Variability in the Environmental Factors Driving Evapotranspiration

Variability in the Environmental Factors Driving Evapotranspiration from a GrazedRangeland during Severe Drought Conditions

JOSEPH G. ALFIERI* AND PETER D. BLANKEN

Department of Geography, University of Colorado, Boulder, Colorado

DAVID N. YATES

Research Applications Laboratory, National Center for Atmospheric Research, Boulder, Colorado

KONRAD STEFFEN

Department of Geography, University of Colorado, Boulder, Colorado

(Manuscript received 9 February 2006, in final form 14 August 2006)

ABSTRACT

Nearly one-half of the earth’s terrestrial surface is susceptible to drought, which can have significantsocial, economic, and environmental impacts. Therefore, it is important to develop better descriptions andmodels of the processes linking the land surface and atmosphere during drought. Using data collectedduring the International H2O Project, the study presented here investigates the effects of variations in theenvironmental factors driving the latent heat flux (�E ) during drought conditions at a rangeland site locatedin the panhandle of Oklahoma. Specifically, this study focuses on the relationships of �E with vaporpressure deficit, wind speed, net radiation, soil moisture content, and greenness fraction. While each ofthese environmental factors has an influence, soil moisture content is the key control on �E. The role of soilmoisture in regulating �E is explained in terms of the surface resistance to water vapor transfer. The resultsshow that �E transitioned between being water or energy limited during the course of the drought. Theimplications of this on the ability to understand and model drought conditions and transitions into or outof droughts are discussed.

1. Introduction

Since almost one-half of the earth’s terrestrial surfaceis susceptible to drought, it is a widespread phenom-enon that has significant social, economic, and environ-mental impacts (Korgan 1997). As a result, it is neces-sary to be able to model and predict the duration, ex-tent, and severity of drought in order to mitigate itsnegative consequences. To accomplish these tasks,however, it is first necessary to have a thorough under-standing of the effects of drought on the processes link-

ing the land surface to the atmosphere. In this study,the authors investigate one of those linkages, the latentheat flux (�E), by exploring how the influence of thefactors driving �E varies with changing environmentalconditions during drought. The understanding gainedthrough this study may then be used to enhance landsurface and atmospheric models so that they are betterable to describe drought.

At the most basic level, drought exists when there isnot enough water to meet the demand for it (Redmond2002). Recognizing drought is a relative conditionwherein there is an imbalance between the supply ofwater and the demand for water (Heim 2002), theAmerican Meteorological Society (1997) has definedfour categories of drought: meteorological or climato-logical drought, agricultural drought, hydrologicaldrought, and socioeconomic drought. Meteorologicaldrought may be defined as a prolonged period duringwhich there is an absence or reduction in precipitation.

* Current affiliation: Department of Agronomy, Purdue Uni-versity, West Lafayette, Indiana.

Corresponding author address: Joseph G. Alfieri, Departmentof Agronomy, Lilly Hall of Life Sciences, 915 West State Street,Purdue University, West Lafayette, IN 47907-2054.Email: [email protected]

APRIL 2007 A L F I E R I E T A L . 207

DOI: 10.1175/JHM569.1

© 2007 American Meteorological Society

JHM569

Agricultural drought is defined as a period duringwhich there is insufficient soil moisture to meet theneed of plants, most typically nonirrigated crops. Hy-drological drought, which can persist long after the me-teorological drought has ended, is defined in terms ofthe effects of meteorological drought on streamflow,groundwater, and the other components of hydrologi-cal systems. Finally, socioeconomic drought is definedin terms of the effects on the availability of economicgoods and services. This study emphasized meteoro-logical drought and, through its impacts on soil mois-ture and �E, agricultural drought. However, it is im-portant to recognize that no one category of droughtcan be considered in isolation from the others becauseall are interconnected and have the potential to havesignificant socioeconomic and environments impacts.

According to data available from the National Cli-matic Data Center (2002), nearly 10% of the total landarea of the United States—approximately 89 000 000ha—was experiencing either severe or extreme droughtat any given time during the last century. Furthermore,there have been several notable exceptions when alarger segment of the United States was affected bydrought. For example, the drought of the 1930s ex-tended from California through the IntermountainWest and the Great Plains into the Great Lakes regionof the United States (Skaggs 1975). A second exampleis the drought of 2002, which, at its peak, encompassedapproximately 39% of the land area of the UnitedStates (Douglas et al. 2003).

According to Obashi (1994), who cites statistics fromthe World Meteorological Organization, for the periodbetween 1967 and 1991, drought impacted 1.4 billionpeople worldwide, the same number of individuals thatwere negatively affected by floods, hurricanes, and allother weather-related natural disasters combined. Dur-ing that same period, 1.3 million of the 3.5 milliondeaths attributable to natural disasters globally weredue to the effects of drought.

Drought also can have substantial economic costs;indeed, it has been suggested that drought is the cost-liest of all weather-related natural disasters (Wilhite2002). As pointed out by Diaz (1983), Riebsame et al.(1991), and others, drought can affect the water supply,water quality, agricultural and timber productivity,power generation, and recreational activities. Droughtalso impacts some unexpected industries. For example,the drought in the United States during 1988 resulted inlosses of revenue in excess of $1 billion to the bargeindustry (Changnon 1989). Overall, the drought of 1988cost 5000 lives (Trenberth and Guillemot 1996) andmore than $30 billion in related damages (Svoboda etal. 2002). The total economic cost of all major drought

events in the United States since 1980 exceeds $100billion (Lawrimore et al. 2002).

Additionally, drought has a significant impact onmost ecosystems. Drought can be devastating to wet-lands and riparian habitats, rangelands, and forestedregions (Riebsame et al. 1991). Perhaps the starkestexample of the relationship between drought and ecol-ogy can be seen in wildfire. For example, the early startand high severity of the 2002 fire season in the westernUnited States can be attributed, at least in part, todrought (Douglas et al. 2003).

Given the effects of drought, it is clear that accuratepredictions of the severity, extent, and duration ofdrought would be beneficial so that the adverse conse-quences of drought can be minimized. Since the qualityof a model’s predictions are dependent on the strengthof the underlying knowledge of those physical pro-cesses represented by the model, a solid understandingof the linkages between the land surface and the atmo-sphere is a key prerequisite for model development. Abetter understanding of the linkages between the landsurface and the atmosphere is the first step toward im-proving the capabilities of land surface models to de-scribe drought.

This study focuses on improving the understanding ofone important linkage between the land surface and theatmosphere, �E. By investigating how the environmen-tal factors driving �E vary with changing environmentalconditions, the key controls on �E were isolated. Inturn, a better understanding of the controls on �E dur-ing drought suggested relationships and methods forimproving land surface models.

The study site, data collection methods, and postpro-cessing procedures, and principal component regres-sion analysis are described in section 2. The results ofthe analyses are presented and discussed in the section3. Finally, conclusions and a brief discussion of ongoingresearch are presented in section 4.

2. Methodology

a. Site description

The data used in this study were collected as a part ofthe International H2O Project (IHOP_2002), a multi-agency field campaign conducted in the southern GreatPlains of the United States during May and June 2002(Weckwerth et al. 2004; LeMone et al. 2007). The datawere collected between 20 May and 16 June 2002 atIHOP_2002 site 10 located in the panhandle of Okla-homa (36.88°N, 100.61°W; Fig. 1) northwest of the cityof Beaver. Although the site, which was dominated bya single species of C4 grass, Andropogon gerardii, wasnot grazed during the field program, it was heavily

208 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8

grazed prior to the observation period. The vegetationwas distributed in an intricate mosaic or patchworkwherein approximately 30% of the surface was coveredwith clumps of vegetation and the remaining surfacewas bare soil. Because of the drought conditions, thevegetation remained dormant until 30 May; this isatypically late since the growing season usually beginsnearly a month earlier (Gould and Shaw 1983).

The site experienced a protracted period of severe toextreme drought prior to and through the duration ofthe field campaign. According to the index developedby the U.S. Drought Monitor and used to characterizedrought in this study, severe drought is characterized bywater shortages, moderate crop and pasture losses, veryhigh fire risk, and precipitation levels 50%–60% belownormal for the preceding 3–4-month period. Similarly,extreme drought is characterized by widespread watershortages, major crop and pasture losses, extreme firerisk, and precipitation levels 60%–70% below normalfor the preceding 4–5-month period (Svoboda et al.2002). (Further information regarding the DroughtMonitor may be found online at http://www.drought.unl.edu.)

According to the Oklahoma Water Resources Board(OWRB), the panhandle region of Oklahoma, whichincludes IHOP_2002 site 10, received only 30%—approximately 216.0 mm—of its long-term average pre-cipitation during the 7 months prior to the beginning ofthe IHOP_2002 field campaign (OWRB 2002a). Addi-tionally, for the period leading up to and including theIHOP_2002 field campaign, 1 March to 1 July 2002, thepanhandle of Oklahoma received 112.5 mm of total

rainfall, which is 45% of the long-term average(OWRB 2002b). For comparison, this same region re-ceived more than twice the 2002 total for the sameperiod during 2003 when the total rainfall was 251.0 mm(OWRB 2003). According to the Oklahoma Climato-logical Survey (OCS), the air temperature was slightlyabove the long-term average for both May and June2002. The mean monthly air temperature for BeaverCounty, Oklahoma, during May 2002 was 20.4°C, ap-proximately 2.1°C above average (OCS 2002a); duringJune 2002 the mean monthly air temperature was 25.0°C,approximately 1.0°C above average (OCS 2002b).

The drought conditions experienced at the study siteare common in the southern Great Plains of the UnitedStates. Historic records, as well as tree-ring analyses,indicate that this region has experienced extended pe-riods of drought at least three times since the mid-1800s(Woodhouse et al. 2002) and numerous short-durationdroughts, such as the 1988 drought (Riebsame et al.1991). The 2002 drought, which extended across 39% ofthe land area of the United States at its peak, wasamong the 10 driest on record. According to the OCS(2002c), it resulted in water shortages, multiple intensewildfire outbreaks, and economic losses exceeding $250million for the state of Oklahoma alone.

b. Micrometeorological measurements

The micrometeorological data were collected usingan eddy covariance micrometeorological station posi-tioned 150 m from both the northern and eastern edgeof the research site. The station was equipped with anarray of instruments, and the data were stored as 30-min block averages in a datalogger (model CR23X,Campbell Scientific, Logan, Utah). The system waspowered via a 12-V, 100 A-h battery trickle-chargedusing a solar panel.

The micrometeorological measurements includedwind speed and virtual air temperature (Schotanus etal. 1983) using a sonic anemometer (model CSAT3,Campbell Scientific) mounted facing due east, the di-rection of the prevailing wind, at a height of 3 m abovethe ground. Water vapor density was measured using akrypton hygrometer (model KH2O, Campbell Scien-tific) mounted facing east at a height of 3 m with ahorizontal displacement of 15 cm from the sonic an-emometer. Both instruments operated at a samplingfrequency of 10 Hz.

A standard suite of transformations and correctionswas applied during postprocessing in order to deter-mine the sensible heat flux (H) and �E. The first ofthese transformations was a coordinate rotation of thewind components such that both the mean crosswind(�) and vertical component (w) were equal to zero

FIG. 1. The location of the research site, IHOP_2002 site 10,along with the land use/land cover of the surrounding region,which includes the entire domain of the International H2O Project2002.


(Pond et al. 1971; Kaimal and Finnigan 1994). Next,several corrections including the correction for attenu-ation of the measured water vapor concentration due tooxygen absorption (Campbell Scientific 1998), the ad-justment for the effects of buoyancy and density (Webbet al. 1980), and the correction for the horizontal dis-placement of the sonic anemometer and the kryptonhygrometer (Horst 2006) were applied to the measure-ments of �E.

Additional measurements included net radiation(Rnet; model Q*7, Radiation Energy Balance Systems,Seattle, Washington) and incident solar radiation(model Eppley pyranometer, Eppley Laboratory, New-port, Rhode Island). Both of these instruments weremounted at a height of 3 m facing due south. Radiativesurface temperature was measured via an infrared ther-mometer (model IRT-4000, Everest Interscience, Tus-con, Arizona) mounted facing due west at a height of 3m and oriented at a 45° angle such that the sensor mea-sured the temperature over a 1-m2 sampling area rep-resentative of the site as a whole. Finally, precipitationwas measured using a tipping-bucket rain gauge (modelTE525WS, Texas Electronics, Dallas, Texas).

Additional measurements used in this research werecollected at IHOP_2002 site 3, which was located ap-proximately 1.25 km southeast of site 10. These mea-surements, which were overseen by the AtmosphericTechnology Division of the National Center for Atmo-spheric Research, included atmospheric pressure, mix-ing ratio, and photosynthetically active radiation(PAR) measured using a digital barometer (model PTB220, Viasala, Helsinki, Finland), an integrated humidityand temperature sensor (model Hummiter 50Y, Vi-asala), and a PAR sensor (model LI-190SA QuantumSensor, LI-COR Biosciences, Lincoln, Nebraska), re-spectively.

c. Soil measurements

Soil properties, including temperature (Tsoil), volu-metric moisture content (�), and heat flux (G), weremeasured with a number of instruments buried ap-proximately 2 m due north of the micrometeorologicaltower at a range of depths from 2.5 to 10.0 cm (Fig. 2).This position was chosen so as to minimize interference

with the measurement of both net radiation and infra-red surface temperature. The depths were chosen inorder to maximize the accuracy of G and to create anear-surface soil moisture profile.

Soil temperature was measured with a soil tempera-ture probe (model STP-1, Radiation Energy BalanceSystems, Seattle, Washington) buried at a depth of 2.5cm. Measurements of G were taken with a pair of soilheat flux plates (model HFT-3, Radiation Energy Bud-get Systems) buried at a depth of 5 cm and separated bya horizontal distance of 30 cm. The measured G at theheat flux plates was corrected to the surface by account-ing for heat storage in the overlying soil layer as de-scribed by Oke (1987).

To determine �, time domain reflectometry–based(TDR) probes (model CS615, Campbell Scientific)were used. Three probes were inserted horizontallyinto the soil at depths of 5.0, 7.5, and 10.0 cm at thesame location as the other subsurface sensors. The soilmoisture measurements at 5.0 cm were used in thisanalysis since it was nearest the surface-atmosphere in-terface; thus, it should have the strongest relationshipwith both the total �E and �E from the soil.

d. Surface measurements

In addition to the continuously monitored data de-scribed above, both leaf area index (LAI) and multi-spectral reflectance were measured on four days (19May, 29 May, 7 June, and 16 June 2002) during the fieldcampaign. LAI was measured using a plant canopy ana-lyzer (model LAI-2000, LI-COR Biosciences) while thereflectance data were collected using a multispectralreflectometer (model MSR5, Cropscan, Rochester,Minnesota). Each of these measurements were made at10 positions located at 5-m intervals along a pair of20-m transects. One of the transects ran north–southbeginning 10 m south of the micrometeorological sta-tion while the other transect ran east–west beginning 10m west of the micrometeorological station (Fig. 3).

In the case of the LAI, the mean of the 10 measure-ments was used for further analyses. In the case of themultispectral reflectance data, the normalized differ-

FIG. 2. The type and position of each of the soil sensors areshown. STP refers to the soil temperature probe; HFT3 refers tothe soil heat flux plates; and CS615 refers to the soil moistureprobes.

FIG. 3. The position relative to the micrometeorological station(S) of each of the measurement points that make up the pair oftransects.


ence vegetation index (NDVI) was first calculated, andthen the mean value was determined for further analy-ses. Vegetation was also characterized using the green-ness fraction (Fg) calculated from the NDVI data fol-lowing the method described by Gutman and Ignatov(1997):

Fg �Ni � Nmin

Nmax � Nmin, �1�

where Ni is the NDVI for the time period of interest,and Nmin and Nmax are the minimum and maximumvalues of NDVI, respectively. Per Carlson and Ripley(1997), domain constants for Nmin and Nmax were de-termined for the IHOP_2002 domain and used in lieu ofGutman and Ignatov’s (1997) global constants to yieldmore accurate, site-specific results. Here Nmin and Nmax

were calculated respectively as 105% of the minimumand 95% of the maximum NDVI value measured overthe IHOP_2002 domain. The resulting values were 0.08and 0.70 for Nmin and Nmax, respectively.

e. Gap-filling method

Because of such factors as instrument failure, inclem-ent weather, and theoretical or practical considerations,micrometeorological datasets are seldom continuous(Falge et al. 2001). In the case of this study, gaps con-stituted approximately 1% and 4% of the H and �Edatasets, respectively. While these gaps made up only asmall portion of the total data collected and the gapswould not impact this analysis, a gap-filling procedurewas utilized to generate a continuous dataset. Gap fill-ing the data allowed the same dataset to be used in thisstudy as will be used in follow-up studies with landsurface models.

Although a number of gap-filling techniques areavailable, the technique ultimately chosen combinesmoving window mean substitution with a scaling tech-nique. The first step of the gap-filling technique was togenerate a lookup table (e.g., Table 1) of mean valuesfor each of the 30-min blocks that span a single day. Anew lookup table was created for each gap using a mov-ing window containing the data for the 5-day periodcentered on the midpoint of the gap. To fill the lookuptable, the mean value (xt) of some variable x for thetime of day t was calculated using the following equa-tion:

xt � �i�1

n xt,i

n, �2�

where xt,i represents a given measurement of x (x couldbe �E, H, G, or any variable requiring gap filling) takenat time of day t within the subset of observations de-

fined by the moving window, and n represents the totalnumber of measurements of x taken at time of day twithin the subset of observations defined by the movingwindow. Once calculated, the lookup table was used asthe foundation for both the gap-filling step via substi-tution and the calculation of the scaling factors ().

The second step of the process was to calculate foreach 30-min block within a given gap of length k. Toaccomplish this, was first calculated for the endpointsas follows:

�b �xb

xb

, �3�

�e �xbk1

xbk1

, �4�

where b is the scaling factor calculated for the firstendpoint, that is, the last valid data point prior to thebeginning of the gap, which was measured at time ofday b; e is the scaling factor calculated for the secondendpoint, which was measured at time of day e, that is,the first valid data point after the end of the gap locatedat time b k 1; xb is the last valid data point prior tothe beginning of the gap, which was measured at time ofday b; xbk1 is first valid data point after the end of thegap, which was measured at time of day e; xb is themean value for time of day b taken from the lookuptable; and xbk1 is the mean value for time of day etaken from the lookup table. Once the endpoint valueswere determined, the for each 30-min block withinthe gap were calculated via linear interpolation as fol-lows:

TABLE 1. A sample lookup table used in the gap-filling processfor the sensible heat flux. The values represent the 5-day averagecentered on local noon for DOY 152.

Midpointtime (LT)

Meanvalue

Midpointtime (LT)

Meanvalue

Midpointtime (LT)

Meanvalue

0015 �33.8 0815 26.0 1615 228.10045 �34.8 0845 49.9 1645 200.60115 �31.8 0915 78.9 1715 185.10145 �25.4 0945 118.9 1745 163.10215 �27.9 1015 133.8 1815 121.90245 �24.4 1045 152.3 1845 89.40315 �18.5 1115 188.4 1915 61.90345 �19.7 1145 192.6 1945 34.10415 �25.5 1215 239.2 2015 1.50445 �23.4 1245 262.6 2045 �25.80515 �22.2 1315 284.4 2115 �27.10545 �22.5 1345 301.6 2145 �39.10615 �28.2 1415 305.8 2215 �27.60645 �22.0 1445 280.3 2245 �35.80715 �24.4 1515 244.7 2315 �34.70745 �7.8 1545 241.2 2345 �33.6


�bl ��e � �b

k 1l �b, �5�

where bl is the lth within the given gap and l is anindex ranging from 1 to k for the given gap.

The final step in the gap-filling process was to sub-stitute a scaled mean into each 30-min block within thegap. The value to be used in the substitution is deter-mined as follows:

xbl � �blxbl, �6�

where xbl is the lth 30-min block within the gap, bl

is the scaling factor the lth 30-min block within the gap,xbl is the mean value for time of day b l taken fromthe lookup table, and l is an index ranging from 1 to kfor the given gap.

f. Analysis techniques

Principal component regression analysis (PCR) wasused to determine the influence of the factors control-ling �E. This method was selected because it minimizedthe effects of multicollinearity while retaining the maxi-mum amount of information regarding the relative im-portance of any given predictor variable (Wilks 1995).Further, PCR allowed the relative influence of the pre-dictor variables to be determined from the loading co-efficients associated with the principal components.

PCR takes advantage of the mathematical transfor-mation commonly referred to as principal componentanalysis (PCA) or empirical orthogonal function analy-sis (EOF). PCA rotates some number, n, of interrelatedvariables through n-dimensional space to generate a setof n new independent variables (Fig. 4), that is, princi-pal components, as a weighted combination of theoriginal variables (Jolliffe 2002). While the family ofprincipal components retains all of the information con-tained within the original variables, principal compo-nents have an ordered hierarchy in which greatestamount of information from the original set of n vari-ables is captured by the first PC and then the amount ofinformation contained within each PC decreases witheach subsequent PC until the least amount of informa-tion is contained in the last or nth PC (Jackson 1991).The hierarchal characteristic of PCs implies that thenumber of PCs used in further analysis can be reducedwith minimal loss of information by eliminating thelower order PCs (Montgomery and Peck 1992).

In this study, only the first principal component wasconsidered because it consistently accounted for at least70% of the variance observed in the response variableand demonstrated a strong correlation with the re-sponse variable. The relative influence for any given

predictor could be determined from the loading coeffi-cients associated with the first principal component bystandardizing the loading coefficient associated with agiven predictor variable against the sum of all of theloading coefficients. This may be expressed as a per-centage as

IF � 100 �WF

�W, �7�

where IF is the relative influence of a given environ-mental factor, WF is the loading coefficient associatedwith that given environmental factor, and �W is thesum of all of the loading coefficients.

Five response variables were regressed against thefirst principal component associated with five environ-mental factors. The response variables included the to-tal �E, one of its two components, soil evaporation(�Esoil) or transpiration (�Eveg), and the surface resis-tances associated with bare soil (rsoil) or a vegetatedsurface (rveg). The five predictor variables were vaporpressure deficit (D), which is defined as the differencebetween the saturation water vapor pressure and am-bient water vapor pressure (Oke 1987), horizontal windspeed (U), net radiation (Rnet), soil moisture content(�), and greenness fraction (Fg). The saturation watervapor pressure used in determining D was calculated as

FIG. 4. A simple two-dimensional example of principal compo-nent analysis is shown. The original variables, X and Y, are ro-tated in two dimensions to yield the principal components X� andY�. The principal components contain all of the information con-tained within the original variables; however, most of that infor-mation is contained within the first principal component, X�.


a function of ambient air temperature using the Clau-sius–Clapeyron equation (Bolton 1980); �E was parti-tioned into �Esoil and �Eveg according to the two-sourcemodel proposed by Shuttleworth and Wallace (1985).The environmental factors were selected using the gen-eral guidelines provided by Draper and Smith (1981)and Montgomery and Peck (1992). The resulting subsetof environmental factors demonstrated a strong practi-cal and theoretical relationship with �E and eliminatedthe redundancy present in the complete dataset.

3. Results and discussion

a. Temporal variability in the environmentalconditions

Figure 5 shows that the components of the surfaceenergy budget varied significantly during the course ofIHOP_2002 field campaign. When only rain-free dayswere considered, as was the case in this analysis, Rnet

(Fig. 5a) demonstrated a consistent temporal patternwith an overnight minimum averaging �51 W m�2 anda midday average peak approaching 660 W m�2. Therelatively small standard deviations of both the middaypeak value (19.1 W m�2) and the overnight minimum(7.2 W m�2) underscore the consistency in the observedRnet. In contrast, H (Fig. 5b) varied not only diurnally,but also over the observational period. Initially, H ex-ceeded 400 W m�2 at midday; but, after rain events(e.g., on Day of Year 144 and 148), the midday H de-creased dramatically to as little as 260 W m�2. Thevariability in the midday H was also demonstrated bythe relatively large standard deviation of 58 W m�2,which was more than 3 times the standard deviation for

the midday Rnet. On a diurnal basis, �E had an averagerange of nearly 160 W m�2 but the diurnal range ex-ceeded 250 W m�2 on days following rain events (Fig.5c). As was the case with H, the variation in �E isclearly demonstrated by looking at the midday peakvalues; the midday peak had an averaged value of 135W m�2 and a standard deviation of 48.9 W m�2. Overthe course of the IHOP_2002 field campaign, three dis-tinct periods related to the behavior of �E could beseen: the period prior to day of year (DOY) 148 whenthe mean daily �E was 40 W m�2, the period fromDOY 148 to 155 when a peak flux of 242 W m�2 wasmeasured on DOY 148 followed by a period of gradualdecline, and the period from DOY 156 to 166 when apeak �E of 363 W m�2 was measured on DOY 156followed by a period of gradual decline in the peakdaily �E. Finally, G (Fig. 5d) showed a consistent pat-tern with an average peak flux of approximately 80 Wm�2 and a standard deviation of 12.1 W m�2.

There were no clear long-term patterns over the ob-servational period for either D (Fig. 6a) or U (Fig. 6b).On both a diurnal basis and across the observationalperiod, D varied over a range of approximately 54 mbwith a mean near 18 mb and a standard deviation of 13mb. Similarly, U ranged from 0 to nearly 13 m s�1 witha mean value over the entire observational period of5.1 m s�1 and a standard deviation of 2.5 m s�1. A clearpattern was apparent for � (Fig. 6c), however, and thispattern was tied to rain events. As can be seen by com-paring Fig. 6c with Fig. 7d, although � was initially lessthan 10%, it increased after rain events to more than39% and 32% on DOY 148 and DOY 156, respectively.(Figure 7 provides a time series of several key meteo-rological variables including air temperature, atmo-

FIG. 5. Temporal variability is shown for each of the compo-nents of the surface energy budget including (a) net radiation, (b)the sensible heat flux, (c) the latent heat flux, and (d) the soil heatflux.

FIG. 6. The time series of the measurements of (a) water vapordeficit, (b) wind speed, (c) soil moisture content, and (d) green-ness fraction.


spheric pressure, relative humidity, and rainfall.)Nearly 14 mm of rain fell on DOY 146 and 147 and 13mm of rain fell overnight on DOY 155 into the morningof DOY 156. Each of the spikes in � was followed by aperiod of slow dry down. Finally, Fg (Fig. 6d) increasedin an exponential fashion, as would be expected duringthe early phases of plant phenology (Barbour et al.1999), from zero at the beginning of IHOP_2002 tonearly 0.30 at the end of the observational period.

b. Principal component regression analyses usingthe total latent heat flux

To understand the relative influence of the environ-mental factors, the data were first sorted chronological-ly and PCR was conducted on each unique time periodusing the total �E as the response variable. The timeperiods selected for this analysis included the entireobservational period, the portion of the field campaignduring May, the portion of the field campaign duringJune, and each of the four weeks during IHOP_2002(Fig. 8). With the possible exception of the increasinginfluence of vegetation as measured via Fg seen in thelatter weeks of the field campaign when the grass hadtransitioned from a dormant to active state, there wereno clear temporal patterns observed in the relative in-fluence of the environmental factors driving �E. How-ever, this analysis did show that � was consistently a keyfactor in driving �E; overall, during the entire observa-tional period the relative influence of � was 39%, whichis more than 10% greater than the next most significantfactor, wind speed, which had a relative influence ofapproximately 28%. Especially during the earliestweeks of IHOP_2002 when the site was driest, the mag-nitude of �E was limited by water availability.

To understand the effects of temporal variability inthe environmental factors driving �E, the data weresorted according environmental conditions into one ofthree subgroups defined according to the standard de-viation of the environmental factor of interest. For ex-ample, when the data were sorted according �, the threesubgroups were

Dry soils: � � s � �i ; �8a�

Moist soils: � � s � �i � � s ; �8b�

Wet soils: � s � �i ; �8c�

where � is the daytime mean � for the whole observa-tional period (15.8%), s is the standard deviation of �for the whole observational period (8.7%), and �i is thedaytime mean � for a given day during the observation.Seven days were sorted into the dry soils category; ninedays were sorted into the moist soils category; and, fivedays were sorted into the wet soils category. PCR wasthen conducted on each of these subgroups. While thisanalysis was conducted with the data sorted accordingto each of the five environmental factors of interest, thefocus here is only on the three most influential factors:�, D, and Fg.

When PCR was conducted on the data sorted accord-ing to �, several patterns became evident (Fig. 9). First,this analysis clarified and reconfirmed the results sug-gested by the PCR analysis when the data were sortedby time, that the influence of � was greatest when thesoils were the driest. Under dry soil conditions, the rela-

FIG. 7. Times series for several common meteorological mea-surements including (a) air temperature, (b) atmospheric pres-sure, (c) relative humidity, and (d) daily total precipitation.

FIG. 8. The PCR results when the total latent heat flux was usedas the response variable and the data were divided temporallyusing periods ranging from a single week to the entire study pe-riod. The investigation periods were as follows: Study: 20 May–16June; May: 20–31 May; June: 1–16 June; week 1: 20–26 May; week2: 27 May–2 June; week 3: 3–9 June; week 4: 10–16 June.


tive influence of � was nearly 40% while under wet soilconditions it was merely 15%. These results suggestthat decreased water availability was an important con-trol on �E during dry conditions.

Also, � appeared to impact the role several otherenvironmental factors played in driving �E. For ex-ample, U had the greatest influence on �E under dryconditions but it played only a minor role when the soilwas wet. This pattern was due to U maintaining a highatmospheric demand for water vapor by continually in-troducing dry air to the surface. Maintaining a highatmospheric demand, in turn, facilitated the maximumrate of evaporation possible given the limited watersupply.

As the soil became moist, the influence of Rnet in-creased sharply. Under dry soil conditions, Rnet has arelative influence of approximately 5% but it increasedto nearly 34% under moist soil conditions and re-mained high under wet soil conditions. The importanceof Rnet under moist and wet conditions suggests athreshold where the amount of energy available toevaporate water is no longer sufficient to evaporate allof the moisture available. Under wet conditions, �E atthe site became energy, not water, limited. In general,wet soils may have an albedo only half that of dry soils(Jury and Horton 2004). While this would tend to in-crease Rnet under wet conditions, the overcast condi-tions during rain events resulted in an overall reductionin Rnet as � started to increase. Immediately followingrain events, clear-sky conditions resulted in an increasein Rnet as � started to decrease. Therefore, the relation-ship between Rnet and � was relatively weak due to the

fast response of the soil to rain events. The weakness ofthe relationship is evidenced by the correlation coeffi-cient (r) of 0.49.

Finally, while the vegetation was dormant during theperiod when the soil was driest, it appeared that impor-tance of Fg increased in direct relation to �. The appar-ent relationship between Fg and � suggests that moistersoils facilitated the uptake of water by the vegetationthrough an increase in the hydraulic conductivity,which provided a second pathway for the transfer ofwater to the atmosphere.

The data were next sorted according to D and thePCR was repeated using the total �E as the responsevariable. The subgroups were defined as follows:

Low: D � s � Di ; �9a�

Intermediate: D � s � Di � D s ; �9b�

High: D s � Di ; �9c�

where D is the daytime mean of D for the entire ob-servational period, 25 mb; s is the standard deviation ofthe daytime D for the entire time period of the re-search, 8 mb; and Di is the daytime mean D for a givenday. Five days were sorted into the low category; 11days were sorted into the intermediate category; and 5days were sorted into the high category.

This analysis clarified the role of vegetation as rep-resented by Fg (Fig. 10). When the vegetation was dor-mant, Fg falls out of the analysis since the relative in-fluence of Fg goes to zero. Once vegetation was present,as D increased, the role of Fg decreased. This inverse

FIG. 9. The PCR results when the total latent heat flux was usedas the response variable and the data were sorted by soil moisturecontent. The criteria for partitioning the data are given in Eqs.(8a)–(8c).

FIG. 10. The PCR results when the total latent heat flux wasused as the response variable and the data were sorted by vaporpressure deficit. The criteria for partitioning the data are given inEqs. (9a)–(9c).


relationship between D and the relative influence of Fg

is to be expected since plants respond to a high D byclosing their stomata to minimize water loss (Hopkins1999). Since transpiration should be minimal whenthere is a high D, the contribution of transpiration tothe total �E is reduced and variations in Fg have lessimpact on the total moisture flux.

Also, while there was no correlation between D andU (r � 0.12), there appeared to be a threshold relation-ship between D and the relative influence of U with Utaking on a much more influential role when there wasa high D. The relative influence of U increased from18% for the low and intermediate subgroups to 36%when D was at its greatest. This relationship between Dand U suggests that the turbulent transport of watervapor is critical when there is a high D, a period thatcorresponds to low or intermediate levels of �.

An analysis was also conducted using Fg as the clas-sification variable. The three subgroups were as fol-lows:

Bare: Fgi � 0.05; �10a�

Intermediate: 0.05 � Fgi � Fg s ; �10b�

Established: Fg s � Fgi ; �10c�

where Fg is the mean greenness fraction for the entireobservational period, 0.11; s is the standard deviationfor Fg across the entire period, 0.04; and Fgi is the green-ness fraction for a given day. Nine days were sorted intothe bare category; 7 days were sorted into the interme-diate category; and 5 days were sorted into the estab-lished category.

While the results show, quite reasonably, that therelative influence for Fg increased from zero whenthere is no green vegetation to 28% once the vegetationwas well established late in the study period, it alsoshowed the relationship between the role of D and Fg

(Fig. 11). The relative influence of D increased fromless than 3% when Fg was at a bare or intermediatelevel to nearly 12% when the vegetation was estab-lished. Given the role D plays in controlling water lossvia transpiration, the increasing importance of D withincreasing amounts of vegetation cover is not alto-gether unexpected. However, the relationship betweenthe influence of D and the magnitude of Fg also sug-gests that the main influence of D on total �E isthrough the impact of D on water transfer via transpi-ration.

c. Analysis of the soil evaporation

After partitioning �E into �Esoil and �Eveg using thetwo-source model developed by Shuttleworth and Wal-

lace (1985), it was found that �Esoil was the primarysource for the moisture flux over the entire observationperiod. (Days with rain events were omitted so that theevaporation of intercepted water could be excludedfrom the analyses.) Between 53% and 100% of the total�E was due to �Esoil (Fig. 12), and on average, �Esoil

contributed 86% of the total moisture flux; �Esoil ac-counted for all of �E during the period from DOY 140to 148 because there was no significant green vegeta-tion at the study site.

Because of the large soil evaporation component(�Esoil), PCR was conducted using only the componentof �E from soil evaporation with the data sorted ac-cording to � as described above. As shown in Fig. 13,the results of this analysis reinforce those of the earlieranalysis and suggest that � was the key control particu-larly under dry or moist soil conditions. Under dry soilconditions the relative influence of � was nearly 50%,which is more than twice the relative influence of thenext most influential environmental factor, Rnet. Undermoist soil conditions, � remained the most influentialenvironmental factor with a relative influence exceed-ing 40%. Even under wet soil conditions when Rnet wasslightly more influential in driving soil evaporation, �maintained a relative influence of approximately 22%.

Since it might seem counterintuitive that Fg has aninfluence on �Esoil, it is important to recall that as Fg

varied, the fraction of the surface area that is bare alsovaried in a complementary fashion (the fractions ofbare soil and vegetated ground must sum to one). Sinceevaporation is proportional to the surface area overwhich moisture exchange can take place, for a given

FIG. 11. The PCR results when the total latent heat flux wasused as the response variable and the data were sorted by green-ness fraction. The criteria for sorting the data are given in Eqs.(10a)–(10c).


level of �, Fg influenced the amount of �Esoil through itsimpact on the amount of the surface area that is baresoil.

Given the dominance of � in controlling �E, empiri-cal relationships between � expressed as a percentageand both the total �E and �Esoil expressed in watts permeter squared were determined (Fig. 14). Both of theseempirical relationships had a sigmoidal form that canbe defined as follows:

� � �0 � � ln� �

� � �0� 1�, �11�

where is either the total �E or �Esoil and the regres-sion coefficients (�, �, 0, �0) are summarized in Table2. The coefficients of determination were 0.90 and 0.95for the total �E and �Esoil, respectively. The similaritybetween these two relationships is quite reasonablegiven that the preponderance of the total moisture fluxwas due to soil evaporation.

The relationship between � and �E (or �Esoil) may beexplained using the physical processes governing soilevaporation. The initial sharp increase in �E may bedue to the transition in soil evaporation resulting fromdiffusion from deeper soil layers (Hornberger et al.1998) to soil evaporation limited by moisture availabil-ity and soil properties (Wallace et al. 1999; Suleimanand Ritchie 2003). Once � exceeded approximately10%, �E (or �Esoil) increased in an exponential fashion.During the latter phase of the soil evaporation process,soil evaporation is proportional to the hydraulic diffu-sivity of the soil, which, in turn, increases exponentiallywith increasing � (Hillel 1998).

Another means of investigating soil evaporation is tofocus on the soil resistance (rsoil). For sparsely veg-etated surfaces where soil evaporation is large, the rsoil

is a major component of the overall surface resistanceto moisture exchange. Thus, a better understanding ofhow changing � impacts rsoil provides insights into howsoil evaporation varies with changing �. The soil resis-tance (rsoil) was determined during rain-free periodsafter Sellers et al. (1996) as follows:

rsoil � ��hsoile*soil � ea��Cp�100 � ��

100E �� ra , �12�

FIG. 13. The PCR results when soil evaporation was used as theresponse variable and the data were sorted by soil moisture con-tent. The criteria for sorting the data are given in Eqs. (8a)–(8c).

FIG. 12. The daytime mean total latent heat flux partitioned intothe flux from the soil, i.e., soil evaporation, and the flux from thevegetation, i.e., transpiration. The gaps are days during which rainevents occurred.

FIG. 14. The empirical relationships between soil moisture con-tent and both the total latent heat flux and the latent heat fluxfrom the soil, i.e., soil evaporation. The histogram shows the num-ber of rain-free days for which the mean daytime soil moisturecontent was in a given of 5% bin.


where hsoil is an estimate of the relative humidity of thesoil pore space, e*soil is the saturation water vapor pres-sure calculated at the soil temperature, ea is the watervapor pressure of the air, � is the density of water, Cp isheat capacity, � is the psychrometric constant, and ra isthe aerodynamic resistance, which was calculated as afunction of the wind speed and friction velocity (Blan-ken et al. 1997). The pore space humidity, hsoil, wasestimated as an empirical function of �.

The relationship between � expressed as percentageand rsoil expressed as seconds per meter had the form

rsoil � 1354 121.5 ln� 25.16� � 5.80

� 1�, �13�

with a coefficient of determination equaling 0.97 (Fig.15). As with the relationship between � and �E (or�Esoil), this relationship can be understood in terms ofthe underlying physical processes. First, as � ap-proached the wilting point (�w), which was estimated tobe 4.7% for the study site, rsoil asymptotes toward in-finity. Under these dry conditions, the small amount ofmoisture remaining within the soil would be boundwithin the soil matrix, and thus, it would be unavailablefor evaporation. At the opposite extreme, once � ex-ceeded field capacity (�fc) and approaches saturation(�sat), rsoil approached zero. Since saturated soils actessentially the same as a free water surface (Dingman2002), the amount of evaporation was limited only bythe meteorological conditions without any soil resis-tance. For the intermediate range of � from approxi-mately 10%–28%, rsoil decreased as an exponential de-cay function of �. Since the rsoil should decrease withincreasing water availability, this is reasonable giventhat water availability is a function of hydraulic diffu-sivity and, as noted previously, hydraulic conductivityincreases exponentially with increasing �.

4. Conclusions

Based on the results of this study, it is clear that � wasan important control on both the total �E and the pri-mary pathway for moisture exchange with the atmo-sphere, soil evaporation. Particularly, under dry and

moist soil conditions, when evaporation from the sitewas essentially water limited, � had a strong control onthe moisture flux. However, other environmental fac-tors, such as Rnet and Fg, also had a major influence on�E when soil conditions were wet, that is, when �Efrom the site was energy limited.

Given the high relative influence of � in controlling�E, which exceeded 40% in some cases, a thoroughunderstanding of soil hydrology is a prerequisite foraccurately describing or modeling the exchange ofmoisture with the atmosphere. To accurately describe�E during drought conditions, it is first necessary toaccurately describe the soil hydrology and other physi-cal processes controlling �. This conclusion is high-lighted by the strong empirical relationships developedbetween � and the total �E, �Esoil, and rsoil.

This study has important implications for the model-ing community. Currently, many land surface models,such as the Noah land surface model (Ek et al. 2003),the Simple Biosphere II (Sellers et al. 1996), and theCommunity Climate Model (Oleson et al. 2004), deter-mine rsoil either directly or indirectly as an exponentialdecay function of �. These results suggest that a sigmoi-dal relationship may yield a more accurate descriptionof rsoil and ultimately the total �E in sparse-canopyenvironments. Such a relationship could be especiallyuseful when describing �E at the soil moisture extremesas was the case during the IHOP_2002 field campaign.

Given the social, economic, and environmental ef-fects of drought, it would be beneficial to be able toaccurately model the extent, duration, and severity ofdrought so that its adverse effects can be minimized.This study reemphasized the importance of � and the

TABLE 2. Regression coefficients for the best-fit sigmoidal re-lationships between soil moisture content, and both the total la-tent heat flux (�Etot) and soil evaporation (�Esoil) are given.

Coefficient �Etot �Esoil

0 84.82 64.98� 11.28 9.76� 25.10 25.23�0 7.21 6.48

FIG. 15. The empirical relationship between soil moisture con-tent and soil resistance. The locations of the wilting point �w, fieldcapacity �fc and saturation (�sat) soil moisture content are labeled.


need to describe accurately soil hydrology to predict �Ecorrectly. This study also provided a practical means ofimproving land surface models by more accurately de-scribing the relationship between � and rsoil under ex-tremes in soil moisture conditions.

These conclusions are particularly salient to the de-velopers of land surface models. While experimentaland field studies have resulted in a number of methodsfor estimating soil evaporation, these studies have oftenbeen conducted over bare soil (e.g., Mahfouf and Noil-han 1991) with little consideration of the role of veg-etation and other environmental factors. As a result,many land surface models estimate soil evaporationwithout fully accounting for all of the factors that drivethe moisture flux.

Although this study was conducted at a specific lo-cation in the panhandle of Oklahoma, the grazed landcover and environmental conditions at the site are notatypical of the southern Great Plains and AmericanWest as a whole. Therefore, the general findings of thisresearch are likely applicable to regions with a similarland use, soil, vegetation, and vulnerability to droughtconditions. Further research should confirm if othersparsely vegetated ecoregions, such as boreal forest, be-have similarly to this study site. To test the broaderapplicability of this study, follow-up research is ongo-ing. The soil moisture–soil resistance relationship is be-ing tested over the entirety of the IHOP_2002 domainby implementing the relationship in the Noah land sur-face model. After the sigmoidal function for rsoil hasbeen successfully validated over the IHOP_2002 do-main, it will be tested over other ecological regions.

Acknowledgments. The authors thank M. LeMone,R. Grossman, and M. O’Connell for their assistanceduring the IHOP_2002 field campaign. The authorsthank F. Chen, T. Horst, S. Oncley, G. McLean, J. Me-cikalski, R. Cuenca, and D. Gochis for their discussionsand insight during the course of this research. The au-thors thank S. Grant for her assistance with preparingthis manuscript. The authors thank the three anony-mous reviewers and the editor of the Journal of Hy-drometeorology for their thoughtful comments and sug-gestions that greatly improved the quality and claritythis paper. The authors would like to acknowledge thefinancial support of the NCAR Water Cycle Initiative,and the National Science Foundation (Awards ATM-0236885 and ATM-0296159).

REFERENCES

American Meteorological Society, 1997: Meteorological drought—Policy statement. Bull. Amer. Meteor. Soc., 78, 847–849.

Barbour, M. G., J. H. Burk, W. D. Pitts, F. S. Gilliam, and M. W.

Schwartz, 1999: Terrestrial Plant Ecology. 3d ed. AddisonWesley, 649 pp.

Blanken, P. D., and Coauthors, 1997: Energy balance and canopyconductance of a boreal aspen forest: Partitioning overstoryand understory components. J. Geophys. Res., 102, 28 915–28 927.

Bolton, D., 1980: The computation of equivalent potential tem-perature. Mon. Wea. Rev., 108, 1046–1053.

Campbell Scientific, 1998: Eddy Covariance System Operator’sManual CA27 and KH2O. 29 pp.

Carlson, T. N., and D. A. J. Ripley, 1997: On the relationshipbetween NDVI, fractional vegetation cover, and leaf areaindex. Remote Sens. Environ., 62, 241–252.

Changnon, S. A., 1989: The 1988 drought, barges, and diversion.Bull. Amer. Meteor. Soc., 70, 1092–1104.

Diaz, H. F., 1983: Drought in the United States. J. Climate Appl.Meteor., 22, 3–16.

Dingman, S. L., 2002: Physical Hydrology. 2d ed. Prentice-Hall,646 pp.

Douglas, A., K. Gleason, D. Phillips, and A. M. Waple, 2003:State of climate in 2002. Bull. Amer. Meteor. Soc., 84, S1–S68.

Draper, N. R., and H. Smith, 1981: Applied Regression Analysis.John Wiley and Sons, 709 pp.

Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V.Koren, G. Gayno, and J. D. Tarpley, 2003: Implementation ofNoah land surface model advances in the National Centersfor Environmental Prediction operational mesoscale EtaModel. J. Geophys. Res., 108, 8851, doi:10.1029/2002JD003296.

Falge, E., and Coauthors, 2001: Gap filling strategies for long termenergy flux data sets. Agric. For. Meteor., 107, 71–77.

Gould, F. W., and R. B. Shaw, 1983: Grass Systematics. 2d ed.Texas A&M University Press, 397 pp.

Gutman, G., and A. Ignatov, 1997: Satellite-derived green veg-etation fraction for the use in numerical weather predictionmodels. Adv. Space Res., 19, 477–480.

Heim, R. R., 2002: A review of twentieth-century drought indicesused in the United States. Bull. Amer. Meteor. Soc., 83, 1149–1165.

Hillel, D., 1998: Environmental Soil Physics. Academic Press, 771pp.

Hopkins, W. G., 1999: Introduction to Plant Physiology. JohnWiley and Sons, 512 pp.

Hornberger, G. M., J. P. Raffensperger, P. L. Wiberg, and K. N.Eshleman, 1998: Elements of Physical Hydrology. John Hop-kins University Press, 312 pp.

Horst, T. W., 2006: Attenuation of scalar fluxes measured withhorizontally-displaced sensors. Extended Abstracts, 17thSymp. on Boundary Layers and Turbulence, San Diego, CA,Amer. Meteor. Soc., CD-ROM, 7.5.

Jackson, J. E., 1991: A User’s Guide to Principal Components.John Wiley and Sons, 569 pp.

Jolliffe, I. T., 2002: Principal Component Analysis. Springer-Verlag, 516 pp.

Jury, W. A., and R. Horton, 2004: Soil Physics. John Wiley andSons, 370 pp.

Kaimal, J. C., and J. J. Finnigan, 1994: Atmospheric BoundaryLayer Flows. Oxford University Press, 289 pp.

Korgan, F. N., 1997: Global drought watch from space. Bull.Amer. Meteor. Soc., 78, 621–636.

Lawrimore, J., R. R. Heim, M. Svoboda, V. Swail, and P. J. Engle-hart, 2002: Beginning a new era of drought monitoring acrossNorth America. Bull. Amer. Meteor. Soc., 83, 1191–1192.

LeMone, M. A., and Coauthors, 2007: NCAR/CU surface, soil,


and vegetation observations during the international H2Oproject 2002 field campaign. Bull. Amer. Meteor. Soc., 88,65–81.

Mahfouf, J. F., and J. Noilhan, 1991: Comparative study of variousformulations of evaporations from bare soil using in situ data.J. Appl. Meteor., 30, 1354–1365.

Montgomery, D. C., and E. A. Peck, 1992: Introduction to LinearRegression Analysis. John Wiley and Sons, 461 pp.

National Climatic Data Center, 2002: U.S. national percent areaseverely to extremely dry and severely to extremely wet.[Available online at http://www.ncdc.noaa.gov/oa/climate/research/2002/may/uspctarea-wetdry.txt.]

Obashi, G., 1994: WMO’s role in the international decade fornatural disaster reduction. Bull. Amer. Meteor. Soc., 75,1655–1661.

OCS, 2002a: Oklahoma monthly climate summary, May 2002.University of Oklahoma, 23 pp. [Available from OklahomaClimatological Survey, University of Oklahoma, 100 E. BoydSt., Ste. 1210, Norman, OK 73019.]

——, 2002b: Oklahoma monthly climate summary, June 2002.University of Oklahoma, 23 pp. [Available from OklahomaClimatological Survey, University of Oklahoma, 100 E. BoydSt., Ste. 1210, Norman, OK 73019.]

——, 2002c: Oklahoma event summary: The Oklahoma droughtof 2001-2002. University of Oklahoma, 29 pp. [Availablefrom Oklahoma Climatological Survey, University of Okla-homa, 100 E. Boyd St., Ste. 1210, Norman, OK 73019.]

Oke, T. R., 1987: Boundary Layer Climates. 2d ed. Routledge, 464pp.

Oleson, K., and Coauthors, 2004: Technical description of theCommunity Land Model (CLM). NCAR Tech. Note NCAR/TN-461STR, National Center for Atmospheric Research,173 pp.

OWRB, 2002a: Oklahoma water resources bulletin and summaryof current conditions, May 8, 2002. Oklahoma Water Re-source Board, 8 pp. [Available from Oklahoma Water Re-sources Board, 3800 N. Classen Blvd., Oklahoma City, OK73118.]

——, 2002b: Oklahoma water resources bulletin and summary ofcurrent conditions, July 3, 2002. Oklahoma Water ResourceBoard, 8 pp. [Available from Oklahoma Water ResourcesBoard, 3800 N. Classen Blvd., Oklahoma City, OK 73118.]

——, 2003: Oklahoma water resources bulletin and summary ofcurrent conditions, July 3, 2003. Oklahoma Water ResourceBoard, 8 pp. [Available from Oklahoma Water ResourcesBoard, 3800 N. Classen Blvd., Oklahoma City, OK 73118.]

Pond, S., G. T. Phelps, J. E. Paquin, G. McBean, and R. W. Stew-

art, 1971: Measurement of turbulent fluxes of momentum,moisture and sensible heat over ocean. J. Atmos. Sci., 28,901–917.

Redmond, K. T., 2002: The depiction of drought: A commentary.Bull. Amer. Meteor. Soc., 83, 1143–1147.

Riebsame, W. E., S. Changnon, and T. R. Karl, 1991: Drought andNatural Resource Management in the United States: Impactsand Implications of the 1987-89 Drought. Westview Press, 174pp.

Schotanus, P., F. T. M. Nieuwstadt, and H. A. R. DeBruin, 1983:Temperature measurements with a sonic anemometer and itsapplication to heat and moisture fluxes. Bound.-Layer Me-teor., 26, 81–93.

Sellers, P. J., and Coauthors, 1996: A revised surface parameter-ization (SiB2) for atmospheric GCMs. Part I: Model formu-lation. J. Climate, 9, 676–705.

Shuttleworth, W. J., and J. S. Wallace, 1985: Evaporation fromsparse crops—An energy combination theory. Quart. J. Roy.Meteor. Soc., 111, 839–855.

Skaggs, R. H., 1975: Drought in the United States, 1931-40. Ann.Assoc. Amer. Geogr., 65, 391–402.

Suleiman, A. A., and J. T. Ritchie, 2003: Modeling soil water re-distribution during second-stage evaporation. Soil Sci. Soc.Amer. J., 67, 377–386.

Svoboda, M., and Coauthors, 2002: The drought monitor. Bull.Amer. Meteor. Soc., 83, 1181–1190.

Trenberth, K. E., and C. J. Guillemot, 1996: Physical processesinvolved in the 1988 drought and 1993 floods in NorthAmerica. J. Climate, 9, 1288–1298.

Wallace, J. S., N. A. Jackson, and C. K. Ong, 1999: Modelling soilevaporation in an agroforestry system in Kenya. Agric. For.Meteor., 94, 189–202.

Webb, E. K., G. I. Pearman, and R. Leuning, 1980: Correction offlux measurements for density effects due to heat and watervapor transfer. Quart. J. Roy. Meteor. Soc., 106, 85–100.

Weckwerth, T., and Coauthors, 2004: An overview of the Inter-national H2O Project (IHOP 2002) and some preliminaryhighlights. Bull. Amer. Meteor. Soc., 85, 253–277.

Wilhite, D., 2000: Drought as a natural hazard: Concepts anddefinitions. Drought: A Global Assessment, D. Wilhite, Ed.,Routledge, 3–18 pp.

Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences.Academic Press, 467 pp.

Woodhouse, C. A., J. J. Lukas, and P. M. Brown, 2002: Drought inthe Western Great Plains, 1845–56. Bull. Amer. Meteor. Soc.,83, 1485–1493.


Documents

Variability in the Environmental Factors Driving Evapotranspiration