour of the mean and
variance in hydrological time series. These stratigraphie records have shown that
even very modest climatic changes have resulted in very important changes in the
magnitudes and recurrence frequencies of extreme floods. A close relationship was
found between the palaeo-flood record of extreme floods in the Upper Mississippi
River system and a palaeo-record of stable isotopes of oxygen and carbon preserved
in speleothem calcite from a local cave. The relationship suggests that isotopic
records elsewhere might be calibrated to provide insight about how future potential
climate changes might impact extreme flood magnitudes and recurrence frequencies
there. Atmospheric global circulation models (GCMs) mainly simulate average
climatic conditions and are presently inadequate sources of information about how
future climate changes might be represented at the extreme event scale. Palaeo-flood
archives, however, provide basic information about how magnitudes and recurrence
frequencies of extreme hydrological events responded to past climate changes and
they also provide a reference base against which GCM simulations can be calibrated
regionally and be better interpreted to decipher hydrological information at the
extreme event scale.

f
Paper presented at the International Workshop on Tracing Isotopic Composition of Past and Present
Precipitation—Opportunities for Climate and Water Studies (Berne, Switzerland, 23-25 January 1995).

* The paper was submitted when the author was with: World Meteorological Organization, Geneva,
Switzerland.

Open for discussion until 1 April 1998

 766 J. C. Knox & Z. W. Kundzewicz

Evénements hydrologiques extrêmes, paléoinformation et

modifications climatiques
Résumé Les débits extrêmes ont été et restent encore l'un des plus importants risques
naturels responsable de la perte de vies humaines et de biens matériels. Les
écoulements extrêmes causent des destructions immédiates et des pertes indirectes,
comme des famines et des épidémies. Dans plusieurs régions, l'évaluation du risque
potentiel n'est pas possible, parce que les séries chronologiques de débits enregistrées
sont trop courtes. La paléoinformation enregistrée dans les sédiments des rivières et des
lacs, les plantes et animaux fossilisés, les glaces et autres archives naturelles prouve
que l'hypothèse de stationnante du climat n'est pas recevable à l'échelle des centaines
ou de milliers d'années. On peut reconstituer les anciens débits extrêmes grâce à leur
signature dans les sédiments déposés par les écoulements, en particulier en estimant les
débits susceptibles de transporter les plus grosses pierres retrouvées dans les sédiments.
La reconstitution de l'information sur les crues anciennes de fleuves tels que le
Mississippi supérieur et le bas Colorado, confirme la nature instationaire des séries
chronologiques hydrologiques. On a pu observer, en étudiant les enregistrements strati-
graphiques, que même de petites variations climatiques avaient entraîné des modifica-
tions très importantes de l'intensité et de la fréquence des débits extrêmes. On a trouvé
Downloaded by [190.91.77.208] at 11:00 04 September 2017

une relation significative entre les anciens écoulements du Mississipi supérieur et la

composition isotopique de dépôts de calcite dans des grottes de la région. Il est donc
possible d'interpréter des signaux isotopiques anciens afin de prévoir de quelle façon
les futures modifications climatiques pourront influencer les intensités et les fréquences
des débits de crue. Les modèles de circulation atmosphérique décrivent des conditions
climatiques moyennes et ne sont donc pas des sources d'information suffisantes sur les
relations entre les futures modifications climatiques et les débits extrêmes. Les archives
de débits anciens contiennent l'information de base concernant l'impact des
modifications climatiques sur les intensités et les fréquences des écoulements extrêmes.
Ils constituent ainsi une base de référence pour le calage régional des modèles de
circulation générale et leur interprétation en termes d'information sur les événements
hydrologiques extrêmes.

INTRODUCTION

Extreme hydrological events represent a focus of considerable interest in hydrology

because such events have the potential to cause considerable harm to human
populations and their related activities. From the beginning of their existence,
humans have suffered from having too little or too much water at particular places
and times. In the present century an additional problem involving dirty water has
emerged. Having water directly unusable because of pollution effectively reduces
already scarce water resources in many areas.
As noted in Kundzewicz et al. (1993), there is no objective and precise definition
of floods and droughts commonly accepted. It is difficult to provide a precise
definition for what can be regarded as a "flood event" or as a "non-flood event".
One obvious basis for definition is to assign arbitrary thresholds of flow amplitude
that identify floods and low flows in such a way that crossing these thresholds (up-
crossing of a flood threshold and down-crossing of a low-flow threshold) is
understood as the onset of an extreme hydrological event. The thresholds could be
chosen either on the basis of hydrological parameters related to flow frequency or in
relationship to such levels that crossing them leads to significant economic losses.
Often the notion of flood is used in combination with a recurrence interval such as,
for instance, a 100-year flood, i.e. a flow magnitude being exceeded, on the average,
in one year in a century. These and other levels associated with different recurrence
 Extreme hydrological events, palaeo-information and climate change 767

intervals lend themselves well as reference thresholds for flood studies.

Extreme floods and droughts have caused and continue to cause high tolls in
deaths and economic losses. Floods are more spectacular as they are short lasting and
violent. The highly visible direct destructive effect of floods may be followed by a
chain of secondary, indirect calamities including famine, epidemics, and fire that
cause concerned populations additional suffering. There have been floods during
historical times that have resulted in death tolls of several hundred thousand human
beings (for example, 900 000 and about 500 000 deaths in the catastrophic floods in
China in 1887 and 1939, respectively). Present-day floods do not take such high tolls
in the number of lives because of modern systems of advanced warning, but present-
day floods continue to cause very high economic losses. Losses associated with the
record breaking Mississippi River flood in 1993 are estimated to be in exceedance of
US$10 billion (Tobin & Montz, 1994).
Downloaded by [190.91.77.208] at 11:00 04 September 2017

Some common flood components that are routinely considered when evaluating
environmental impacts from floods include: runoff volume, peak discharge magni-
tude, flood stage elevation, time distribution of runoff, area inundated, and flow
velocities (Dunne & Leopold, 1978). Collectively, these variables provide informa-
tion necessary for the management of flood control works, water supply, irrigation,
land use policy and zoning, as well as many other factors that relate to the storage
and routing of flood waters through channels, flood plains, and reservoirs of the
watershed system.
The maximum flood discharge and its exceedance probability are the commonly
used parameters for describing the process of threshold crossings. The maximum
peak discharge is a basic parameter for designing spillways, bridges, and flood relief
channels. However, other flood characteristics, as previously noted, are of direct
importance in other applications. For example, information about probable or poss-
ible maximum flood stage elevations and flood durations are of critical concern for
levee designs, and for specifying areas of probable inundation by flood waters during
time of maximum stage. In turn, predicted stage heights and extent of expected
inundation are important in planning for evacuation and loss assessment. Under-
standing the probable limits of flood magnitudes and recurrence frequencies also is
an important part of environmental planning. For example, knowledge of past flood
histories provides a critical information base for choosing between flood protection
alternatives including structural measures such as storage reservoirs, levees, channel
and flood plains modifications etc. vs non-structural measures such as improved
flood forecasting, insurance, proofing, etc.

WHY DO WE NEED TO KNOW PAST EXTREME EVENTS?

Many human catastrophes are associated with extreme natural events. What has
happened in the past can, and probably will, happen again in the future. Therefore,
knowledge about the characteristics of extreme hydrological events is valuable for
siting, planning, and designing numerous objects within the human infrastructure.
Furthermore, because the memories of experiences of the living human population
 768 J. C. Knox & Z. W. Kundzewicz

are short term and because historical instrumental records at most locations are also
short term, there is a tendency to perceive incorrectly even moderate magnitude and
relatively high recurrence frequency natural events as more rare and extreme than
they are. For example, the great flood of 1993 on the Mississippi River at St. Louis,
Missouri, USA was generally perceived by local inhabitants and the media as an
exceptionally rare event, but historical data show at least six floods of nearly the
same size occurred there since the 1780s with the most recent large flood having
occurred only twenty years previously (Bhowmik et al., 1994; Chin et al., 1975). A
careful study of the characteristics of past events can significantly improve the degree
of preparedness for hydrological extreme events such as floods and droughts.
Historical instrumental records of extreme hydrological events for most regions
are typically shorter than desired for a thorough assessment of potential magnitudes
and recurrence frequencies. Short duration records not only weaken the theoretical
Downloaded by [190.91.77.208] at 11:00 04 September 2017

background of design, planning and management, but they also limit the ability to
test scientific theories that describe connecting linkages between climatic change and
hydrological systems.
Standard practice in planning and design for flood and drought protection, and in
siting and dimensioning water projects, typically involves the assumption of a stable
and stationary climate. The stationarity assumption requires that the mean and
variance of climatic conditions do not change over time. If a shift in either or both
the mean and variance should occur in the future, then the recurrence probabilities
for hydrological events of a specified magnitude will no longer be correctly speci-
fied. A change in the variance (heteroscedascity) can be quite important even if the
mean state does not change. This can be illustrated by the direct influence of non-
stationary variance of runoff on reservoir design. However, even if the assumption of
stationary conditions holds, the length of available instrumental data may not be
adequate to represent correctly the mean and variance over the long term. Too often
the background information for a costly investment stems from a meagre few years,
or a few decades at best, of relevant hydrological observations.
Predicting the magnitude of a 100-year flood (e.g. for dam design) based on five
years of data is definitely improper. Yet, in modern highly risk-averse societies,
design recurrence intervals of 30 000 years or more are at stake when planning very
large dams or cooling systems of nuclear power plants. As reported by Klemes
(1986), even a request for the determination of a 1 000 000-year flood has been
made. Estimation of such extreme probability events involves enormous extrapola-
tions based on statistical variability observed in existing instrumental records.
Because event magnitudes predicted in the tail regions of probability density distri-
butions (pdfs) may vary significantly from one distribution to another, the choice of a
pdf (from a set of distributions, none of which has a physical basis) has major
consequences. The characteristics of extreme hydrological events used in design and
planning have therefore largely a public-relations flavour rather than being rigorous
and theoretically sound accurate values.
However, in the light of geological evidence indicating that major climatic
changes have occurred over time scales involving thousands of years and longer, it is
evident that the stationarity assumption and the procedures of long-term extrapolation
 Extreme hydrological events, palaeo-information and climate change 769

based only on a relatively short instrumental record are likely to be erroneous. The
idea that stationarity is unlikely to be maintained over long periods is further
substantiated by the records of climatic disasters and anomalies in ancient Chinese
documents. Fang (1992) found that frequencies of floods, droughts, cold years, and
warm years commonly displayed abrupt temporal changes in recurrence frequency.
He attributed much of the abrupt changes to climatic causes, but he noted that some
of the change toward increasing floods during the past 2000 years was related to
vegetation destruction and soil erosion associated with agricultural development.
Starkel (1987) concluded that human activity accelerated runoff and sediment yields
as early as about 10 500 years ago in steppe regions while many forested regions
were similarly affected about 6800 years ago. Although extreme floods tend to be
less sensitive to land use than moderate to small magnitude floods, it is clear that
human impacts on hydrological systems can also be a cause of nonstationarity of the
Downloaded by [190.91.77.208] at 11:00 04 September 2017

mean and variance of flood records.

The palaeo-archives of the Earth, contained in stratigraphie records of sediments,
fossil plant and animal matter, ice layers, and other materials, indicate that climate
and water resources have never been stationary over a long time interval (geological
scales), even without anthropogenic impacts (Wohl & Enzel, 1995). As recently
demonstrated by Greenland Ice-core Project Members (1993), several very rapid
shifts of climate with mean temperature changes of up to 10°C in as little as a couple
of decades took place in the previous interglacial period (known as Eemian), which
was previously believed to have had a stable climate. The high (decadal scale)
resolution of past climates from the central Greenland Summit ice core, was made
possible by analyses of stable isotope ratios, chemical and physical properties, and
greenhouse gas concentrations in trapped air bubbles contained in annual ice layers.
Extension of the instrumental data base is much needed so proxy data and palaeo-
information are of increasing interest in hydrology and water resources planning and
management to augment existing hydrological records. Evidence of historical flood
peaks which reflect the frequency of large and extreme floods is therefore being
sought. Wohl & Enzel (1995) summarize methods for reconstructing river flows into
five divisions that include: (1) nonsystematic historical records typically represented
by high water marks; (2) regime-based methods using properties of channels or
drainage networks; (3) palaeo-competence methods which relate characteristics of
channel sediments to flow parameters; (4) geobotanic methods which employ
characteristics of vegetation along a channel or flood plain; and (5) palaeo-stage
indicators such as slackwater deposits or silt lines. Reconstructing long records of
extreme floods has been mainly achieved by the application of palaeo-competence
and palaeo-stage methodologies.

EXPLORING PALAEO-FLOOD ARCHIVES OF EXTREME EVENTS

Two late-Holocene records of large flood events provide example applications of the
palaeo-competence and palaeo-stage methods that illustrate how extreme flood
occurrences can be episodic and highly sensitive to even modest climatic changes
 770 ./. C. Knox & Z. W. Kundzewicz

(Knox, 1993; Ely et al., 1993). The illustration of the palaeo-competence method
uses tributaries of the upper Mississippi River in the north-central United States, and
their records have been described in greater detail by Knox (1993). The 7000 years
flood records for the upper Mississippi River tributaries were compiled by computing
the minimum depths of flood water necessary to have transported the coarsest
sediment particles found in cobble and boulder gravels deposited by former large
overbank floods and now buried in overbank floodplain alluvium. The depths of
water competent to transport the relict flood deposits were estimated using an
empirically derived hydraulic equation that, when calibrated on modern data,
explained 74% of the variance and was statistically significant at the 0.0001
probability level. The equation expresses competent flood depth as a function of the
mean intermediate axis of the coarsest five particles in a relict flood gravel and the
palaeo-flood energy gradient approximated by the slope of the floodplain surface
Downloaded by [190.91.77.208] at 11:00 04 September 2017

containing the relict gravels (Knox, 1993). Ages for relict floods were determined
either by radiocarbon dating of the organic materials found in the flood gravels or by
dating the deposits that stratigraphically bracket the flood gravels.
The reconstructed record of approximately 70 large overbank floods that
occurred on upper Mississippi River tributaries, mainly in southwestern Wisconsin,
during the past 7000 years shows very clearly that both the mean and variance of the
data have been subject to abrupt changes over time (Fig. 1). Note that the depths of
relict floods shown on Fig. 1 are expressed as a ratio to their respective bankfull
stage depth at the site to facilitate plotting all floods on a single graph. The bankfull
stage is defined as the flood depth that corresponds approximately with a flood event
having a recurrence probability of one to two years. The bankfull stage also tends to
correspond with the top of lateral accretion sedimentation as illustrated by point bar
deposits. Since ratios of flood depths to bankfull stage depths for rivers with alluvial
floodplains are known to vary systematically with flood recurrence probabilities

i 1 1 ' 1 ' 1 ' I

4.5 AV6r3C)6 95 percentile CI for Age

Average 95 percentile C! for Flood Ratio -
: -
(p)

sz 2.
1 4.0
A
A A

3.5 A
1
Q
£
•=>• 500-Yr Flood »
A

g" 3.0 A * A
?o D
2.5 A
\
' AA
A A »
A

* * ~-
nkfu
of Fl

A
^ 50-Yr " A A A* A A _
2.0 ' Flood A
A

10 A
ê œ 1.5
A
A
A A
A * / * :
o A
*
& 1.0
A
A A A

0.5 1
5500 3300
4 1200 t + : 500
0.0 i ! i 1 i 1 i

7000 6000 5000 4000 3000 2000 1000

Calendar Years
Fig. 1 Stratigraphie record of 70 large overbank floods m the upper Mississippi
River drainage basin of southwestern Wisconsin and northwestern Illinois as
reconstructed by Knox (1993) based on water depths required for competence to
transport the largest boulders and cobbles found in the relict flood deposits.
 Extreme hydrological events, palaeo-information and climate change 77 ]

(Dunne & Leopold, 1978), the Fig. 1 arrangement allows for the interpretation of
computed relict flood depths in relation to recurrence intervals of modern floods of
similar depths. For example, in the upper Mississippi River system represented in
Fig. 1, a flood depth ratio of 2 implies a water stage above the channel bed that is
twice the modern bankfull channel depth. The regional overbank floods with flood
depth ratios of about 2 are expected as frequently as once in 30-50 years, whereas
floods with depth ratios of about 3 are expected to recur only about once in 500 years
or longer under modern conditions (Fig. 1).
The most striking example of nonstationarity in the upper Mississippi River
system record occurs about 3300 years ago when an abrupt shift from a 2000 year
episode characterized by small floods is suddenly replaced a new hydrological regime
of large floods. The small floods of the period between about 5500 and 3300 years
ago generally did not exceed magnitudes of modern floods expected to recur as
Downloaded by [190.91.77.208] at 11:00 04 September 2017

frequently as once in 50 years (Fig. 1). Climate change is presumed to be the cause
of the change in flooding behaviour because there is no evidence of human influence
or from other natural environmental causes. Climate change also is suggested as the
principal cause because analyses by others show that the period of small floods here
corresponded with mean annual precipitation about 15% less than modern rates and
July temperature about 0.5°C warmer than before or after (Winkler et al., 1986;
Bartlein et al., 1984). These climatic changes were apparently associated with
changes in hemispheric patterns of atmospheric circulation responsible for the low
incidence of large summer floods. Fossil proxy climate data indicate that between
about 4000 and 3000 years ago a shift to cooler wetter conditions occurred with
annual precipitation gradually increasing by about 16% to its modern value but with
mean annual temperature first decreasing by about 0.7°C and then increasing grad-
ually by about 0.4°C. Further evidence that climatic change is the driving force
behind the abrupt changes in flood magnitudes is indicated by palaeo-records of the
oxygen and carbon isotopic composition of speleofhem calcite from a cave in the
local region (Dorale et al., 1992; Baker et al, 1996). The isotopic record indicates
that major changes occurred about 5900 and 3300 years ago. Dorale et al. (1992) and
revisions by Baker et al. (1996) using oxygen isotopic ratios suggest that the climate
warmed rapidly by about 1.5°C about 5900 years ago and then cooled abruptly about
3300 years ago when savanna vegetation replaced prairie. Allowing for differences in
dating methodology and control, the major discontinuities in the isotopic record
agree rather closely with the discontinuities at 5500 and 3300 years ago in the record
of large overbank floods (Fig. 1). It is important to recognize that changes in mean
temperature and seasonal precipitation were not the direct causes of the changes in
recurrence frequencies of the extreme floods. However, the changes in mean
conditions are symbolic of changes in air mass boundaries and trajectories of storm
tracks that favour either the enhancement or suppression of extreme hydrological
events. In this way persistent departures from long-term average climatic conditions
can be expected to be reflected in the record of extreme hydrological events. Note on
Fig. 1, for example, that between about 700 and 500 years ago, during a period of
cooling climate leading into the Little Ice Age, the frequency and magnitude of large
overbank floods on the upper Mississippi River tributaries was anomalously high.
 772 J. C. Knox & Z, W. Kundzewicz

Use of slackwater deposits as palaeo-stage indicators of past floods is the other

most commonly practiced method of compiling a long-term history of extreme
floods. Slackwater sediments typically are relatively fine-grained deposits that
accumulate as sediment falls from suspension in areas of slowed flow velocities.
An extensive discussion of the methodology and its limitations has been presented
by Kochel & Baker (1988). Common sites for slackwater deposition include:
tributary mouths, depressions and cavities in bedrock canyon walls, downstream
margins of valley obstructions and other features that associate with flow
separation in velocity streamlines. Palaeo-stage heights of floods are indicated by
tracing slackwater sediments to their highest vertical extent. Kochel & Baker
(1988) found that slackwater flood sediments commonly occur in stacked
individually fining-upward sequences that may display abrupt grain size and
mineralogical changes between units.
Downloaded by [190.91.77.208] at 11:00 04 September 2017

Ely et al. (1993) have used slackwater sediments to reconstruct a 5000-year record of
floods in the Colorado River system of southwestern United States (Fig. 2). They found,
as in the long-term record of the upper Mississippi River system, that nonstationarity of
the mean and variance characterized the record of extreme flood occurrences. Their
results show that large floods cluster into distinct times that coincide with the prevalence
of cool, moist climate and frequent El Nino events, and that large floods were
concentrated between about 5700-3800 years ago and since about 2100 years ago. Large
floods were extremely rare or absent for the period between about 3800 and 2100 years
ago (Fig. 2). The apparent dramatic increase in frequency of large floods in the last 500
years may partially reflect a tendency for the stratigraphie record to be more available
and better preserved with decreasing age, although Ely et al. (1993) concluded that the
pattern shown on Fig. 2 was not merely an artifact of differential preservation.
The long-term post-glacial records of large floods in the upper Mississippi River and
Colorado River drainage basins show that relatively modest shifts in climate regimes can

I 132

30

Number
4900
20
Of 5650 I 3750

10
Floods

5600 4800 4000 3200 2400 1600 800 0

Radiocarbon Years
Fig. 2 Stratigraphie record of large floods on rivers in the Colorado River drainage
basin of Arizona and southern Utah as reconstructed by Ely et al. (1993) from
slackwater sediments. (Dates above arrows are calendar year ages BP based on
CALIB calibration of Stuiver & Reimer, 1993).
 Extreme hydrological events, palaeo-information and climate change 773

result in large differences in magnitudes and recurrence frequencies of extreme floods.

These records also emphasize the importance of long-term perspectives because short
records, such as represented by the period of instrumentation, often fail to capture the
range of possible hydrological extremes. Long-term geological records of extreme
hydrological events also provide a useful baseline for judging regional responses to
global-scale climatic changes. Regional responses to global warming or global cooling
can be very different because causal factors responsible for extreme floods vary
regionally. An illustration of this principle is evident when comparing the flood
chronologies shown on Figs 1 and 2. Note that the period of frequent large floods
between about 5700 and 3800 years ago in the southwestern United States corresponds in
time approximately with the north-central United States flood episode between about
5500 and 3300 years ago when few floods exceeded the magnitude of a relatively modest
flood of 50-year recurrence probability. The relationship suggests that hemispheric
Downloaded by [190.91.77.208] at 11:00 04 September 2017

circulation regimes that were responsible for the enhancement of extreme flood
occurrences in the southwest United States also were responsible for suppression of
extreme flood occurrences in the northern midwest United States.
As previously noted, the examples of post-glacial variations in floods from middle
latitude watersheds in the United States show that very significant changes in magnitudes
and recurrence frequencies of extreme events can be expected even from relatively
modest climatic changes. The long history of Nile River extreme floods indicates that the
tropical and subtropical latitudes behave in similar fashion. For example, by about 10 000
years ago, when most large-scale vegetation communities were beginning to approach
their modern locations, the seasonal cycle of radiation in the Northern Hemisphere was
reaching a relative post-glacial maximum of about 8% more than today in response to the
Milankovitch cycle of precession of the Earth's axis (Kutzbach, 1981). The increased
warming over the Northern Hemisphere continents apparently accelerated summer
monsoonal circulation in North Africa and increased rainfall on the Equatorial Lake
Plateau in East Africa. The resulting early post-glacial dominance of the Nile River by
high stage flows and frequent extreme floods cleared the White Nile of dunes which had
choked its course during the aridity of the last glacial maximum. These flows apparently
degraded the Nubian Nile by about 30 m (Said, 1993). Said (1993) suggests that the post-
glacial history of Nile flows can be divided into a period of high flows that dominated
before 2400 BC (about 4400 years ago) and a period of low flows that have dominated
since then, although each of these episodes was characterized by shorter term flood
episodes associated with changes in climatic wetness and dryness. Noteworthy of the
shorter term variation in the Nile records is a 110 year period in the apparent globally
cool 14th and 15th centuries AD when exceptionally high magnitude floods occurred,
while the apparent globally warm 10th, 11th, and 20th centuries are characterized by
exceptionally small floods (Said, 1993).

EXPLORING PALAEO-ISOTOPES AS A SOURCE OF PALAEO-

HYDROLOGICAL INFORMATION

A basic insight into reconstructions of the past climate is gathered by the study of
stable isotope composition (e.g. oxygen-18, hydrogen-2) of past precipitation.
 774 J. C. Knox & Z. W. Kundzewicz

Analysis of deep ice cores or laminar lake deposits proved the an ability to re-
construct past climates, the behaviour of the water cycle and its links with climate
change, with improving resolution and reliability. Baseline data collected since 1961
by the IAEA/WMO Global Network "Isotopes in Precipitation" (GNIP), has made it
possible to determine significant relationships between the isotope compositions of
precipitated water and surface air temperature, as well as relationships between the
amount of rainfall and prevailing atmospheric circulation patterns. However, these
relationships have been mostly developed to explain associations between hydro-
logical and climatological variables measured at the mean annual scale rather than
developing an understanding of individual occurrences of extreme events.
It remains an open question as to how these methods are applicable to studies of
hydrological extremes. If, in the mean sense, a wetter and colder climatic period can
be detected some time in the past, it may be possible to reconstruct flood charac-
Downloaded by [190.91.77.208] at 11:00 04 September 2017

teristics indirectly, assuming, for instance, an annual distribution of precipitation

equal to the present one. Unfortunately there is little long-term data where the history
of large floods can be matched against temporally equivalent isotopic records to
determine properly if the isotopic record can be calibrated to provide a useful proxy
measure of hydrological extreme events. However, the analysis by Dorale et al.
(1992) and Baker et al. (1996) of the oxygen and carbon isotopic composition of
speleothem calcite in a northeast Iowa cave showed that local climatic history of the
past 8000 years can be divided into three distinct stages that are separated by abrupt
change, and that these stages tend to correspond closely with the 7000 year history of
large floods in the region as noted previously above. For example, Dorale et al.
(1992) and Baker et al. (1996) note that between about 7800 and 5700 years ago 8180
values average about 24.76%c, but at the beginning of stage 2 about 5700 years ago,
8180 values rose abruptly to an average value of 25.81 %>. They found that values of
8180 remained close to the stage 2 average value until about 3300 years ago (the
beginning of stage 3) when ô180 dropped abruptly to an average of 24.14%o. Dorale
et al. (1992) and revisions by Baker et al. (1996) indicated that the middle post-
glacial isotopic stage 2 excursion represented a mean annual temperature warming of
about 1.5°C from stage 1 conditions, and that the isotopic stage 3 recovery
represented an abrupt drop in mean annual temperature from stage 2. Dorale et al.
(1992) and Baker et al. (1996) attributed the large temperature changes to post-
glacial adjustments in air mass boundaries because the speleothem sampling site is
located on a prairie-forest ecotone known also to be associated with seasonal air mass
boundaries. It is noteworthy that the isotopic stage 2 excursion indicative of warming
matches rather closely with the fluvial episode when low frequency extreme floods
were of very small magnitude in the upper Mississippi Valley (Fig. 1). The sample
record of former large overbank floods for the period corresponding with isotopic
stage 2 shows that maximum flood magnitudes were about the size of the modern
equivalent 50-year flood or smaller. The cooler climate episode of isotope stage 3
beginning about 3300 years ago, on the other hand, corresponds closely with a
period in which extreme floods of modern 500-year return frequency are common
(Fig. 1). These associations suggest that the proxy record ®f temperature represented
in the palaeo-isotopic record might also be calibrated to provide a proxy record of
 Extreme hydrological events, palaeo-information and climate change 775

large flood events because the isotopic record is sensitive to air mass characteristics.
Air mass dominance reflects not only potential precipitable water vapour content, but
it also is indicative of large scale trajectories of atmospheric circulation and storm
track positions. These climatic variables have strong influences on flood magnitudes
and their recurrence frequencies.
Isotopic indices of annual precipitation and mean temperature of remote past
periods also can help decipher the properties of droughts. If the yearly rainfall total
in a period of interest was lower than at present and the temperature higher, droughts
were more likely to occur and some quantitative assessments can be made.

CLIMATE CHANGE AND EXTREME EVENTS: LOOK INTO THE FUTURE

Downloaded by [190.91.77.208] at 11:00 04 September 2017

The recent interest in climate change builds upon the observation that human
interventions in the natural fluxes of energy, in the last century, attained dimensions
which significantly altered the natural regime. The ratios of components of energy
fluxes have changed at the global scale. Increased combustion of fossil fuels,
deforestation and production of some synthetic compounds have caused a rise of
concentration of "greenhouse" gases in the atmosphere resulting in a warm-up effect.
As the climate and water systems are interwoven, it may be expected that the
warming effect may influence the dynamics of the hydrological cycle (Rind, 1988;
Gleick, 1989)
Atmospheric global circulation models provide a general understanding of
probable directions of climatic and hydrological change, but they emphasize average
conditions. These models to date generally have been inadequate sources of informa-
tion about how future climate changes might be represented at the "event" scale. In
particular they are especially poor at representing expectations of extreme events.
Recent results from climate simulations of global warming in association with
increased atmospheric loading with greenhouse gases, while indicating that tempera-
ture variability can be expected to decrease, also indicate that the proportion of
precipitation occurring in extreme events can be expected to increase (Karl et al.,
1995). The occurrence or non-occurrence of natural events in the statistical "tails" of
frequency distributions is of critical importance because economic impact and other
potential negative impacts on human life tend to increase exponentially as the magni-
tude of a natural event deviates from the average.
Disaggregating results of general atmospheric circulation models into smaller
hydrologically-relevant areas may produce results largely dependent on the choice of
a particular GCM. McCabe & Wollock (1992) analysed the moisture conditions in
the Delaware River basin, based on different GCM simulations of possible climatic
change scenarios. One GCM predicted a drop in precipitation by 46 mm, whereas
another GCM predicted a rise in precipitation by 143.9 mm. This sort of uncertainty
may even become amplified when translated into runoff analyses. In another study of
GCM model predictions for 18 major United States watersheds, Knox (1990) found
that results from three commonly used GCM models indicated both agreement and
disagreement at the large watershed scale. On the assumption of a doubling of CO,,
 776 J. C. Knox & Z. W. Kundzewicz

however, there was a general tendency for most middle latitude US watersheds to
experience warming and a relatively modest average increase in annual runoff of
about 21 mm per year, but with a sample standard deviation of about 60 mm. The
large standard deviation reflects the substantial difference in monthly watershed
precipitation predicted by the different models. Not surprisingly, with predicted
higher mean temperatures and only modest increases in predicted precipitation, a net
decrease of about 0.4 mm per year runoff was predicted for the average of the three
models. The sample standard deviation for predicted annual runoff was 32 mm, again
reflective of the large differences in predicted monthly runoff between the models.
While ongoing research by climate modellers suggests that the magnitudes of
warming suggested by first generation GCM models probably need to be scaled back
somewhat, the models nevertheless clearly illustrate that the average hydrological
cycle is highly sensitive to even a modest climate change. The sensitivity of
Downloaded by [190.91.77.208] at 11:00 04 September 2017

hydrological extreme events such as floods and droughts, however, is not presently
captured in GCM models. The palaeo-record of extreme hydrological events
therefore is not only important as a source of basic information regarding how
extreme natural events responded to past climate changes, but also is critical infor-
mation for the calibration of GCM and other regional climate models as their
technology and capabilities improve in the future.
What could be the climate change consequences on extreme hydrological events?
Although much scientific effort has been dedicated to studying this issue, there are
still strong uncertainties in the results. Scientists keep looking for greenhouse signa-
tures in long series of observed hydrological records. However, no general and
unambiguous message relevant to flow data can be asserted yet. Figure 3 presents a
long time series of flows of the River Warta, in Poznan, Poland (annual mimima,
means and maxima). It is clear that strong natural variability dominates in this time
series and that no clear and unambigous greenhouse component may be dis-
tinguished. Karl et al. (1995) analyses of amounts of annual and seasonal precipi-

— 1000 1

"B 1000 , •• • ,

min mean max

Fig. 3 Long time series of minimum, mean and maximum annual flows for the
Poznan gauge at the River Warta (instrumental record).
 Extreme hydrological events, palaeo-information and climate change 777

tation delivered in extreme events during recent years of global warming show a
significant increase in the proportion of precipitation derived from extreme precipi-
tation over the USA, but they did not find such trends over the former Soviet Union
and China. No persuading and scientifically meaningful evidence of a changed
regime of floods and low flows and droughts has yet been found. However, in past
assessments the following conclusions are frequently offered:
(1) In middle latitudes of the Northern Hemisphere the winter precipitation is likely
to increase and may fall more as rain rather than snow. Thus a risk of increased
winter floods arises. Lower water storage in the snow pack and earlier snowmelt
may cause reduced spring streamflow.
(2) Prophecies of an increase in the intensity and severity of a pronounced drought
summer period are also quite frequent. Higher temperatures will cause higher
évapotranspiration which, in turn, would outweigh the increase in precipitation
Downloaded by [190.91.77.208] at 11:00 04 September 2017

(3) There is a common belief that climatic changes may have a stronger effect on
extremes than on mean values of hydrological variables. Extreme events ob-
served recently are increasingly taken as manifestations of such an effect.

CONCLUSION

In summary, available historical and proxy data on extreme floods imply quite
clearly that the recurrence frequencies and geographical distributions of these events
are highly sensitive to changes in large-scale atmospheric and oceanic circulation
regimes. Currently available historical and proxy data on extreme floods are too
limited, both with regard to length of records and to geographical locations, to test
adequately important hypothesized climate-flood relationships.
A few immediate questions come about. Is it possible to recognize globally
synchronous episodes of extreme hydrological events for historical and post-glacial
time scales? Are times of major climatic change more prone to experience hydro-
logical extremes than periods before and after the change? Does the relative direction
of a climatic change play a significant role in occurrences of hydrological extremes?
Occurrences of extreme floods in the middle latitudes often follow large volcanic
eruptions in tropical latitudes as occurred in the early 1880s and in the early 1990s.
What is the role of large volcanic eruptions in influencing hydrological extremes?
Because different flood generating processes operate in different regions, as for
example, snow melt, localized convectional rainfall, frontal rainfall, extreme mon-
soonal rainfall, etc., local and regional responses to global climatic change obviously
vary. While long historical flood records from the Nile and certain Chinese rivers are
extremely helpful, the real need is to expand the proxy flood record to
environmentally sensitive regions. Because flood generation processes are under-
stood, and because of the large regional scale nature of atmospheric and oceanic
circulation regimes, long-term proxy flood records from strategic regions would
provide important advances in understanding how the magnitudes and frequencies of
specific types of circulation regimes responsible for extreme hydrological events
varied during past global environmental changes. In turn, these understandings are
 778 J. C. Knox & Z. W. Kundzewicz

useful for prediction of possible hydrological consequences related to potential future

environmental changes.
It is important to determine whether the isotopic methods of palaeo-
environmental reconstructions that have worked so well in marine and ice-core
records can also be applied with similar success to define better past occurrences of
extreme hydrological events in continental fluvial deposits. The parallel trends in
palaeo-isotopic ratios and palaeo-flood indices described above for the midwestern
USA suggest that isotopic data might be calibrated to provide useful proxies of high
amplitude short term hydrological variability. More study is needed to define the
limitations and constraints in the applications of isotopic data to studying extreme
hydrological events. These limitations should be considered in the context of
environment type (humid, arid, polar, tropical) and in the context of time scale
(event basis, seasonality effects, and accumulated departures from the mean).
Downloaded by [190.91.77.208] at 11:00 04 September 2017

Hydrology and water resources specialists also need better definition of how well
extreme hydrological events may be deciphered from various forms of palaeo-data.
There is a strong need to better articulate how palaeo-data in the fossil record can be
applied for reconstructing flood parameters such as: peak stage, runoff volume, event
duration, and flood recurrence frequencies. Similar needs occur for the application of
palaeo-data in the fossil record for the study of droughts. Particular attention should
be given to defining the confidence limits of such methodology, including evaluations
of the credibility and accuracy of findings. Attention also should be given to define
better the limits of temporal discrimination and spatial transferability associated with
various types of hydrological palaeo-data. Finally, there is a strong need to integrate
the instrumental hydrological records with proxy palaeo-data. These integrations
could greatly improve the quantification of proxy hydrological palaeo-data. In turn,
better quantitative evaluations of extreme hydrological responses to past
environmental changes will lead to improved methodology for predicting extreme
hydrological responses to potential future environmental changes.

Acknowledgements The research on the long-term palaeo-flood chronologies in the

USA was supported by the US National Science Foundation.

REFERENCES
Baker, R. G., Bettis, E. A. Ill, Schwert, D. P., Horton, D. G., Chumbley, C. A., Gonzalez, L. A. & Reagan, M. G.
(1996) Holocene paleoenvironments of the northeast Iowa. Ecol. Monographs 66, 203-234.
Bartlein, P. J., Webb, T., Ill & Fieri, E. (1984) Holocene climatic change in the northern Midwest: pollen-derived
estimates. Quatern. Res. 22, 361-374.
Bhowmik, N. G. et al. (1994) The 1993 Flood on the Mississippi River in Illinois. Illinois State Water Survey
Miscellaneous Publication 151.
Chin, E. H., Kelton, J. & Guy, H. P. (1975) The 1973 Mississippi River Basin Flood: Compilation and Analyses of
Météorologie, Streamflow, and Sediment Data. US Geological Survey Professional Paper 937.
Dorale, J. A., Gonzalez, M. K. Reagan, Pickett, D. A., Murrell, M. T. & Baker, R. G. (1992) A high-resolution
record of Holocene climate change in speleothem calcite from Cold Water Cave, northeast Iowa. Science 258,
1626-1630.
Dunne, T. & Leopold, L. B. (1978) Water in Environmental Planning. W. H. Freeman, San Francisco.
Ely, L. L., Enzel, Y., Baker, V. R. & Cayan, D. R. (1993) A 5000-year record of extreme floods and climate change
 Extreme hydrological events, palaeo-information and climate change 779

in the southwestern United States. Science 262, 410-412.

Fang, J.-Q. (1992) Establishment of a data bank from records of climatic disasters and anomalies in ancient Chinese
documents. International Journal of Climatology 12, 499-519.
Gleick, P. H. (1989) Climate change, hydrology, and water resources. Rev. Geophys. 27, 329-344.
(GRIP) Greenland Ice-cores Project Members (1993) Climate instability during the last interglacial period recorded in
the GRIP ice core. Nature 364, 203-207.
Karl, T. R., Knight, R. W., & Plummer, N. (1995) Trends in high-frequency climate variability in the twentieth
century. Nature ill, 217-220.
Klemes, V. (1986) Diletantism in hydrology: transition or destiny? Wat. Resour. Res. 22, 177S-188S.
Knox, J. C. (1990) Hydrologie and Geomorphic Implications of Climatic Change. Contract Report DACW 72-88-M-
0754, Institute for Water Resources, US Army Corps of Engineers, Fort Belvoir, Virginia.
Knox, J. C. (1993) Large increases in flood magnitude in response to modest changes in climate. Nature 361, 430-432.
Kochel, R. C. & Baker, V. R. (1988) Paleoflood analysis using slackwater deposits. In: Flood Geomorphology ed. by V.
R. Baker, R. C. Kochel & P. C. Patton, 357-376. John Wiley & Sons, New York.
Kundzewicz, Z. W., Rosbjerg, D., Simonovic, S. P. & Takeuchi, K. (1993) Extreme hydrological events in
perspective. In: Extreme Hydrological Events: Precipitation, Floods and Droughts, ed. by Z. W. Kundzewicz,
S. P. Simonovic & K. Takeuchi, 1-7. IAHS Publ. no. 213.
Downloaded by [190.91.77.208] at 11:00 04 September 2017

Kundzewicz, Z. W. & Somlyôdy, L. (1993) Climatic change impact on water resources—a systems view. IIASA
WP-93-30, IIASA, Laxenburg, Austria.
Kutzbach, J. E. (1981) Monsoon climate of the early Holocene climate experiment with the earth's orbital parameters
for 9000 years ago. Science 214, 59-61.
McCabe, G. J. & Wollock, D. M. (1992) Effects of climate change and climatic variability on the Thornthwaite
moisture index in the Delaware river basin. Climatic Change 20, 143-153.
Rind, D. (1988) The doubled CO, climate and the sensitivity of the modeled hydrologie cycle. /. Geophys. Res. 93,
5385-5412.
Said, R. (1993) The River Nile: Geology, Hydrology, and Utilization. Pergamon Press, Oxford, UK.
Starkel, L. (1987) Man as a cause of sedimentological changes in the Holocene: Anthropogenic sedimentological changes
during the Holocene. Striae 26, 5-12.
Stuiver, M. & Reimer, P. J. (1993) Extended l4C database and revised CALIB radiocarbon calibration program.
Radiocarbon 35, 215-230.
Thorp, M. & Thomas, M. (1992) The time of alluvial sedimentation and floodplain formation in the lowland humid
tropics of Ghana, Sierra Leone and western Kalimantan (Indonesian Borneo). Geomorphol. 4, 409-422.
Tobin, G. A. & Montz, B. E. (1994) The Great Midwestern Floods of 1993. Saunders College Publishing, Harcourt
Brace College Publishers, New York.
Winkler, M. G., Swain, A. M. & Kutzbach, J. E. (1986) Middle Holocene dry period in the northern midwestern
United States: lake levels and pollen stratigraphy. Quatern. Res. 25, 235-250.
Wohl, E. E. & Enzel, Y. (1995) Data for palaeohydrology. In: Global Continental Palaeohydrology ed. by
Received
K. J.3 Gregory,
April 1996;
L. Starkel
accepted& 9V.February
R. Baker,1997
23-59. John Wiley & Sons, Ltd., London.