Hydrological consequences of land use change a riveiw ofthe stat of scicen

In this chapter, 1 1 summarize various scientific methods that have been used to determine the hydrological effects of land use change; 2 review the state-of-science with respect to u

Hydrological Consequences of Land Use Change:

A Review of the State-of-Science

Keith N Eshleman

University of Maryland Center for Environmental Science, Appalachian Laboratory,

Frostburg, Maryland

Rates o f deforestation, agriculturalization, urbanization, wetland drainage, and several other types o f land use change have accelerated as a function o f the growth

o f human populations Hydrologists have recognized for nearly a half century that such land use changes can substantially affect hydrological processes at the scale o f the research plot, the hillslope, and the small experimental catchment The hydro-logical consequences o f land use change are o f interest not only to the academic hydrologist and ecologist, but they are o f critical importance to the practicing civil engineer At the beginning o f the 21st century, these studies are increasingly being incorporated into multi-scale analyses used to address both scientific and manage­

ment questions at landscape, river basin, and regional scales Such efforts are being supported by major technological developments in collecting, analyzing, and mod­

eling hydrological data, as well as new capabilities for observing and quantifying land use and land cover changes using remote sensors In this chapter, 1 ( 1 ) summarize various scientific methods that have been used to determine the hydrological effects

of land use change; (2) review the state-of-science with respect to understanding the effects o f several different types o f land use change on hydrological processes; and (3) identify key research issues related to uses o f specific methodologies and to improved understanding, detection, quantification, and prediction o f the hydrolog­

ical consequences o f specific land use changes

INTRODUCTION Land use change has been a ubiquitous component of human

settlement of the earth's surface, particularly in Europe, North

America, and Asia during the previous millenium The con­

version of land to support growing human populations is a

major component of human modification of the environment

Rates of deforestation, agriculturalization, urbanization, wet­

lands drainage, and several other types of land use change

Ecosystems and Land Use Change

Geophysical Monograph Series 153

Copyright 2004 by the American Geophysical Union

10.1029/153GM03

have accelerated as a function of the growth of human popu­ lations and economic development In recent decades, dra­ matic land use changes in Southeast Asia, South America, Africa, and elsewhere have been carefully documented by on-the-ground observations and through the use of modem remote sensing technologies In this century, it is likely that rates of expansion and intensification of agriculture, growth of urban areas, extraction of timber and other natural resources, and development of freshwater resources will all increase to meet the demands of expanding human populations for higher stan­ dards of living

Hydrologists have recognized for nearly a half century that land use changes and a variety of types of land cover distur­ bances can substantially affect hydrological processes at the

13

scale of the research plot, the hillslope, and the small exper­

imental catchment Experimental manipulations of vegeta­

tive cover—in particular deforestation and reforestation—have

demonstrated that water yields, interception losses,

evapo-transpiration rates, flood peaks, sediment transport rates, and

concentrations of many water quality constituents can be

dramatically modified through human activities on the land

surface Not only are these and other consequences of land use

change of interest to the academic hydrologist and ecologist,

but they are of critical importance to the practicing civil engi­

neer The correct statistical interpretation of long time series

of data from observational networks; the proper design and

operation of flow conveyance devices, flood control struc­

tures, drainage networks, and water systems; and the appro­

priate use of mathematical models for forecasting future

hydrological conditions are all critically dependent upon

understanding systems that have been affected by human

activities As the spatial extent and intensity of land use

changes increase in systems for which we have observational

records, our ability to casually assume stochastic stationarity

becomes more and more questionable Development of the

next generation of hydrological models will almost certainly

include a new conceptualization and parameterization of land

use change effects on land surface processes Such models will

almost certainly need to go beyond our current "static" view

of governing processes by considering the influence of both

spatial extent and pattern of different land use changes on

the watershed, as well as their t e m p o r a l v a r i a t i o n , on

"dynamic" hydrological parameters

At the beginning of the 21 st century, there is still much to

be learned from small-scale experimental and observational

studies of land use changes These studies can be made far

more valuable, however, if they are part of a multi-scale

analysis focused on how the smaller scale impacts are prop­

agated to larger river basin scales where management often

occurs Within the last dozen years, at least three independ­

ent groups of hydrologists have issued concerted pleas for

continuation, better coordination, and augmentation of basic

hydrological observation networks to better understand

h u m a n effects on the hydrological cycle One of these

groups—the Eagleson committee—specifically identified

"hydrologic effects of human activity" as one of its top five

scientific priorities in the overall field of hydrological sci­

ence [NRC, 1991]

Several important technological developments have recently

converged to facilitate greater progress in understanding the

hydrological consequences of land use change The first devel­

opment was obviously the extraordinary improvement in data

collection, data archiving, data distribution, and computa­

tional (both hardware and software) capabilities to support

such analyses The importance of the Internet as a tool for

transmitting archived, as well as real-time data, to hydrolo­ gists cannot be overstated As recently as ten years ago, just the gathering of data to perform time series analysis may have taken ten times as long as it does presently, freeing up precious human resources that can now be focused on other aspects

of the problem

A second critical development is the unprecedented feasi­ bility of observing land cover changes using remote (e.g., satellite-based) sensors Landsat data are being acquired sys­

tematically around the world [Goward and Williams, 1997]

and data from moderate resolution sensors such as MODIS are

now widely available [Justice et al, 2002] Scientists are now

able to efficiently store, manage, and process these images

in ways not possible just a decade ago Satellite data, appro­ priately calibrated and validated with ground data, can now provide information on the spatial distribution of land cover

types, as well as resolve changes over time [Hansen and

DeFries, in press] Whereas previously such information could

only be obtained for small areas using ground surveys or aer­ ial photography, satellite data can expand the spatial coverage

to much larger areas and permit analysis at more frequent time intervals Remotely-sensed imagery can also be used to quantify the spatial distribution of many land cover parame­

ters, such as vegetation cover [Nemani et al, 1993], vegeta­ tion change [Nemani et al, 1996], and imperviousness

[Slonecker et al, 2001] that are important elements of change

Finally, satellite remote sensing is increasingly being used to provide extensive coverage of key hydrological variables, such

as precipitation [Smith et al, 1996; Sturdevant-Rees et al, 2001], soil moisture [Sano et al, 1998], and river flooding

[Portmann, 1997; Townsend and Foster, 2002] Eventually it

may be feasible to subject these data to time series analysis as well Each of these types of data would have been virtually impossible to obtain through traditional field methods

[Entekhabietal, 1999]

The purposes of this chapter are to: (1) summarize various scientific methods that have been used to understand and quantify hydrological effects of land use change; (2) review the state-of-science with respect to understanding the effects

of several different types of land use change on hydrological processes, particularly those processes that are observable at the land surface; and (3) identify key research issues related

to uses of specific methodologies and to improved under­ standing, detection, quantification, and prediction of the hydro-logical consequences of specific land use changes

SCIENTIFIC METHODS FOR ASSESSING HYDROLOGICAL EFFECTS OF LAND USE CHANGE Identifying, quantifying, and predicting the hydrological consequences of land use change have proven quite

chal-ESHLEMAN 15

lenging for several reasons The relatively short lengths of

most hydrological records, superimposed on the relatively

high natural variability of hydrological systems, make it dif­

ficult to isolate a land use signal from hydroclimatological

"noise" This issue is most significant in "real" systems for

which the timing and patterning of land use changes are not

controlled, but it is even an issue for those small-scale stud­

ies for which land use modifications can be carefully imposed

The problem is also complicated by the normal paucity of

detailed hydrometric measurements that are typically used to

characterize the hydrological consequences of land use change

Stream discharge data from one location in a watershed are

most often the only data available to assess the hydrological

impacts of land use change The relatively small number of

controlled small-scale experimental studies that have been

performed has also limited progress in extrapolating or gen­

eralizing results from such studies to other systems Given

the diversity and complexity of land use changes that are tak­

ing place around the world, satisfactory techniques for ana­

lyzing the hydrological consequences of land use must be

considered to be in an early stage of development The devel­

opment of mathematical tools (i.e., models) for reliably pre­

dicting the hydrological effects of future land use changes is

in its infancy [Beven, 2000]

In this section, I classify and discuss the primary scientific

methods that have been used to assess the hydrological con­

sequences of land use change It must be emphasized that for

purposes of this chapter I do not distinguish between land

use change and land cover change (see Loveland and DeFries,

this volume), since the two issues are closely related and

essentially identical methods can be employed to address

either issue The majority of research projects that have exam­

ined the hydrological consequences of land use change have

employed scientific methods that can be described by one of

the following six classes:

• plot-scale experiments and observations

• small watershed experiments and observations

• empirical modeling

• time series analysis

• physically-based (lumped and spatially-distributed)

modeling

• landscape, river basin, and regional-scale analysis

A discussion of these six types of methods is provided in the

following sections

Plot-Scale Experiments and Observations

Experiments (and observations) made at the plot-scale are

perhaps best exemplified by a large number of studies of the

evapotranspiration process Many of these studies were aimed

at estimating consumptive use and consumptive use coeffi­ cients of different agricultural crops and natural vegetation

In these studies, consumptive use (potential evapotranspira­ tion) can be directly measured using 0.6-3.0 m (2-10 ft.) diameter tanks known as lysimeters, in which inputs of water

can be controlled and outputs measured by weighing [Blaney

etal, 1930; Pruitt and Angus, 1960; Van Bavel, 1966] Two

important advantages of plot-scale studies include (1) the ability to replicate measurements in both space and time, thus subjecting data to rigorous statistical tests of signifi­ cance; and (2) the ability to control environmental condi­ tions to some degree A major disadvantage of them is that they are best suited for measuring processes that can be (or are assumed to be) essentially occurring in the vertical dimen­ sion Direct extrapolation of plot-scale results to larger sys­

t e m s is often q u e s t i o n a b l e , u n l e s s the full r a n g e of environmental and site conditions that are found in nature were encountered in the experimental study Interception

losses by different plant species [Helvey, 1967], snow accu­ mulation and melting [Goodison et al, 1981; DeWalle and

Meiman, 1971], and soil infiltration [Black, 1991] are other

land surface processes—considerably influenced by land use—that are commonly addressed using plot-scale tech­ niques Many other hydrological studies have been conducted

at somewhat larger (i.e., hillslope or field) scales, but few of these have dealt with the issue of land use change

Small Watershed Experiments and Observations

Understanding of the effects of land use changes on hydro-logical processes has also occurred through controlled, exper­ imental manipulations of land cover in small watersheds Studies of the effects of forest management practices (includ­ ing cutting, removal activities, and regrowth of forest vege­ tation) on annual and seasonal water yields, evapotranspirative losses, interception rates, and flood peaks have been con­

ducted in forests throughout the world [Bosch and Hewlett, 1982] The development and application of the small exper­

imental watershed technique beginning in the 1950s—exem­

plified by the careful, pioneering work conducted at Coweeta

[Swank and Crossley, 1988] and Hubbard Brook [Hornbeck

et al, 1970; Likens et al, 1977] in the eastern U.S.—clearly

revolutionized the study of the hydrological effects of land use change and land management activities These studies typically involve instrumenting a pair of adjacent, small watersheds that (initially) have similar vegetation, soils, aspect, slope, and other geomorphic properties; precipita­ tion (Figure 1) and stream discharge (Figure 2) are meas­ ured using on-site instrumentation The watersheds are typically monitored for several years (the calibration period)

to establish statistical relationships for key state variables, and

one of the watersheds is then subjected to an experimental

manipulation or land use change Subsequent monitoring of

precipitation and discharge (in some cases both quantity and

quality) is used to determine differences in hydrological (and

ecosystem) responses that can be attributed to the experi­

mental manipulation The major advantage of the small exper­

imental watershed technique is that effects on important land

surface processes such as stream discharge and

evapotran-spiration can be directly measured under somewhat "con­

trolled" conditions The major limitation of the technique is

lack of replication: the resources required to properly instru­

ment two small watersheds and perform a single manipula­

tion normally preclude true experimental replication in a

probabilistic sense The literature contains far fewer exam­

ples of controlled studies of the effects of permanent land con­

versions (e.g., forest to agriculture, agriculture to urban,

etc.), however

Paired-watershed studies of urbanization and

agricultur-alization are less common in the literature, but both of these

types of land use change have generated an extensive liter­

ature from analysis of observational data from comparative

or case studies of watersheds Similar to the small experi­

mental watershed technique, these studies involve the use

of similar types of field instrumentation, but they suffer from

the lack of a pre-manipulation calibration In some cases,

these studies involved instrumenting multiple watersheds—

an advantage of the strictly paired approach of the experi­

mental technique

Empirical Modeling

A variety of types of empirical models for representing

watershed-scale runoff responses to rainfall have been devel­ oped over the years, including (1) the rational method, (2) the unit hydrograph method, (3) the USDA-SCS Curve Number method (hereafter the SCS method), and (4) the Modified Universal Soil Loss Equation (MUSLE) All four of these models have long been considered the state-of-the-art in hydrology and have been thoroughly summarized and reviewed

by Chow [1964], among others Despite major improvements

in computing and data availability, many of these empirical

methods are widely used even today Beven [2000] recently dis­

cussed some of the well-known limitations and assumptions

of empirical models, particularly the simplifying assumption

that routing of water can be assumed to occur in a linear way,

thus enabling straightforward solution by applying the principle

of superposition The assumption of linearity imposes some severe constraints on the application of these models to fairly small watersheds, although models of this form have been extended to larger, more complex watersheds by treating them

as if composed of a set of linear subsystems Such an approach

thus assumes spatial linearity as well The lack of sufficient

internal data for calibration, however, often calls into question the overall reliability of models of such complex systems The most widely used method for estimating infiltration (and storm runoff) from watersheds subjected to land use

change is the SCS method developed by the U.S Soil Con­

servation Service [1973] using data from small

experimen-Figure 1 A shielded, weighing-type precipitation gage located on the Neff Run watershed near Frostburg, Maryland

(photo courtesy of Timothy Negley)

ESHLEMAN 17

Figure 2. A truncated Parshall flume, stilling well, and digital water level recorder used to provide continuous measure­

ment of stream discharge on the Neff Run watershed near Frostburg, Maryland (photo courtesy of Timothy Negley)

tal watersheds in the U.S This empirical method was later

extended to urbanized watersheds as well The basic theory

underlying the method is that direct (storm) runoff increases

with increasing storm rainfall, according to a functional

relationship that uses curve numbers that are assigned to

different hydrologic soil groups for different land use sce­

narios: curve number (CN) values can theoretically range

from 0 to 100 The higher the value of CN, the greater the

storm runoff (all else being equal) Empirically-derived CN

values have been tabulated on the basis of land use, land

treatment or practice, hydrologic condition, and hydrologic

soil group; CN values can also be modified to account for dif­

ferences in antecedent moisture conditions A major advan­

tage of the SCS method is that hydrologic soil groups have

been mapped for the U.S.; when combined with local data on

land cover and land management practices, storm runoff can

be easily estimated for either actual or design storms using

the published CN values

Chow et al [1988] provide an especially nice example of

using the SCS method (using curve numbers tabulated from

other studies) to address the problem of increased runoff

from a watershed due to urbanization Fogel et al [1974]

also used the SCS method to predict the likely effects of dif­

ferent degrees of urbanization on the return periods of storm

runoff from desert shrub watersheds While flood frequency

analysis is most commonly performed as a component of

hydrological time series analysis, in this example flood fre­

quency curves were generated using the SCS method and

estimated parameter values that corresponded to different

land treatments The SCS method has been incorporated into

a variety of other "event-based" runoff models (including

TR-20 and HEC-HMS) and (with MUSLE) provides the con­ ceptual basis for several other empirically-based models that have been developed to address water quality problems asso­ ciated with land use changes Examples of such models

include AGNPS [Young et al, 1989], SWRRB [Williams et

al, 1984], SWAT [Arnoldetal, 1993; Bingner, 1996], and

SWIM [Krysanova etal, 1998]

Beven [2000] emphasized another modeling problem that is

peculiar to land use change modeling: in predicting the effects

of change using these models, at least some of the model

parameters must also be expected to change Therefore, unless

the effects of change can be independently accounted for through empirical study, one must expect that the uncertain­ ties associated with model parameters must increase relative

to the case of stable land use Fortunately, extensive experience with some of these models (the SCS method and MUSLE, in particular) has resulted in detailed tables of data that can be used to parameterize the model for a specific application, ostensibly improving model predictability

Time Series Analysis

Time series of data for some important hydrological variables (e.g., river flow) have been collected using consistent methods

for some watersheds for over a century by the United States

Geological Survey; data collected prior to the 1960s were typ­ ically associated with very large U.S rivers and these data are most useful in addressing issues associated with water resources development, water resources management, and climate change

in major river basins (e.g., Figure 3) Since the 1960s, however, larger numbers of relatively long time series have been

col-lected for smaller watersheds that have experienced important

land use changes, particularly urbanization in the U.S [McCuen,

2003] The improvements in data collection, data archiving,

data distribution, and computational capabilities now make

such time series analyses feasible

To be useful in identifying and quantifying hydrological

change statistically, time series methods must be able to dis­

tinguish between episodic and secular (i.e., gradual) effects in

a long-term record [McCuen, 2003] Construction of a deten­

tion basin, diversion of streamflow to another channel, removal

of a dam, or straightening of a natural river are all examples

of episodic changes that could produce a sudden transition

from one state to another Secular changes would be expected

to dominate in watersheds that are, for example, gradually

urbanized or agriculturalized One can easily imagine a water­

shed that is influenced by both types of land use change In all

of these cases, land use change can be expected to induce

nonhomogeneity or nonstationarity into the data records; a

critical issue is the amount (both intensity and extensity) of

change that can be detected, given a certain amount of random

variability in a long-term record

The flood frequency estimation problem for short (rela­

tive to the desired return period) records has been widely

discussed in the literature, but the estimation problem for

nonstationary time series attributable to land use change has

not received as much attention McCuen [2003] discusses

the use of graphical and statistical methods for detection of

nonhomogeneity in hydrological time series, and Beighley

and Moglen [2003] recently suggested a method for model­

ing changes in a flood frequency curve for a watershed in Maryland that experienced extensive urbanization during the

gaging period Entekhabi et al [1999] described an important

complication in statistical quantification of hydrological effects of land use change: hydrological variability can also

be caused by long-term climate fluctuations and climate change In systems subjected to both types of change, it is nec­ essary to isolate the individual contributions of human activ­ ities at the land surface from those that are due to climate change, particularly in any analysis of extreme events (e.g., floods, droughts, etc.) In essence, identifying a linkage between hydrological change and land use change often involves finding a signal within a substantial amount of cli­ matic variability or "noise."

Physically-Based (Lumped and Spatially-Distributed) Modeling

Computational improvements in the last thirty years have enabled far faster processing of hydrologic data and thus more rigorous testing of hydrological paradigms Modeling of land use change has thus rapidly evolved from simple empir­ ical approaches that often involved "event" modeling [e.g., unit

hydrographs, Jakeman et al, 1993] to more complex, phys­

ically-based models that could be applied to provide

"con-700

1895 1905 1915 1925 1935 1945 1955 1965 1975 1985 1995 Figure 3. Mean annual discharge in the Potomac River near Point of Rocks, Maryland (water years 1895 through 2000)

Data from http://nwis.waterdata.usgs.gov/nwis/sw

ESHLEMAN 19

tinuous" outputs or simulations based on application of the

water balance continuity equation Lumping or spatial aver­

aging is the most common simplification in watershed mod­

eling, which often limits application of the model to small

watersheds with relatively homogeneous vegetation, soils,

topography, and geologic strata [Blackie and Eeles, 1985]

The first generation of physically-based models was com­

posed of lumped-parameter models such as the Stanford

Watershed Model of Crawford and Linsley [1966], later known

as Hydrologic Simulation Program—Fortran or HSPF

[Bick-nell et al, 1997] Lumped-parameter models such as HSPF

include explicit soil moisture accounting, computation of

actual evapotranspiration rates, and routing of flow through

various compartments on a regular time step

There are quite a few examples in the literature of the use

of lumped-parameter models for evaluating and predicting

hydrological response to land use change One of the earliest

examples was the use of the USDAHL (U.S Department of

Agriculture Hydrograph Laboratory) model by Langford and

McGuiness [1976] to simulate runoff from a watershed that had

experienced both reforestation and forest thinning operations

Swift et al [1975] used a lumped-parameter model (PROS­

PER), based on the Penman-Monteith equation, to model

changes in evapotranspiration due to clear-cutting and refor­

estation of research watersheds in the southern Appalachi­

ans Huff and Swank [1985] later employed PROSPER to

evaluate effects of clear-cutting on water yields from six

research watersheds located throughout the U.S Federer and

Lash [1978] described the use of another lumped-parameter

model (Brook) to evaluate the effects of different species of

hardwood trees on the timing of transpiration and runoff from

forests in the eastern U.S Eeles and Blackie [1993] used the

Institute of Hydrology (IH) lumped model to evaluate the

effects of afforestation on streamflow within the Balquidder

catchments in the U.K An important improvement in this

work was the use of a parameter optimization scheme [e.g.,

Nelder and Mead, 1965] to automatically fix 23 of the 30

parameters in the model

Two important limitations of lumped-parameter models are

(1) their inherent inability to represent the spatial variability

of hydrologic processes and parameters [Moore et al, 1991]

and (2) the effective parameters used by these models are not

directly related to measurable watershed characteristics [Storck

et al, 1998] Therefore, it has long been thought that greater

progress could be made through the development and appli­

cation of spatially-distributed physically-based models that

can ostensibly make use of high-resolution information on

land use patterns and processes [e.g., Abbott et al, 1986;

Dunn andMackay, 1995; Adams et al, 1995] As pointed out

by Beven [1985; 2000], the primary rationale behind the move

to such models is that land use change almost never occurs both

suddenly and uniformly over an entire watershed Spatially-distributed models can, in theory, correctly implement any changes in parameter values and place them in their correct

spatial context [Beven, 1985] Recent developments in geo­

graphic information system (GIS) technology have clearly

e x p e d i t e d a d v a n c e s in s p a t i a l l y - d i s t r i b u t e d m o d e l i n g

[DeVantier andFeldman, 1993; Maidment, 1993] The most

important current limitations are the general lack of data with which to parameterize these models, the difficulties and expenses associated with gathering necessary data for para­ meterization, and the relatively large spatial variability in

those data that are available Beven [1989] also questioned

whether some of the "small-scale" physical parameters that are endemic to these models can actually be used to model phys­ ical processes at the larger grid scale, a problem common to both lumped and spatially-distributed models

The literature includes descriptions of at least five spatially-distributed watershed models that have been used to address the problem of land use impacts on hydrological processes,

including (1) IHDM [Beven, 1985]; (2) CASC2D/GRASS

[Doe et al, 1996]; (3) Macaque [Watson et al, 1999], based

on the distribution function modeling concept common to

RHESSys [Bandetal, 1993] andTOPMODEL [Beven and

Kirkby, 1979]; (4) SHE/ SHETRAN/ARNO/NELUP [Abbott

et al, 1986; Adams and Dunn, 1995; Dunn and Mackay, 1995; Dunn et al, 1996]; and (5) DHSVM [Storck et al, 1998; Van-Shaar et al, 2002]

The issue of "overparameterization" of rainfall-runoff mod­

els has been brought up in the literature [e.g., Beven, 1989;

Jakeman and Hornberger, 1993], but this issue may be some­

what less relevant to land use change modeling where the main objective is to predict how the intensity, extensity, tim­ ing, and patterning of land modifications alter hydrological

variables Beven [2000], however, points out a more important

modeling issue that must be reemphasized here:

[T]here have been many studies that report predictions

of the hydrological impacts of change but, as yet, I know

of none where catchment-scale predictions made before

a change have later been verified

In other words, without more extensive field validation of model predictions, the overall reliability of physically-based models of hydrological response must be seriously questioned

Landscape, River Basin, and Regional-Scale Analysis

Most of the techniques that have already been described in the preceding paragraphs have been used to address hydro-logical consequences of land use change at larger (i.e., land­ scape, river basin, or regional) spatial scales While land use changes only occur uniformly in space and time under

con-trolled experimental conditions, some approaches have been

used to address the consequences of land use changes in real

systems where temporal and spatial patterning is a compli­

cating factor This is a particularly important issue for river

basins—both as a research focus and as a management ques­

tion A second topic relates to the quantification or predic­

tion of hydrological consequences at even larger regional

scales where data may be particularly sparse News on and

Colder [1989] provided an excellent review (and history) of

some of the key problems in generalizing about the effects

of forest management practices on the hydrological cycle at

regional scales

HYDROLOGICAL CONSEQUENCES OF LAND USE

CHANGE

In this section, I provide a review and discussion of the

hydrological consequences of six major types of land use

change—in each case a "conversion" from one type to another

The six land use changes are: (1) deforestation/reforestation,

(2) agriculturalization, (3) urbanization, (4) wetland drainage,

(5) water r e s o u r c e s d e v e l o p m e n t , and (6) surface

mining/reclamation It must first be pointed out that these

categories are not actually mutually exclusive; for example,

in some cases deforestation may be considered one element

of agriculturalization (i.e., conversion of previously forested

land to agricultural land) The same holds true for wetland

drainage: the process of draining a marsh or swamp may be

one element of a larger process through which a wetland is

converted into agricultural land (i.e., agriculturalization) or

urban land (i.e., urbanization) Secondly, my goal in this sec­

tion is not to provide an exhaustive summary of the literature,

since such a review would be impossible given the limited

pages that are available to complete this task Moreover, the

diversity within and among these land use changes, as well

as the condition and inherent variability of natural systems

onto which these changes are imposed, makes it virtually

impossible to provide a truly "global" generalization—either

in textual or mathematical form Rather, my primary goal in

this section is to provide the reader with a reasonable summary

of the present understanding of the hydrological consequences

of each of the six major change types I have also included

under each of the change categories a brief discussion of the

effects of different management practices on hydrological

processes, where such information can be gleaned from the

scientific literature

Deforestation/Reforestation

Human conversions of forested lands—either to provide

fuel and fiber or to merely provide land for agricultural pro­

duction—has received significant attention by both ecolo-gists and hydroloecolo-gists Forests cover roughly 30% of the land surface of the earth, less than half of their original areal cov­ erage prior to human settlement Deforestation is widespread, but the extent of forest clearing is most significant in south­ ern Europe, northern Africa, and western Asia where humans have been able to make use of fertile forest soils for agricul­

tural crops or as pastures for grazing animals [Dansmann,

1972] In recent decades, the rate of clearing of tropical forests

in Africa, Asia, Central America, and South America has greatly accelerated, causing global concerns about effects on climate and local concerns about increased denudation of the

landscape, flooding, and downstream sedimentation [Werth

and Avissar, 2002]

The most significant hydrological effects of permanent

deforestation are at least qualitatively well known for many

forests Since forests naturally occur in areas of relatively high precipitation (and the density of forest vegetation varies roughly

as a function of precipitation), forest clearing causes an imme­ diate loss of interception capacity and subsequent reductions

in interception and transpirational losses of water to the atmos­

phere [Swift et al, 1975] Forested lands subjected to clearing

are also exposed to the direct impact of raindrops, inducing par­ ticle detachment from unprotected surfaces Forest soils— particularly those in temperate and boreal systems—are also characterized by relatively high infiltration capacities, owing

to the presence of undecomposed organic matter at the surface and large quantities of organic matter within the upper mineral soil layers Under these conditions, infiltration-excess overland flow is minimized Forest clearing, and the subsequent loss

of the surface organic layer and decline in soil organic matter, results in an increase in overland flow, raindrop detachment of soil particles, sheet erosion, rill erosion, gullying, and down­ stream sedimentation

While the hydrological responses of lands following defor­ estation are highly variable, forest clearing at the scale of the experimental watershed generally results in a significant

increase in annual water yield [Hibbert, 1967; Bosch and

Hewlett, 1982; Whitehead and Robinson, 1993], depending

upon a host of different factors These factors include the spe­ cific method of deforestation including cutting, wood removal,

and road-building practices [Reinhart et al, 1963; Harr et

al, 1975; Likens et al, 1978; Beschta, 1998]; the extent of

forest cover removed within the watershed [Hewlett et al, 1969; Bosch and Hewlett, 1982]; the rate and type of revege-tation or reforesrevege-tation [Federer and Lash, 1978; Swank et al,

1988]; climatic conditions, especially the temporal distribution

and magnitude of rainfall and snowmelt [Chow, 1964; Bosch

and Hewlett, 1982; Whitehead and Robinson, 1993]; and

chemical and physical properties of the soil, especially of the

forest floor [Likens et al, 1978] Other small watershed

stud-ies have examined in considerable detail changes in other

measures of streamflow response, such as flow duration curves

[Hornbecket al, 1970; Burt and Swank, 1992]; baseflow, low

flow conditions, or flow minima [Hicks et al, 1991; Bowling

et al, 2000]; peak flows [Hornbeck et al, 1970; Harr et al,

1975; Harr, 1981; Harr, 1986; Jones and Grant, 1996; Bur­

ton, 1997; Storck et al, 1998; Jones, 2000]; and the timing

and magnitude of snowmelt runoff [Harr, 1986; Beschta,

1998; Storck et al, 1998] Despite the large numbers and types

of analyses that have been conducted, it has been difficult to

generalize across these different streamflow response measures

Increases in streamflow can cause further denudation of

the land surface through increased rill and channel erosion, as

well as downstream sedimentation, but these responses are

often episodic and thus not easily predicted [Grant and Wolff,

1991] On deforested slopes in particularly steep terrain that

are subsequently subjected to intensive cropping practices or

fire or both, rates of denudation of the landscape are maxi­

mized The "end result" of this process is exemplified by bar­

ren, unstable wastelands unable to support any vegetation,

exemplified by the foothills north of Mexico City described

by Starker Leopold in 1959 [see Dansmann, 1972] Globally,

it has been estimated that the average erosion rate over the

continents has doubled or tripled due to all human activities,

but the most important factor has been forest clearing for

agriculture and forest harvesting practices

In better-managed systems where revegetation rapidly

occurs, the effects of forest clearing on hydrological processes

are transient and less extreme In these cases, hydrologists

have focused on how the changes in vegetation type and den­

sity affect the rates and timing of runoff, evapotranspiration,

sediment losses, and nutrient export from watersheds [Likens

et al, 1978; Swank et al, 1988] Experimental conversions of

natural forests to grasslands, to forest plantations, and back to

native vegetation have been used to quantify the hydrological

impacts of deforestation and to understand the transient con­

sequences of various forest management practices and natu­

ral disturbances [Swank et al, 1988]

Reforestation (or afforestation) can be conceptualized to

some extent as reversing the hydrological responses to defor­

estation Eeles and Blackie [1993] predicted that runoff would

decline in an approximately linear proportion to the increase

in the area of forest within the watershed, with a 50% increase

in forest area producing a 12% reduction in runoff

Flow-duration analysis of the model results further suggested that

these percentage reductions occurred uniformly across the

entire range of flow conditions [Gustard and Wesselink, 1993]

More recently, Colder [2003] proposed several modifications

of the Penman-Monteith equation that allowed an assessment

of the effects of afforestation on water resources In a com­

panion paper, Colder et al [2003] employed the Hydrologi­

cal Land Use Change (HYLUC) model which predicted dra­ matic declines in average annual recharge plus runoff in response to conversion of grasslands to forests in the lowland areas of the U.K Complete reversal of the effects of defor­ estation may be dependent upon restoration of both vegetation

and soil properties that are characteristic of native forests

within a particular climate on a particular parent material Given the relatively slow speed of forest soil development, particularly in temperate climates, it is expected that hydro-logical processes under these conditions would be restored quite gradually

At the river basin scale, the ability to predict integrated responses of hydrological variables over large areas has a

fairly brief history Gentry and Lopez-Parodi [1980] presented

empirical evidence that deforestation of the Upper Amazon River basin was causing increased flooding downstream, but

Nordin and Meade [1982] refuted the study's conclusions on

methodological grounds Shukla et al [1990] used a coupled

atmosphere-biosphere model to predict that complete con­ version of the Amazon tropical rainforest to degraded pas­ ture would ultimately cause a 643 m m decrease (-26%) in mean annual precipitation, a 496 mm decrease (-30%) in mean annual evapotranspiration, and a 147 mm decrease (-18%) in mean annual runoff Other modeling studies have produced similar, if less dramatic, responses to Amazon deforestation

[Werth and Avissar, 2002] Regional responses of river basins

or large regions to forest management practices have been examined using models for the Tyne River basin (U.K.) by

Dunn andMackay [1995], in the Columbia River basin (U.S.)

by Matheussen et al [2000], and in the Central Sierra region

of California (U.S.) by Huffet al [2002]

Agriculturalization

The conversion of grasslands and forested lands to agri­ cultural use is quantitatively the most dramatic land use change that has yet taken place Roughly 10-12 million k m2 of land are currently thought to be under cultivation (compared to the 18 million k m2 of land that are potentially suitable for agricultural use) Obviously, the term "potentially suitable"

is subject to debate, since some agriculture can be practiced successfully for short periods of time on even marginally-suitable lands Subsidies of energy and materials (water, fer­ tilizers, pesticides, herbicides, etc.) can extend production in these systems for longer periods Western cultivation prac­ tices—involving deep plowing, liming, manuring, mulching, crop rotation, and fallowing—had their origin in the brown for­ est soils of Europe, and were later successfully practiced in the eastern U.S When properly used, these practices can protect the soil from erosion, maintaining soil structure, fertility, and productivity for long periods of time When improperly used,

or practiced in regions that once supported tropical forests, soils

are often quickly depleted of nutrients—causing the "shift­

ing agriculture" that is so common in these areas In the

droughty grassland regions of the mid-western U.S., improper

cultivation, combined with a period of extreme drought, caused

the Dust Bowl of 1931, during which 2 to 12 inches of topsoil

were literally blown off of the land, together with thousands

of farmers The call for soil conservation was a major result

of the 1931 Dust Bowl, but water and wind erosion are still

major problems that are largely attributable to poor agricultural

practices [Dansmann, 1972]

Arnold et al [1995] simulated the changes in streamflow and

sediment yields for several watersheds in Texas that have been

"recovering" from the Dust Bowl episode In the White Rock

Creek watershed (area of 257 k m2) north of Dallas, both the

observational data and modeling results indicated a significant

decline in erosion and reservoir sedimentation during the

period 1910 to 1984 The decline in sediment yields was attrib­

uted to the combined effects of conversion from rural land to

77% urban during this period, changes in agricultural crops

(replacement of cotton by wheat and grain sorghum), and

implementation of soil and water conservation measures begin­

ning in the 1940s

In general, it is thought that those agricultural practices that

can retard soil erosion are practices that will also increase

infiltration, thus reducing and delaying surface runoff Con­

tour plowing, terracing, strip cropping, and no-till cropping are

four newer practices that attempt to reduce erosion of the land

by water Applications of the SCS method indicate that storm

runoff from agricultural lands is almost always greater than

runoff from natural forested or range lands, all else being

equal, while reductions in CN values are predicted through the

use of better or best management practices There are few

observational or experimental studies that have demonstrated

the effects of agriculturalization or agricultural management

practices on hydrological processes at any scale One excep­

tion is the Goodwin Creek Research Watershed, a 21.3 k m2

basin in north central Mississippi where the percentage of

cultivated land decreased from 26% to 12% over an 8-year

period from 1982 to 1990 Using a combined experimen­

tal/modeling approach, Kuhnle et al [1996] showed that 42%

of the decline in total fine sediment concentrations could be

explained by simulated reductions from agricultural lands

The researchers were able to attribute the decline in sediment

concentrations to reductions in peak discharge during the

growing season, which was shown to vary as a function of

percent cultivation

Several studies of changes in hydrological processes have

also been conducted in the Driftless Area of southwestern

Wisconsin, an area dominated by agriculture Trimble [1981]

showed that rates of upland sheet and rill erosion and down­

stream sedimentation declined by nearly 30% in the Coon Creek watershed (area of 360 k m2) between the periods

1850-1940 and 1940-1975 Potter [1991] showed a signifi­

cant decrease in flood peaks and winter/spring flood volumes

in the East Branch of the Pecatonica River (area of 572 k m2)

during the period 1940 to 1986 Gebert andKrug [1996] per­

formed a similar time series analysis of annual flood peaks and 7-day low flows for gaging stations in the Driftless Area, thus demonstrating significant reductions in flood peaks and increases in low flows during the 20th century (but no trends for forested basins in northern Wisconsin during the same period) All three studies attributed these effects to changes in agricultural management practices, particularly the wide­ spread use of soil conservation practices

Hydrologists have also concerned themselves with how agricultural practices can increase rates of evapotranspiration from crops (i.e., consumptive losses) In much of the U.S., for example, irrigation is a major off-stream use of both sur­ face and groundwater resources Consumptive water use by irrigated crops is a major component of the water balance at

many scales [Solley et al, 1998] Among the most signifi­

cant hydrological impacts of agricultural activities, ground­ water withdrawals for irrigation in the western U.S have contributed to dramatic increases in evapotranspiration, exces­ sive declines in water tables, surface subsidence, and soil salinization

Urbanization

The conversion of forest or agricultural land to urban use has major ramifications for hydrological processes In a review

published nearly a half century ago, Savini and Kammerer

[1961] summarized the hydrological effects of urbanization and qualitatively described the effects based on an analysis of sev­ eral different stages of land use change: (1) transition from pre-urban to early-pre-urban, (2) transition from early-pre-urban to middle-urban, and (3) transition from middle-urban to

late-urban Savini and Kammerer [1961] also tried to distinguish

between the hydrological effects associated with the human uses of water from effects associated with human uses of the land, but they noted that relatively few studies had yet been conducted in order to quantify the effects of urbanization on hydrological systems Common to all three stages of urban­ ization are: (1) decreases in transpiration from loss of vege­ tation, (2) d e c r e a s e s in infiltration due to d e c r e a s e d perviousness, (3) increases in storm runoff volumes, (4) increases in flood peaks, (5) declines in water quality from dis­ charges of sanitary wastes to local streams and rivers; and (6) reductions in baseflow Design, installation, and maintenance

of urban drainage systems for collection and disposal (i.e.,

"routing") of stormwater to reduce damage from floods is a